Why Follow the Leader?

Publc Dsclosure Authorzed Publc Dsclosure Authorzed Publc Dsclosure Authorzed Publc Dsclosure Authorzed Polcy Research Workng Paper 6179 Why Follow the Leader? Collectve Acton, Credble Commtment and Conflct The World Bank Development Research Group Macroeconomcs and Growth Team August 2012 Phlp Keefer WPS6179

Polcy Research Workng Paper 6179 Abstract Most analyses of conflct assume that conflctng groups act n a untary fashon. Ths assumpton s often volated: to reduce ther rsk of replacement, group leaders prevent both group members and solders from actng collectvely, makng t dffcult for leaders to make credble commtments to them. Lftng the assumpton that groups are untary shfts the analyss of a wde range of conflct ssues. The effects of ncome shocks and rents on conflct rsk become contngent on collectve acton. Leader decsons regardng collectve acton explan the forcble recrutment of chld solders and predaton on cvlans: leaders who prefer to lmt mltary organzaton are more lkely to pursue these tactcs. Leader decsons regardng collectve acton also ntroduce an unexplored mechansm by whch state capacty s created and a specfc reason to regard state capacty as endogenous to conflct rsk. Ths focus, fnally, suggests that nterventons to reduce conflct rsk, such as safety net payments or servce delvery, are lkely to be most dffcult to delver precsely where leaders are most reluctant to allow collectve acton and where, therefore, conflct rsk s hghest. Ths paper s a product of the Macroeconomcs and Growth Team, Development Research Group. It s part of a larger effort by the World Bank to provde open access to ts research and make a contrbuton to development polcy dscussons around the world. Polcy Research Workng Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at pkeefer@worldbank.org. The Polcy Research Workng Paper Seres dssemnates the fndngs of work n progress to encourage the exchange of deas about development ssues. An objectve of the seres s to get the fndngs out quckly, even f the presentatons are less than fully polshed. The papers carry the names of the authors and should be cted accordngly. The fndngs, nterpretatons, and conclusons expressed n ths paper are entrely those of the authors. They do not necessarly represent the vews of the Internatonal Bank for Reconstructon and Development/World Bank and ts afflated organzatons, or those of the Executve Drectors of the World Bank or the governments they represent. Produced by the Research Support Team

Why follow the leader? Collectve acton, credble commtment and conflct Phlp Keefer Prepared for the Oxford Handbook of the Economcs of Peace and Conflct Mchelle Garfnkel and Stergos Skaperdas, edtors. JEL codes: D70, D72, D74 Key words: Conflct, cvl war, credble commtment, collectve acton Sector Board: Publc Sector Governance (PSM) Acknowledgements: Ths paper has benefted greatly from the generous and nsghtful comments of Gary Mlante, Jacob Shapro and the edtors. Contact nformaton: Development Research Group The World Bank 1818 H St. NW Washngton, DC 20433 pkeefer@worldbank.org

2 Credble commtment s at the heart of the study of conflct. However, most of the lterature s concerned wth the credblty of assurances by opposng groups not to take up arms aganst each other: when each group s unwllng to trust the other to lay down ts arms or to refran from a preemptve attack, conflct s more lkely (Azam 1995 and Fearon 1995). Less attenton has been pad to commtments between group leaders and members, although the credblty of these commtments nfluences every aspect of conflct, from a group s ablty to moblze mltary capacty to ts ablty to make agreements wth other groups. Ths paper focuses on these commtments. The central argument here s that leader commtments are credble to the extent that group members and armed forces can act collectvely to enforce them. However, to nsulate themselves from the rsk of expulson by ther own supporters or armed forces, leaders mpose lmts on collectve acton. These lmts make t more dffcult for them to wn the hearts and mnds of ctzens and to feld an effectve mltary, each of whch ncreases a group s vulnerablty to attack by opposng groups. Fearon (2008) descrbes the modal cvl war n the last 60 years as persstent, small and relyng on guerrlla tactcs, rather than as a conflct between large conventonal armes, as n the Unted States Cvl War. However, restrctons on collectve acton also seem to dstngush the modal conflct from the U.S. Cvl War. The Confederacy allowed substantally more collectve acton by (whte) ctzens n ts terrtory than dd the Taml Tgers n northern Sr Lanka, the Shnng Path n the Peruvan Andes, or the Democratc Forces for the Lberaton of Rwanda n the eastern part of the Democratc Republc of Congo (DRC). Unon forces were not comprsed of a well-funded Presdental Guard and a poorly-funded regular army, as n the DRC, or closely drected by the presdent, as n Sudan. The dscusson below therefore examnes the decsons of leaders to allow ctzens and solders to act collectvely. Revews of qualtatve and quanttatve evdence, partcularly related to poltcal partes, ndcate that ctzens n countres vulnerable to conflct exhbt less ablty to act collectvely. Ample qualtatve evdence also demonstrates sgnfcant dfferences n the extent to whch governments allow mltares to organze collectvely: nter-unt cooperaton, nformaton transmsson wthn the mltary, and mltary control over promotons and honors are all more heavly restrcted n some countres than others. The dynamcs of wthn-group collectve acton have sgnfcant mplcatons for debates across the conflct lterature. Recent research concludes that ncome shocks encourage conflct by changng the costs of conflct (the wages of combatants) more than the rents at stake n conflct. The dscusson here concludes that these arguments apply most strongly n precsely those countres where armed forces cannot act collectvely to enforce a clam on future rents or where ctzens cannot act collectvely to prevent the armed forces from makng such a clam. The dscusson s also relevant to partcularly dsturbng features of conflct, such as predaton on cvlans and coerced recrutment of chld solders. The frst could be a drect consequence of leader efforts to undermne organzaton and dscplne wthn the armed forces n order to defuse coup threats. The latter s usefully seen as part of a class of measures that leaders take to reduce coup threats, snce adolescents are less lkely than adults to organze a rebellon aganst them. Scholars have hghlghted the role of state capacty n shapng the vulnerablty of countres to conflct. That research generally focuses on the fscal costs of capacty-buldng and only recently has recognzed that capacty s not an exogenous determnant of conflct

