Media Networks and Political Accountability: Evidence from Radio Networks in Brazil

Transcription

1 Meda Networks and Poltcal Accountablty: Evdence from Rado Networks n Brazl Horaco A. Larreguy A. Joana C. M. Montero June 4, 2014 We examne how dfferent types of meda structure contrbute to poltcal accountablty, analyzng the role of local rado statons, regonal rado networks and televson statons n compensatng for the bas of Brazl s federal government aganst non-algned muncpalty mayors when allocatng drought relef ad. Theoretcal and emprcal evdence shows that rado networks play a central role n reducng federal government bas due to ther ablty to exchange nformaton among afflated rado statons operatng n dfferent muncpaltes. Rado networks, however, are only effectve when they contan non-algned muncpaltes that receved federal ad; only nformaton from such muncpaltes reveals the federal government s bas. JEL: H84, D72, I38 Key words: meda networks, poltcal accountablty, dsaster relef. Ths paper benefted from helpful conversatons and suggestons from Daron Acemoglu, Esther Duflo, Danel Hdalgo, John Marshall, Dan Posner, Pablo Querubn, Ashsh Shenoy, Jm Snyder, Felpe Soutello, and Tavneet Sur. Partcpants at the Harvard Brazl Studes Program Semnar, the MIT Poltcal Economy Workshop and LACEA- PEG 2013 provded essental feedback. Lva Almeda and John Marshall provded excellent research assstance. We are greatly ndebted wth Danel Mederos who facltated the meda network data from Donos da Mda. Alejandra Menchaca and Rafael Aquno provded support and patence throughout the project. All errors are our own. Department of Government, Harvard Unversty, Cambrdge, MA (hlarreguy@harvard.edu) Brazlan Insttute of Economcs, Getúlo Vargas Foundaton, Ro de Janero, Brazl (joana.montero@fgv.br) 1

2 1 Introducton A central queston n poltcal economy s how to get elected offcals to represent the preferences of ther consttuents. 1 A recent lterature stresses the role of meda n holdng poltcans accountable. 2 However, meda outlets do not necessarly unformly dstrbute neutral nformaton to everyone. Technology can determne who gets access to the meda, whle meda ownershp and poltcal afflaton can nfluence ts content (Djankov et al. 2003; Besley and Prat 2006; Gentzkow et al. 2006; Gentzkow et al. 2011; Stromberg 2004). However, lttle s known about the way meda networks affect the transmsson of nformaton, and thus ts mpact on poltcs. Ths s partcularly mportant because changes n technology and large fxed costs of producng news create ncentves for meda frms to own or afflate wth other meda outlets, whch nfluences how nformaton s generated and dstrbuted (Hamlton 2006). In ths artcle, we show that networks that allow for nformaton exchange between meda outlets operatng n dfferent muncpaltes enhance poltcal accountablty. Addtonally, we hghlght that the partsan tes between local and federal government n the dfferent muncpaltes to whch nformaton flows s central for meda to hold poltcan to account, an ssue prevously overlooked by the lterature. Motvated by our Brazlan emprcal applcaton, we develop a smple model n whch a potentally based federal government allocates dsaster relef support to affected muncpaltes. Muncpaltes dffer n that some are governed by mayors that are poltcally algned wth the federal government. Furthermore, relef s effectve only f a muncpalty s mayor has hgh ablty, where a mayor s ablty s observed by the federal government but not voters. Whle an unbased federal government allocates ad to all hgh ablty mayors, a based federal government whch maxmzes the number of co-partsan mayors, rather than voter welfare may wthhold ad from hgh ablty mayors n non-algned muncpaltes. However, the federal government s behavor s constraned by meda networks whch report the occurrence of dsasters, federal ad and resource management to muncpaltes wthn ther network. Ultmately, voters must decde whether to re-elect ther 1 Semnal work n ths area ncludes Barro (1973); Fearon (1999); Ferejohn (1986); Holmstrom (1999). 2 See Prat and Stromberg (2013) for a revew on the poltcal economy lterature on the nfluence of mass meda on poltcs and polcy. 2

3 ncumbent mayor on the bass of ther belefs about the mayor s ablty, whch n turn depends upon belefs regardng the federal government s bas. The model shows that whether a based federal government supples ad to non-algned muncpaltes wth hgh ablty mayors depends on the geographc coverage of the meda networks operatng locally. The key factor s ther coverage of other non-algned muncpaltes affected by dsasters. If network coverage s large, a based federal government needs to provde support to a large enough fracton of non-algned hgh ablty mayors to prevent voters from nferrng that t s based. If network coverage s small, the federal government can get away wth provdng no support to non-algned hgh ablty mayors. More specfcally, the model predcts that the probablty that a non-algned muncpalty covered by meda networks receves ad from a based federal government after experencng a dsaster s ncreasng n the number of non-algned muncpaltes covered by the networks and the number of other non-algned muncpaltes covered by the network that smultaneously receve ad. To test these predctons, we examne how rado networks can contrbute to poltcal accountablty by compensatng for poltcal bas n the allocaton of federal drought relef ad n Brazl. Lke those n many other countres, Brazl s federal government often favors algned consttuents n order to ncrease ther lkelhood of stayng n offce. 3 In partcular, Brollo and Nanncn (2012) show that the federal government transfers fewer resources to non-algned mayors to hurt ther re-electon chances. Ths prevous evdence and the wde varety of meda sources n the country ndcate that Brazl represents an deal settng to nvestgate the effects of meda structure on poltcal accountablty. The sgnfcant red tape nvolved wth provdng federal drought relef ad ensures that t can be subject to consderable poltcal dscreton. We present evdence suggestng that the federal government explots the features of ths bureaucratc process to favor algned muncpaltes and wthhold assstance from non-algned muncpaltes. Brazlan voters receve nformaton about dsasters and the responsveness of the federal govern- 3 Theoretcal models ratonalzng such behavor nclude: Acemoglu and Robnson (2001); Cox and McCubbns (1986); Dxt and Londregan (1996); Lndbeck and Webull (1987); Lzzer and Persco (2001); and Robnson and Verder (2013). Extensve emprcal evdence ncludes Arulampalam et al. (2009) n Inda, Vega and Pnho (2007) n Portugal, Costas-Pérez et al. (2011) n Span, and Ansolabehere and Snyder (2006) and Berry et al. (2010) n the Unted States. 3

4 ment to these events from the meda. Local and regonal ssues are usually dscussed by commercal rado statons, whch broadcast the bulk of local news n Brazlan muncpaltes. In 2009, 41% of Latnobarometro respondents frequently lstened to the rado, whle 66% of these reported regularly dscussng poltcs wth frends, famly and colleagues, and thus were lkely to spread poltcal nformaton they lsten n the rado. 4 Whle most rado statons are ndependent and just reach a local audence, 5% are connected to regonal networks coverng around 20 muncpaltes on average. These rado statons are connected to a central staton that collects nformaton, whch t then dstrbutes by satellte to local statons. Therefore, a rado staton that s connected to a network can both receve nformaton about other places and add content that s consdered relevant to a larger regonal audence. Conversely, televson statons are almost always afflated to a network, whose content they retransmt and usually do not affect. We then study how the presence and geographc coverage of network-connected rado statons compensate for the federal government bas n the dstrbuton of drought relef ad aganst non-algned muncpaltes through ther ablty to spread local news to a larger audence. We test the model s emprcal mplcatons by nvestgatng the probablty of recevng federal relef after experencng low ranfall. The man outcome varable s an ndcator of whether the federal government declared a drought-related state of emergency, whch s a necessary condton for a muncpalty to receve federal support, n a muncpalty between 2002 and Our dentfcaton strategy leverages plausbly exogenous wthn-muncpalty varaton n precptaton levels. Lower ranfall ncreases the lkelhood that a muncpalty receves federal support upon request. To test the model predctons, we explot varaton across tme and space n muncpal poltcal algnment wth the federal government, the presence network-connected rado statons, and network s coverage of other non-algned muncpaltes. The results show that rado networks play an mportant role n supportng poltcal accountablty. Frst, we establsh the exstence of federal government bas aganst non-algned muncpaltes, demonstratng that such muncpaltes are fve percentage ponts less lkely to receve relef after 4 Addtonally, 41% of Latnobarometro respondents read a newspaper, whle 77% watched televson. 5 Data on the actual transfers s not avalable, but declaraton muncpal state of emergency s a good proxy for fund release. 4