3 rsk, but one that s determned, jontly wth conflct rsk, by other factors. The dscusson here emphaszes that capacty s a functon of organzatonal choces that leaders make, whch are ndependent of fnancng. The constrant on capacty s not budgetary, n ths case, but the threat that capacty (a better organzed publc admnstraton or mltary) mght pose to leader tenure. The conjectures developed here justfy a closer examnaton of current strateges to mnmze conflct rsk. For example, n vew of ther fndng that ranfall (and, therefore, ncome) shocks trgger conflct n Afrca, Mguel et al. (2004) propose the adopton of crop nsurance and safety net payments. The donor communty has begun to focus on mproved servce delvery n post-conflct settngs as a way to reduce tensons and buld up state legtmacy. However, f conflct s most lkely n countres where leaders are reluctant to allow ctzen collectve acton, these strateges may be least lkely to succeed precsely where they are most needed. Leaders who do not allow collectve acton are less lkely to provde servces effectvely, to target safety net payments approprately, or to enforce nsurance contracts. Credble commtment, collectve acton and conflct Most of the conflct lterature examnes the effects of poltcal nsttutons on conflct from the perspectve of the credblty of nter-group agreements. As Garfnkel and Skaperdas (2007) observe, scholars tend to assume that antagonsts are untary actors and abstract from problems of ntra-group collectve acton, regardless of whether they use barganng models that focus on the nablty of antagonsts to make credble commtments, or contest functons that examne the sacrfces that groups make n terms of productve actvty n the pursut of conquest. There are exceptons to ths rule. Kaplan (2010) uses evdence from Colomba to support hs argument that vllages wth greater ablty to organze collectvely are better able to resst the pressures placed on them by armed forces (nsurgent or government). Garfnkel (2004) analyzes decsons governng group sze. Larger groups are more lkely to preval n conflct, but ndvduals n larger groups have to devote more resources to protect ther share from other group members. Larger groups are therefore more lkely to emerge when nsttutons of conflct management, exogenously determned, are better establshed wthn a group, attenuatng ntra-group conflct. The focus here s: To what degree do leaders of nsurgent or government forces encourage collectve by ctzens and armed forces, and how do the efforts of group leaders to reman n power shape group nsttutons? In partcular, the next secton descrbes leader decsons to allow collectve acton by supporters and examnes the consequences of these decsons for the ablty of leaders to wage the battle for hearts and mnds. The secton followng consders the decson to lmt collectve acton by armed forces, even f ths degrades mltary effectveness. The fnal sectons of the paper then outlne the lkely mplcatons of varaton n ntra-group nsttutonal arrangements for the role of ncome and ncome shocks n precptatng conflct, predaton on cvlans, relance on chld solders, state capacty, and mproved servce delvery as a strategy for defusng conflct. Collectve acton and the battle for hearts and mnds Berman, Shapro and Felter (2009a) pont to a consensus among promnent practtoners of nsurgency and counternsurgency, from Mao Tse-Tung to Davd Petraeus,

4 on the mportance of ganng popular support n conflct outcomes. Accordng to practtoners, and scholars such as Kalyvas (2006) and Fearon (2008), the key strategc advantage from ganng the support of non-combatants s the provson of nformaton about the armed forces of the opposton. Problems of wthn-group credble commtment (such as whether ctzens can rely on leaders to pursue ther nterests n the event that they preval n the conflct) are not the central focus of these analyses. Nagl (2002) places great weght on the mportance for counternsurgency strategy of ganng the trust of local actors. He focuses most on how trust matters and less on how t emerges. One way t can emerge, however, as s mplct n Kaplan s (2010) analyss, s f local actors are able to act collectvely, allowng them to mpose a larger sancton on leaders who renege on ther commtments. The lterature portrays the battle for hearts and mnds n two ways; collectve acton by ctzens s relevant to both. One s the struggle by competng groups to make credble commtments to cvlan populatons to reward them for cooperaton. Another s ther struggle to delver the benefts of good government to these populatons, such as publc securty, socal servces, or predctable and non-predatory taxaton. Collectve acton by cvlan populatons plays a central role n both cases: collectvely organzed cvlans can more easly punsh groups that renege on ther commtments or who fal to provde the benefts of good government. The most-studed nsttutonal arrangement through whch ctzens can hold leaders accountable s votng. In fact, the absence of compettve electons s a notable feature of conflct countres. Systematc data supports ths clam n the case of governments; casual observaton offers lttle ndcaton that nsurgent-controlled terrtores are any dfferent. The Sambans (2004) conflct database records 71 conflcts from 1975 2000 for whch data on compettve electons (from the Database of Poltcal Insttutons) are avalable. In only 15 of these cases were countres governed by compettvely elected leaders n the year before the conflct started. In the remander, governments were not compettvely elected or not elected at all. Indeed, ths emprcal regularty s as strkng as the assocaton of conflct wth ncome: of these same 71 conflcts, agan only 15 (not the same 15) occurred n the rchest 50 percent of countres. Although the paucty of electons n conflct countres s an ndcaton that leaders n conflct countres mpose lmts on collectve acton by ctzens, electons are not a suffcent condton for effectve collectve acton. Even ctzens wth unfettered votng rghts are handcapped n ther ablty to hold leaders to account when they are constraned by mperfect nformaton (the nablty to observe leader actons or the effects of those actons on ther welfare). More mportantly for the analyss here, though unfettered votng rghts lower the ndvdual costs of holdng leaders accountable, they do not guarantee that challengers wll emerge who can credbly commt to pursung dfferent polces than those of the ncumbent. Electoral accountablty of leaders depends on the exstence of such challengers. Ferejohn (1986) and Persson and Tabelln (2000) examne the case where credble challengers do not exst. Ther absence substantally attenuates ncumbent ncentves to pursue the publc nterest (e.g., to provde publc goods). These results extend mmedately to the conflct settng: such leaders, even f elected, have lmted ncentves to wn hearts and mnds through the provson of publc servces.

5 An mportant open queston s how poltcal compettors develop the capacty to make credble promses. A lkely answer, though, s rooted n the ablty of ctzens to act collectvely to choose and support canddates, workng especally through programmatc poltcal partes that are organzed to represent the nterests of lke-mnded ctzens (Keefer 2007b). Such partes allow members to coordnate wth each other; offer them fora at whch they can select canddates; requre potental new members or canddates to nvest n costly sgnals that they share the goals of the exstng members (e.g., as n Snyder and Tng 2002); and exhbt procedures to expel members who do not pursue the group s collectve nterests. Challengers chosen n such a process are better able to make credble commtments because the exstence of the party ensures the contnung ablty of ctzens to act collectvely to choose a dfferent canddate f challengers renege on those commtments. Other research extends ths logc to non-democraces. Even f ctzens cannot vote, autocrats can nevertheless decde to admt a certan fracton of the socety nto the rulng party and to permt members of the rulng party to act collectvely. Gehlbach and Keefer (2009), for example, argue that when autocrats allow rulng party members to share nformaton about autocrat behavor that s not avalable to others, members can act collectvely to punsh leader expropraton of ther nvestments. By usng rulng party nsttutonalzaton to lmt ther rents from expropraton, autocrats can attract greater prvate nvestment and can elct greater effort from party members n the pursut of leader objectves. These objectves could nclude nsurgency (for leaders of rebel groups) or counter-nsurgency (for government leaders). Ths dscusson suggests, then, that whether leaders try to wn hearts and mnds depends not only on the nsttutons that affect nter-group contractng and that are the focus of the conflct lterature, such as electons, electoral rules (proportonal representaton or pluralty electons, for example), or poltcal checks and balances. They also depend on the ntra-group nsttutonal arrangements that allow leaders to credbly commt to pursue the nterests of supporters. These arrangements are those that allow large groups of supporters to act collectvely. Ths logc predcts that governments should confront a greater rsk of nsurgency by opposng groups when they are reluctant to allow ther own supporters to organze, whch n turn s most lkely to be the case when competton for poltcal offce s not medated by nsttutonalzed partes (partes, for example, that facltate collectve acton by members regardng the selecton of party canddates). In the absence of such partes, the probablty of nsurgency success rses snce government ablty to wn the battle for hearts and mnds through the provson of publc servces s attenuated. Keefer (2008) presents evdence to ths effect. In that analyss, each of the followng proxes for the ablty of ctzens to act collectvely has a sgnfcant effect on conflct rsk: the contnuous years of compettve electons (capturng the ablty of poltcal compettors to make broadly credble commtments to ctzens, as n Keefer 2007); the degree to whch partes convey a programmatc stance to ctzens (snce such a stance s only credble f partes have organzatonal arrangements that force out leaders whose actons are nconsstent wth the party program); and the age of the rulng party relatve to the years that the leader has been n offce (rulng partes created by leaders are both lkely to be younger than the leader s years n offce and under the control of the leader). When values of any of these varables are lower, conflct s sgnfcantly more lkely, partcularly n the poorest 50 percent of countres

6 where more than 75 percent of all conflcts occur. Every addtonal year by whch the age of the governng party exceeds the years a ruler has been n offce reduces the odds of conflct by about 2 percent a year. The Democratc Republc of Congo provdes a specfc llustraton of the assocaton of weak collectve acton wth conflct. In the 2006 electons, the frst after the cessaton of wdespread conflct (although guerrlla actvty perssted n the eastern provnces), 213 partes competed for parlamentary seats. Followng the electons, 14 partes were represented n the government and 70 n the Natonal Assembly. Such fragmentaton s consstent wth the nablty of partes to make credble commtments to serve the nterests of broad groups of ctzens. It s also consstent wth the fact that, except for the bref perods from 1960-1967 and 1990-1997, poltcal partes had been banned n the DRC, offerng no hstorcal bass for party-based poltcal competton n 2006 (Internatonal Crss Group 2008 p. 15). If partes are based on personal relatonshps and clentelst tes, as n the DRC, leadershp changes should have a dramatc effect on party stablty. Consstent wth ths, when Jean-Perre Bemba, the leader of the opposton Movement for the Lberaton of Congo, left the country, a sgnfcant number of the party s legslators crossed to the government coalton (Oxford Analytca May 9, 2009; Congo-Knshasa: Kabla gans from opposton eroson. ). Fnally, f clentelst tes are the foundaton of partsan organzaton and party members are not able to act collectvely, legslators from a party confront lttle electoral rsk n castng votes that are nconsstent wth the preferences of the party s voters. Consstent wth ths, after the 2006 electons, n nearly all provnces legslators elected governors from the presdent s coalton, despte the fact that n many provnces, the presdent s coalton won only a small fracton of the vote ( Congo-Knshasa: Clashes hghlght post-poll challenges, OA, March 26, 2007). Ths dscusson ponts to a strong assocaton between conflct and the ablty of ctzens to act collectvely. It has not touched on the queston of why ctzens are organzed for collectve poltcal acton n some countres, but not n others. Ths s agan a largely open area n the lterature. In an early effort, Keefer and Vlacu (2008) argue that poltcans compare the costs of buldng poltcal credblty to the poltcal advantages of beng able to make broadly credble commtments to them. When costs are hgh, poltcans choose to make credble pre-electoral commtments only to narrow groups. Such poltcans then favor low publc good provson; hgh rates of prvate good provson to those narrow groups; and to engage n hgh rates of rent-seekng all contrary to the objectve of wnnng hearts and mnds. They are partcularly lkely to make ths decson where, as n many conflct countres, patron-clent relatonshps are deeply-rooted n socety, makng narrow appeals cheap relatve to broad-based appeals. In many parts of Afghanstan or Iraq, for example, t s easer for poltcal compettors to buld support by makng commtments to patrons, who have no nterest n broad publc good provson, than to nvest n the ablty to make credble commtments to broad groups of voters. Alternatvely, though, the choce for poltcans may not be how much to nvest n buldng ther credblty. Instead, t may be whether to form a party n whch members, rather than they themselves, are allowed to choose canddates. The less sure they are that they wll have the collectve support of party members, the less lkely they are to organze a party n whch members have free ren to act collectvely. Ths tradeoff re-emerges n the