5 experencng low precptaton. Second, we show that the presence of a network-connected rado staton almost exactly compensates for ths bas on average. Importantly, nether the presence of local rado nor the presence of natonally-orented televson statons compensate for the federal government bas. Thrd, we show that the effect of network-connected rado statons ncreases wth the number of non-algned muncpaltes covered by the network. Furthermore, as predcted by the model, the presence of a network-connected rado staton n a non-algned muncpalty compensates for poltcal bas only when there are other non-algned muncpaltes wthn the network that receve ad at the tme the muncpalty s ht by a drought. Our fndngs are further supported by addtonal tests demonstratng the robustness of our results. We frst explot nformaton on the tme when rado statons enter and leave rado networks to construct a placebo network connecton. The placebo consders as connected muncpaltes that are not afflated to a network at a gven tme but are at some pont of our perod of analyss. Our man results vansh when we conduct ths placebo exercse, supportng the clam that the presence of network-connected rado statons captures ther ablty to dsclose nformaton about local problems rather than unobservable muncpal characterstcs. Second, we show that our results cannot be explaned by concentraton n the ownershp of network-connected rado statons, or by ther manpulaton by local poltcans. Lastly, we show that covarates do not predct the presence of network-connected rado statons n non-algned muncpaltes or changes n the presence of network-connected rado statons. Ths paper s man contrbuton s to the poltcal economy lterature that addresses how dfferent types of meda contrbute to poltcal accountablty (Besley and Burgess 2002; Stromberg 2004; Snyder and Stromberg 2010; Gentzkow 2006; Gentzkow et al. 2011). Ths lterature has emphaszed that the man dfference among meda types s the extent to whch they cover local poltcs. 6 However, we offer a conceptually dstnct explanaton for our fndng that the effect of meda on poltcs depends on meda type. Our study ponts out that some meda types have a dfferental role n mprovng poltcal accountablty relatve to others because ther structure allows the exchange of nformaton among meda outlets, whch facltates the dssemnaton of nformaton. Addtonally, 6 See Prat and Stromberg (2013) for a dscusson of ths matter. 5

6 we emphasze the mportance of the partsan algnment of the mayors of the muncpaltes covered by these dfferent meda outlets n voters belef formaton process. The remander of the paper proceeds as follows. In Secton 2, we provde background nformaton on the poltcal context, the dsaster relef polcy and the meda market n Brazl. In Secton 3, we develop a model to gude the emprcal analyss. In Secton 4, we present the data before detalng our emprcal strategy. In Secton 5, we shows our man results, present evdence on the mechansm that explans the effect of meda, and rule out alternatve explanatons. Secton 6 concludes. 2 Insttutonal Background 2.1 Poltcal Context Brazl s a federal republc wth three layers of government: federal, state and muncpal. Brazl s 5,565 muncpaltes whch are contaned n ts 26 states and the federal dstrct of Brasla are governed by mayors who are mportant poltcal players n the country s decentralzed poltcal system. In 2013, the federal government transferred 25% of ts tax revenue to muncpaltes to provde publc servces such as basc educaton, prmary health care, and local nfrastructure. 7 In addton to ther role n provdng local publc servces, mayors are mportant poltcal brokers because they draw votes for ther party canddates n other poltcal races. Ames (1994) fnds evdence that n the frst round of the 1989 Presdental electon every canddate receved a bg boost n muncpaltes headed by a mayor from the canddate s party. Avelno et al. (2012) analyze close races n the 2010 electons for state and natonal congresses, and fnd that canddate performance was stronger n muncpaltes governed by a mayor from the same party. Snce the Workers Party (PT) ganed control of the federal government n 2003 t has ncreased ts local poltcal base. Scholars argue that the PT recognzes the correlaton between local support and performance at state and federal electons, and delberately nvests to strengthen ts local presence. Snce the 1980s, the PT has opened local offces to ncrease membershp, mprove the 7 Federal and state transfers represents on average 92% of muncpal revenue, whle tax revenue corresponds to only 6%. 6