7 next secton: leaders are more lkely to allow the armed forces to act collectvely, ncreasng both ther mltary effectveness and ther ablty to launch a coup, the lower the rsk that the armed forces wll want to launch a coup. 1 Collectve acton and mltary effectveness Mltary capacty s naturally a central ssue n conflct, but most analyses abstract from the leader s dlemma that armed forces able to undertake collectve acton, and therefore more effectve mltarly, can also demand a hgher share of rents from leaders and threaten the leader wth replacement. Ths secton ponts to the advantages that wellorganzed securty forces offer to leaders; documents decsons by leaders that ntentonally mpede the effcacy of ther armed forces; and traces these decsons to leaders fear of overthrow by well-organzed troops. It s easy to see that the ablty of a leader s armed forces to act n a coordnated manner collectvely s essental to effectve performance; examples below show how leaders n some countres mpose large barrers to jont exercses by ther country s own ar and ground forces, wth obvous mplcatons for mltary readness. Another advantage, dscussed n Wensten (2005) and Keefer (2008), relates to the contractng of solders. Leaders who cannot credbly promse future rewards to ther solders (government or nsurgent) must pay them n the form of spot payments that fully cover the reservaton wage of the fghters. 2 Ther ablty to recrut a fghtng force depends entrely on the rents that they control durng the conflct perod. Snce conflct tself tends to degrade the productve capacty of countres, leaders who cannot make credble commtments to ther solders regardng future compensaton are more lkely to be compelled to rely on natural resource rents and external fundng. One of the most successful nsurgent groups, the Vet Cong, recognzed ths: long after the Vetnam War, veterans of the North Vetnamese army and ther chldren contnued to receve prvleged access to jobs and housng. 3 Keefer (2008) argues further that even f leaders have szeable rents at ther dsposal, they mght stll be reluctant to rely on spot contracts wth solders. Effectve armed forces are rarely constructed on the bass of spot contracts because mltary effort s dffcult to observe and bravery hard to purchase. One way that leaders can promote effort s to credbly commt to pursung natonal goals that are compatble wth those of the armed forces. 4 Another s to offer ex post rewards for successful mltary acton, whether medals or promotons. However, these promses must be credble f they are to have an effect. 1 A smlar tradeoff s central to the democratzaton lterature. Under what condtons do eltes allow electons (e.g., as n Acemoglu and Robnson 2006 and Box 2003)? The decson s dffcult because of a well-known double-edged commtment problem: eltes cannot promse to refran from expropratng non-eltes f they do not allow electons; non-eltes cannot promse to refran from establshng hgh tax rates at the expense of eltes f electons are allowed, promptng eltes to resst electons. 2 The logc here s smlar to that n Schultz and Wengast (2003), who argue that leaders who can credbly commt to repay loans have an advantage n conflcts wth countres whose leaders cannot. 3 Author ntervew, Hano, February 24, 2006. Ths could reflect the possblty that former Vet Cong enjoy nsder status and prvleged access to rents; the examples gven suggest that even former solders wth lttle nfluence on government decsons enjoy these prvleges. 4 Wensten (2005) and Humphreys and Wensten (2006) emphasze that pecunary compensaton matters less when solders and leaders share the same deologcal objectves. Even here, however, solders requre some

8 Although organzed mltares, capable of collectve acton, are more effectve n conflcts wth regme opponents, they also pose greater threats to leaders themselves. Leaders who are more fearful of coup threats than of attacks by regme opponents are therefore more lkely to make decsons that lead to mltary dsorganzaton. These decsons can take many forms. One s to starve the armed forces of resources. Prevous research has emphaszed access to resources as the key to mltary effectveness, to meet the usual requrements of salares, supples and weapons. Consstent wth ths emphass, mltary dysfuncton and low salares seem to go together. In the Democratc Republc of Congo, for example, salares of top offcers n the army, known for ts neffectveness, were approxmately US$80 per month (Internatonal Crss Group 2006, pp. 5, 11). However, under-fundng may not sgnal the leader s lack of resources, as the lterature emphaszes. Instead, t could result from the leader s reluctance to fund a potental coup threat. In the DRC, although resources were not avalable for army offces, mnsters salares approached US$4,000 per month. Leaders concerned about coup threats can also create separate and competng securty forces and prevent them from cooperatng wth one another. De Atkne (1999) reports that jont commands and exercses are rare n many Mddle Eastern mltares and efforts by one branch of the mltary to secure the assstance of another (for example, f an army wshes to obtan arcraft from an ar force for army arborne tranng), must be coordnated by heads of servces at the mnstry level, and lkely requres presdental approval. Sadaam Hussen s management of the Iraq armed forces provdes the most extreme example of ths. Hashm (2003) summarzes hs dscusson of Hussen s polcy throughout hs rule to rotate offcers, to replace even successful offcers wth ncompetent loyalsts, and to establsh competng armed forces, even at substantal cost to mltary readness. Even though Iraq faced a hgher threat envronment n the 1990s vs à vs both external and nternal threats, n [Saddam Hussen s] mnd t was the most proxmate force to hm, the Iraq mltary, whch contnued to represent the gravest threat (p. 14). The DRC mltary was also dvded. In 2006, the army was 140,000 strong and poorly funded, but the Presdental Guard, wth the sole purpose of protectng the presdent, had 12,000 solders and was relatvely well-funded. In July 2006, for example, 42 tanks and armored vehcles were offcally delvered to the Congolese army, but only the Presdental Guard had forces traned to use them (Oxford Analytca (2006); Congo-Knshasa: Hstorc poll not end of transton, October 27). Another way to nhbt ntra-mltary coordnaton s to prohbt nformaton-sharng across mltary unts. 5 Focusng on armes n the Arabc countres of the Mddle East, De Atkne (1999) ndcates that wthn-mltary flows of nformaton are tghtly restrcted, another ndcaton of lmtatons on collectve acton. In partcular, promotons, transfers, names of unt commanders, and unt desgnatons are all frequently classfed. Observers also fnd that leaders n the regon prevent ndependent collectve acton wthn the mltary assurance that leaders wll not pursue goals that are ncompatble wth those objectves; collectve organzaton s agan mportant. 5 As Gehlbach and Keefer, 2009 demonstrate, lmts on nformaton flows are a sgnfcant obstacle to collectve acton wthn rulng partes

9 by lmtng the delegaton of authorty: solders and offcers who are dened the dscreton to undertake ndependent acton can less easly ntate collectve acton aganst the leader. Gehlbach and Keefer (2010) contrast the decsons of some unelected leaders, who make coordnaton wthn the mltary dffcult to mnmze coup threats, even at the expense of mltary effectveness, wth the experence of Indonesa under Suharto. After defeatng the Communst nsurgency, Suharto made sgnfcant efforts to unfy a fragmented mltary. In the mmedate post-nsurgency era, securty threats loomed larger than coup threats. Consequently, Suharto ntroduced a jont command and permtted strateges that gave substantal dscreton to lower level offcers to collaborate wth local cvlan authortes. These decsons were well-suted to combatng the re-emergence of nsurgent threats, at the cost of mprovng the mltary s coup capablty. In the later years of hs regme, however, Suharto changed strateges, concerned that he had more to fear from a mltary from whch hs peers had retred and less to fear from nsurgents after years of fast growth. He nserted hmself nto the mltary s promoton decsons, placng close assocates n hgh postons; he more overtly played powerful generals off one another, probably resultng n multple nformal chans of command that led only to Suharto (Callahan, p. 13). Dscplne broke down among the lower ranks of solders and was non-exstent n many regons (Callahan, p. 15). 6 The effects of densttutonalzaton were evdent n May 1998, when the mltary played lttle role durng the popular uprsngs that brought the regme down (Callahan, p. 15). These examples come from the government sde, but the logc apples as strongly to nsurgent armed forces: coup threats constran nsurgent leaders decsons about the organzaton of nsurgent mltary forces. For example, n 2002, an ethnc Tuts, Laurent Nkunda, commanded the 7 th Brgade of the Congolese Natonal Army the rebel army fghtng the government of Laurent Kabla n the DRC. After a faled coup attempt by non- Tuts members of the brgade, he purged the brgade of non-tutss, executng at least 160, reducng the coup threat at the expense of hs capacty to project force (Oxford Analytca, October 24, 2007, Congo-Knshasa: Nkunda key to eastern securty ). The fear of coup threats (aganst ether government or nsurgent leaders) also provdes a complementary explanaton of the organzatonal decsons of rebel leaders that Wensten (2005) analyzes. He focuses on Uganda s Natonal Resstance Army and Mozambque s Renamo. The frst began wth an ethncally homogeneous leadershp that ntally recruted co-ethnc (Banyaloke) fghters loyal troops, n the context of the model, to whom leaders could most easly make credble commtments and whch were least lkely to engage n coup actvty. The NRA later expanded ts forces by recrutng non- Banyalokes, but focused prmarly on the Baganda, the group wth whch Banyaloke leaders had developed strong tes (e.g., the capacty to make credble commtments). Renamo, n contrast, began wth a hghly ethncally fragmented leadershp drawn from the ethncallymxed exle communty; ths leadershp group recruted from many ethnc groups and areas of Mozambque. Compared to the NRA leaders, they had a more lmted ablty to make credble commtments to ther troops. 6 The motvaton for densttutonalzaton s not documented, but s plausbly one of the followng: a declne n the rsk of nsurgency (reducng the value of an nsttutonalzed mltary); an ncrease n nternal threats to the regme from the mltary; and, related to ths, a potental declne n the wllngness of the regme to share rents wth the mltary.