7 party s vote share and ncrease partsan attachment. (Amaral 2011; Rbero 2010). Ths strategy was renforced after the successful 2002 Presdental electon, whch allowed the party to strengthen ts party label and ts fnances (Rbero 2010). Electoral fgures from the 2000s ndcate that the PT experenced the return of ths strategy. The number of muncpaltes controlled by the PT doubled n the 2004 electon, and then ncreased an addtonal 37% n the 2008 electon to reach 10% of Brazlan muncpaltes. In general, muncpaltes wth PT mayors are less prone to drought and score slghtly above average n terms of socoeconomc development. Together, ths evdence ndcates that the PT has strong electoral ncentves to ncrease the number of algned mayors. In a context where the PT only drectly controls a small share of muncpaltes, a potental strategy for ncreasng ths number s to lmt the resources avalable to non-algned mayors. Ths could damage the mage of rval partes n local poltcs. Supportng ths logc, Brollo and Nanncn (2012) analyze close electons n Brazl between 2000 and 2004 and show that the federal government transfers fewer dscretonary funds for nvestment n nfrastructure to non-algned mayors. 2.2 Dsaster Relef Polcy Droughts are the most common natural dsaster n Brazl and requre emergency actons from federal, state and local governments. 8 A central responsblty of the Natonal Cvlan Defense and Emergency System, managed by the Mnstry of Natonal Integraton, s to support local governments to deal wth the consequences of water scarcty. Ths federal-level support ncludes the supply of water trucks, food dstrbuton, and temporary cash transfers ( bolsa estagem ). 9 Crucally, muncpaltes can only receve ths support f they request and obtan a declaraton of a state of emergency from the federal government. A state of emergency s declared n 8% of muncpaltes every year. The process of declarng a state of emergency s subject to sgnfcant red tape, whch allows room for dscreton. To declare an emergency, a muncpalty has to send to the federal government 8 Droughts represent 57% of the cases n whch the federal government declared a state of emergency n Brazlan muncpaltes between 2003 and The federal government may also allow farmers to renegotate agrculture debts or redeem agrculture nsurance. 7

8 several documents that prove the severty of the dsaster and ther lack of fnancal ablty to deal wth t. These documents encompass nformaton on the characterstcs of the dsaster, the area affected, the estmated losses, and the measures adopted by the muncpal and state governments, as well as data on populaton, economc ndcators and muncpal budget (CEPED 2012). 10 Because of the bureaucratc nature of the emergency declaraton process there are several ways through whch the federal government can ad alles and restrct help to non-alles. Whle the federal government can expedte a declaraton of emergency and approve t before even analyzng the documentaton, t can also delay the process by requrng addtonal nformaton or even deny assstance by clamng that the muncpal government has enough fnancal capacty to deal wth the consequences of the drought. 11 The popular press provdes examples of the poltcal dscreton that takes place n the emergency declaraton process. On February 17th, 2012, an artcle n Globo Newspaper ndcated that federal audtors found evdence that the Mnster of Integraton authorzed relef transfers to several muncpaltes before recevng a techncal report that quantfed the damage and the resources necessary for reparatons. Ths decson benefted sx muncpaltes n Baha n 2009, the state of the Mnster at the tme, and four of them were algned to the Presdent s or Mnster s party. 12 Federal audtors found notes where techncans clarfed that the techncal reports should be fled wth dates that preceded the authorzaton of transfers. Federal audtors pont to the understaffng of the Mnstry of Natonal Integraton as a key explanaton for dscreton n the allocaton of publc funds. However, the meda exposure of dsasters may constran poltcal dscreton. Recent research has shown that voters are senstve to nformaton on dsasters, and punsh or reward poltcans at the ballot box dependng upon dsaster relef response (Achen and Bartels 2004; Bechtel and Hanmueller 2011; Chen 2013; Cole et al. 2012; Healy and Malhotra 2009). In Brazl, ths s no 10 In addton, the mayor has to enact a decree declarng a state of calamty at the muncpal level, whch also allows the muncpalty to expedte muncpal procurement. 11 The federal government recognzed that the natonal system of natural dsaster relef was too bureaucratc, and smplfed the rules n 2010 and agan n Snce data were unavalable beyond 2008, we cannot explot ths nsttutonal change. 12 The decson benefted Caru (R$ 1,2 mllon), Lauro de Fretas (R$ 7mllons), Mascote (R$ 600 k), Valenca (R$ 700k), Conde (R$ 1 mllon) and Smoes Flho (R$ 1 mllon). Source: ntegracao-lberou-verba-contra-enchentes-antes-de-parecer-tecnco