10 Wensten (2005) emphaszes rents as the explanaton for the dfferent recrutment strateges. The NRA had no access to rents and could only compensate solders wth promses of future compensaton. It recruted solders wth whom the leaders had ethnc tes that made promses of future compensaton more credble. Renamo, n contrast, ntally receved sgnfcant support from Rhodesa and dd not need to rely on promses of future payment. When Rhodesa fell and support declned Renamo then turned to coercon to ensure troop performance. 7 The arguments here are consstent wth ths explanaton, but they also suggest that the NRA and Renamo mght have pursued smlar strateges even f they had had equal access to rents. Only the NRA had an ethncally homogeneous leadershp. Ths gave t the opton of reducng the barrers to collectve acton among ts solders wth an ethncally homogeneous recrutment strategy. Moreover, when t sought to expand the rebel force, t focused on only one group, the Baganda, ensurng that new rebels were smlarly able to act collectvely to enforce agreements. In contrast, the more ethncally fragmented early leaders of Renamo could not easly pursue ths strategy. Dsproportonate recrutment of any one ethnc group would have put leaders from any other ethnc group at a dsadvantage, snce they would be more subject to a coup threat than the co-ethnc leaders. Renamo was correspondngly unable to make credble commtments to solders, nether durng the perod when they had access to ample external assstance, nor durng the perod when they dd not. They reled nstead on hgh current payments when they had access to external assstance and resorted to coerced recrutment not the recrutment of co-ethncs when they dd not. In general, conflct models abstract from coup threats and leader decsons regardng the organzaton of the mltary organzaton. The qualtatve evdence suggests that ths s a potentally mportant omsson. Garfnkel and Skaperdas (2007) revew a seres of conflct technologes. All share the feature that the probablty that one party wns a conflct ncreases n ts own materal nputs (guns) and decreases n the guns deployed by the other party. The effcacy wth whch guns are deployed on each sde s a parameter of these models, not a choce varable. Furthermore, n most models, all rents are avalable for leaders to compensate mltary effort, whether or not leaders control them. Ths mples that leaders can credbly commt to share rents wth troops n the event of vctory. A key determnant of both effcacy and the credblty of compensaton commtments s the degree to whch solders can act collectvely. One straghtforward way to formally ntroduce the organzaton of the mltary nto standard conflct analyses s to allow leaders to choose to rely on two separate armed forces. One of the forces mght be personally loyal to the leader (such as a presdental guard). Leaders can rely on these personally loyal solders not to undertake a coup and to defend them f a coup attempt by other solders occurs. Whle they confront no coup rsk f all of ther armed forces are drawn from ths group, loyal troops are less effectve n conflct. In addton, the supply of loyal troops s lkely to be lmted. 7 An outsde group, such as Rhodesa, that wants to exert strong nfluence over one group n a conflct mght even encourage ethnc dversty, snce the less credble are leader commtments to rebels, the more that leaders must rely on outsde fnancng.

11 The probablty of a successful coup aganst the leader s then gven by pc G, G L g G, G L, pc 0 for G 0. The probablty ncreases n G (the sze of the contngent of professonal solders), falls n G L, and the cross dervatve s negatve: an ncrease n G rases the probablty of coup by less the larger s the number of solders personally loyal to and wllng to defend the leader, G L. The probablty that group wll preval n conflct aganst group j s then gven by the usual functon p G 1, G 2, G L1, G L 2 L, f G, G f G, G f G, G Lj j Lj where f s the conflct functon that translates materal nputs nto mltary effectveness and s assumed dentcal for both groups. The probablty of group s success rses wth ts own conflct effectveness, f G, G L, and falls wth group j s, f G j, G Lj. Unlke the coup functon g, the conflct functon f ncreases n both G and G L, though ncrements to G contrbute more to the probablty of conflct success than fundng for loyal solders G L ( f G > f G L for G = G L ). Agan the cross-dervatve s negatve: addtonal solders contrbute less to vctory the larger s a leader s relance on two, ndependently organzed groups of solders. In the lterature, antagonsts typcally choose G subject to the foregone rents or producton losses they ncur by transferrng resources to appropratve actvtes, as n the models revewed by Garfnkel and Skaperdas (2007). Coup rsk ntroduces a second constrant. The greater s coup rsk, the more reluctant are leaders to mprove ther capacty to conduct nsurgency or counter-nsurgency. To see ths most smply, t s useful to assume that all conflct actvtes are fnanced out of exogenous rents. The leader of group chooses G and G L to maxmze the returns from conflct, V G1, G 2, G L1, G L 2 p pc R G G L, where expected leader rents are a functon of the probablty of prevalng n conflct, p, less the probablty of beng expelled by a coup, p c. Ths yelds frst order condtons (holdng constant the response of group j), p G p c (1) R 1 0 p G G c (2) R 1 0 p c G L p G L. From (1), the greater the threat that addtonal regular forces pose to the leader,, the fewer of them the leader wll recrut. From (2), the larger the contrbuton that addtonal expendtures on personally loyal solders make to coup preventon, more of them the leader hres, even though ther addton to the leader s armed forces contrbutes less to conflct success than addtonal expendtures on regular forces. p G c L, the

12 p G Coup threats also attenuate leader responses to rents. Rewrtng (1) as p c G 1 R and under the usual assumptons that p G s declnng and p c G ncreasng n G, t follows mmedately that coup rsk suppresses leader nvestments to pursue rents, holdng constant the response of the other group. The leader can only ncompletely offset ths effect by nvestng n a parallel securty servce, snce ths nvestment contrbutes less to overall mltary effectveness than nvestments n a untary mltary staffed by regular troops. 8 Ths smple setup makes two substantal assumptons about rents. Frst, coup threats are exogenous. If they were determned wthn the model, they would lkely rse wth the rents at stake. Ths would tghten the coup constrant confrontng the leader and make hm more lkely to devate from organzatonal choces that optmze mltary effectveness. Second, leaders ablty to fnance mltary expendtures s assumed to be lmted by the total rents at stake, not the rents that the leader controls. Ths s reasonable f the leader can make credble commtments to pay out of future rents n the event that hs group prevals. However, future commtments to the professonal solders, G, become less credble as the leader reles more on loyal solders, G L. The cost of relyng on professonal solders s therefore not constant, but rsng n the fracton of loyal solders that the leader hres. The model abstracts from another mportant characterstc of real world nsurgences to whch future research mght turn: coup threats and the cost n terms of mltary effectveness of relyng on loyal troops (that s, the functon f ) are heterogeneous across groups. The NRA could rely on an ethnc strategy as a bass for credble commtment between leaders and troop; Renamo could not. Laurent Nkunda could rely on the ethnc strategy (many of hs relatves were klled n ant-tuts uprsngs, makng hs appeals to Tuts solders partcularly credble), but also on relgon. In Mass [n 2004] Nkunda a longstandng born-agan Chrstan began to receve heavenly vsons nformng hm that he was to be a savor for all Congolese Tutss. Hs men belevng they were on a dvne msson entered Bukavu n June 2004 followng Nkunda s clam that a genocde of Tutss was takng place there... (Oxford Analytca, October 24, 2007, Congo-Knshasa: Nkunda key to eastern securty ). In none of these cases could government leaders rely on smlar appeals to expand ther stock of loyal troops. These consderatons are partcularly mportant for the emprcal analyss of conflct, whch tends to rely on measures of fnancal strength or access to rents to compare army and nsurgent capacty. However, a small, poorly funded group of solders that beleves the commtments of ts leaders and presents lttle coup rsk may have greater mltary capacty than a large and well-funded group for whch these condtons are not met. Credble commtment, nvestment and ncome shocks Leader decsons to allow collectve acton by ctzens and securty forces nfluence a number of debates n the lterature. One of these concerns the roles of ncome and rents n conflct. The most well-establshed emprcal regularty n the study of conflct s that poor countres are more lkely to experence conflct than rch countres: 75 percent of conflcts 8 The model n the prevous secton abstracts from strategc consderatons by coup-plotters. However, rents rase the payoffs to coups just as they do n the case of nsurgency, exacerbatng the reluctance of leaders to allow ther armed forces to act collectvely.

13 from 1975 2000 occurred n the poorest 50 percent of countres. Negatve ncome shocks are also frequently assocated wth a hgher rsk of conflct. Smlarly, natural resource rents seem to be assocated wth greater rsk of conflct, though sgnfcant outlers (e.g., Norway) rase persstent ssues about the mechansms through whch ths occurs. 9 The ncome-conflct assocaton s not easy to explan. Fearon (2008) observes that n poor countres, the costs of buldng mltary capacty (the reservaton wages of potental recruts) are low, but so also are the potental rewards from conflct. Several explanatons have emerged to grapple wth ths paradox, ether by pontng out that ncome shocks need not fall proportonately across all sectors (labor- and captal-ntensve) of the economy, or that temporary ncome shocks have a larger effect on the costs of wagng conflct than on the future rents earned by conflct success. All abstract from ntra-group collectve acton. Dal Bó and dal Bó (forthcomng) observe that ncome shocks may occur n the captal- or labor-ntensve sectors of the economy. Because conflct s labor-ntensve, ncome shocks to captal-ntensve sectors have a larger effect on the stakes n conflct; shocks to the labor-ntensve sectors have a larger effect on the costs of wagng conflct. Dube and Vargas (2007) fnd evdence for ths logc n Colomba. Coffee s labor-ntensve, and volence fell n coffee-producng regons when the nternatonal prce of coffee ncreased; ol s captal-ntensve, and volence ncreased n petroleum-rch regons of the country when the prce of ol rose. Besley and Persson (2009a) make a smlar argument and show that ncreases n the GDP-weghted prces of countres commodty exports ncrease the probablty of conflct. These analyses are focused on ncome shocks, rather than the large average dfferences n conflct ncdence across rch and poor countres. However, to the extent that poor countres are more relant on commodty producton than rch countres, they help to explan the hgher frequency of conflct n poor countres. Dfferences across countres n the ablty of ctzens and armed forces to act collectvely can also explan these results, however. The decson by leaders to lmt the organzaton of ctzens and armed forces has four effects. Three of these arse because ctzens who cannot act collectvely are more exposed to opportunstc behavor by leaders. Frst, ctzens are more reluctant to nvest, so ncomes are lower. Second, they are more lkely to engage n labor-ntensve producton for the domestc market, whch s less vulnerable to expropraton. Thrd, any captal-ntensve producton that occurs s lkely to be concentrated n hgh-return actvtes, offsettng expropraton rsk, such as copper or gold extracton where the copper or gold are closer to the surface or the ores are partcularly rch. The fourth effect of leaders restrctons on collectve organzaton s clear from the earler dscusson: t reduces mltary readness (makng nsurgency less costly) and forces the leader to rely on current rents to sustan mltary effort (hence the assocaton of rents and conflct). Taken together, these could explan both the ncome-conflct assocaton (countres where collectve acton by ctzens and armed forces s more dffcult are both poorer and more vulnerable to nsurgency), and the assocaton of ncome shocks wth conflct (countres where collectve acton s dffcult are lkely to have larger labor-ntensve sectors sttng sdeby-sde wth hgh rent commodty exports). 9 The mcro evdence on ncome and volence s less clear, however. Berman, et al. (2009b) fnd robust evdence of a negatve relatonshp between unemployment and attacks on government forces n both and the Phlppnes.

14 Some evdence s consstent wth ths alternatve vew. One s that weak constrants on leader s actons towards ther own group are characterstc of conflct countres. Data already cted show that conflcts occur dsproportonately n countres wthout compettve electons. Subjectve measures of opportunstc behavor by governments (the rule of law and corrupton ndcators from Internatonal Country Rsk Gude, avalable for 39 conflct countres) tell a smlar story: they stood at 2.1 and 2.6 the year before conflct, compared to 3.8 and 3.5 for all non-conflct country/years (hgher scores ndcate greater rule of law and less corrupton; the maxmum s sx). Chassang and Padró Mquel (2009) observe that a negatve productvty shock reduces the reservaton wages of potental mltary recruts. However, because shocks are transtory, they have a neglgble effect on the value of future producton around whch conflcts are fought. Two groups each choose a subset of ther members to fght the other group. Non-fghters contnue to be productve, payng wages to fghters suffcent to cover ther opportunty costs of foregong productve actvty. A negatve economc shock reduces ths opportunty cost more than t does the future rents from productve actvty, trggerng conflct. However, ths asymmetrc effect depends on another, mplct assumpton: that armed ctzens refran from usng ther mltary advantage to exproprate a share of all future rents. The greater the share of future rents that armed members of the group can capture, the less t s the case that temporary ncome shocks reduce the costs of conflct more than the stakes of conflct. Ths mplct assumpton holds precsely n countres where armed forces are dsorganzed and less able to enforce ther clams on future rents. Among countres where ths s lkely to be the case, the observed lnk between ncome shocks and conflct s predctably strong. Mguel, et al. (2004) show that negatve ncome shocks, nstrumented by ranfall, have a large, postve effect on the probablty of cvl war n Sub- Saharan Afrca. They emphasze that ths regon s approprate for ther test because t reles heavly on ran-fed agrculture. It s also approprate, however, because n ths regon, armed groups are partcularly unable to enforce clams to future rents. In settngs where they can, ncome shocks would be expected to have a smaller effect. In OECD countres, the armed forces are exceptonally well-organzed and yet do not make large clams on economc rents. The condtons under whch armed forces exercse restrant n predatng on the populaton are another area where more research s needed. However, one promsng explanaton for ths restrant s the level of organzaton of unarmed ctzens. In these countres, the capacty for collectve acton by ctzens, whether through well-establshed poltcal partes or local governments, allows them to more easly resst mltary efforts to extract a share of future rents. It also allows them to punsh leaders who tolerate undscplned armed forces, however, an explanaton for hgh levels of organzaton of securty forces. The evdence n Besley and Persson (2009) can be nterpreted n a smlar lght. They demonstrate that hgher prces for a country s commodty exports ncrease the rsk of conflct only n countres wth low scores on the Polty IV ndex of executve constrants. They take ths as support for ther argument that rents cause conflct when poltcal checks and balances are absent that would compel governments to share rents wth the opposton. The alternatve nterpretaton the emerges from the dscusson here s that weak executve constrants are an ndcaton that ctzens cannot act collectvely to restran leaders, so that leaders refran from sharng rents wth everyone, and not only the opposton. In such

15 countres, conflct rsk ncreases followng a rse n commodty rents because leaders who do not allow collectve acton are more lkely to have dsorganzed mltares. Insurgents confront lower costs of volently pursung hgher rents n these countres than n countres that exhbt sgnfcant constrants on the executve. The valdty of ths alternatve nterpretaton depends on what, exactly, the Polty measure of executve constrants captures. Keefer (2010b) shows that the subjectve executve constrants measure s hghly correlated wth objectve measures of poltcal checks and balances, consstent wth the nterpretaton tradtonally gven the varable n the lterature. However, even after controllng for checks and balances, t s just as strongly assocated wth the age of the largest government party and the years of contnuous compettve electons, both of whch relate to the degree to whch ctzens can act collectvely to hnder opportunstc behavor by leaders. Other mplcatons: Predatory behavor, chld solders, state capacty, ethncty and servce delvery Leaders decsons to allow collectve acton affect numerous other ssues n the study of conflct. Two are the problems of predaton on cvlan populatons and the forced recrutment of chld solders. Others relate to arguments that state capacty or ethnc fragmentaton rase conflct rsk. A ffth s the degree to whch external agences should focus on mproved servce delvery as a way to defuse the rsk of conflct recurrence n postconflct countres. Ths secton brefly revews these. Predaton Predaton by both solders and rebels on the local populaton s well-documented n many conflct settngs, ncludng the DRC. 10 Gates (2002) argues that predaton s more lkely when leaders cannot easly control shrkng by ther solders n ther conflct effort, makng them more tolerant of lootng. 11 Humphreys and Wensten (2006) argue that organzaton wthn fghtng unts (rather than the ablty of leaders to supervse fghtng unts) drves predaton. The evdence they assemble from Serra Leone shows that fghtng unts that exhbted less dscplne (e.g., more drunkenness or wthn-unt conflct) were more lkely to predate on cvlans. Kalyvas (2006) argues that volence aganst cvlans could also be spurred by prvate efforts to settle scores. In all three approaches, predaton s the consequence of the nablty of leaders to prevent ndvdual fghtng unts or solders from takng prvate actons that undermne, or at least do not contrbute to, overall conflct success. The earler dscusson about collectve acton bears drectly on the contractual relatonshp between leaders and solders. If predaton hurts conflct success, and f leaders can credbly promse future rewards based on conflct success, solders are less lkely to predate. They are also more lkely to support leader punshment of solders who predate, snce such solders threaten the rewards of all solders. However, leaders (ether of the entre group or of fghtng unts wthn the group) who fear a coup threat are lkely to dscourage 10 Predaton here s understood as non-strategc volence undertaken by fghtng unts or ndvdual solders n pursut of ther own prvate objectves. 11 Padró Mquel and Yared (2009) examne a related ssue n dervng the optmal strategy of a government nterested n quellng local dsturbances through the use of local actors who can do ths at lower cost.