9 dfferent. Accordng to a meda specalst that we ntervewed, professonals workng n poltcal marketng strongly advse poltcans that they cannot gnore natural dsasters. Ths mght explan why at the outbreak of major dsasters the Presdent and Governor themselves usually vst affected regons. 2.3 Meda n Brazl Rado statons are an mportant source of poltcal nformaton n Brazl. Accordng to Latnobarometro, 40% of Brazlans use rados as a source of poltcal nformaton. Furthermore, voters use the nformaton they receve through rados to evaluate poltcans performance and to remove corrupt poltcans from offce (Ferraz and Fnan 2008). Most rado statons are ndependent and just reach a local audence, but about 600 (17%) have been connected to rado networks between 2002 and The crucal characterstc of network-connected rado statons for our paper s that they share content. A central staton produces programs that are transmtted by all the rado statons connected to ts network. To that end, t collects from ther afflates nformaton on local ssues that are relevant for a wder audence. The ntensty of ths nformaton flow vares by rado network. Rado networks such as Globo and CBN organze perodc meetngs wth journalsts from afflated rado statons to share nformaton, whle others nteract more durng extraordnary local events lke the ones we study. An example of the type of news broadcast by Rado Globo on October 18th 2011 llustrates how network-connected rados dsclose nformaton about the federal government relef response to natural dsasters and exert poltcal pressure on the federal government to release support. 14 Rado Globo reports an ntervew wth the mayor of Londrna and mentons that the muncpalty had requested support from the federal and state government to fx fve brdges that were damaged by a flood. The news s anchored by a reporter n Rado Globo Londrna and transmtted to the whole 13 In 2008, there were 3,445 commercal rado statons dstrbuted across 1,970 (out of 5,565) muncpaltes. Another 890 muncpaltes were only covered by communty rados, whch are low-power statons wth a maxmum broadcast range of one klometer. Communty rados are normally operated by local cvc groups such as neghborhood assocatons. Our emprcal analyss ncludes all muncpaltes. 14 ( htm) 9

10 Rado Globo AM network that covers 46 muncpaltes n 14 states. The capacty of network-connected rado statons to receve and transmt content dfferentates them from other types of meda. Local rado statons that are not connected to networks provde nformaton on local ssues, but cannot reach other localtes. Televson networks operate under a smlar structure to rado networks, but prmarly retransmt natonal and regonal programs. Network-connected TV statons rarely contrbute to the content of ther network because they have a more structured program and less room to dsclose nformaton on extraordnary local events. 15 Because of ths structure, we antcpate that televson networks cannot play the role that rado networks do; we explot ths as a placebo test n Secton 4. Our dataset comprses 34 rado networks that, altogether, had 700 afflated rado statons n 2012, a marked ncrease from about 350 network-connected rado statons n There s lmted evdence about the drvers of rado network expanson n Brazl. The few artcles that analyze the subject argue that rado network expanson n Brazl s busness-orented (Comasseto 2006; Franquet 2003). Intervews wth managers of rado networks ndcate that new afflatons occur ether by a drect nterest of the central staton or by the desre of local statons to be part of a network. 16 Another factor that fostered the expanson of rado networks was the dffuson of dgtal technology, whch allows the central staton to easly receve and send content to ther afflates. From the pont of vew of local rado statons, the afflaton s of nterest because t helps to reduce the costs of producng content and enables specalzaton n nche markets such as news and specal musc genres (Comasseto 2006). In Secton 4 we assess the correlates of the locaton of networkconnected rado statons and the drvers of network expanson. We show that network expanson s not correlated wth observable muncpal characterstcs that could confound our results. 15 Local content n televson s very lmted: only 11% of the programmng of televson channels s flled wth regonal nformaton provded by the regonal headquarter (Valente 2009). 16 For nstance, four of the rado networks that we ntervewed (Rado Globo, Gaucha Sat, Rede Catolcas de Rados and Banderantes) mentoned that they seek afflates n areas wth low coverage by the rado network and wth growth potental, and receve proposals from local rado statons nterested n jonng the network. 10

11 3 Model Ths secton develops a model to drect our emprcal analyss. In the model, voters are uncertan about the federal government s type whether t s unbased (maxmzes voter welfare) or based (maxmzes the number of co-partsan mayors) and whether the ablty of ther muncpalty s mayor to manage federal dsaster assstance s hgh or low. Only f a mayor has hgh ablty s federal ad effectve. Voters decde whether to re-elect ther ncumbent mayor based upon ther belefs about the mayor s competence. Crucally, such belefs depend upon whether the federal government provdes relef upon experencng a drought, but also on nformaton broadcasted by rado networks about the federal government s response to droughts n other muncpaltes wthn the network. The ntuton of the model s straghtforward. A based federal government must decde whether to provde assstance to hgh ablty mayors n non-algned muncpaltes. Wthholdng ad can cause voters to fal to re-elect non-algned mayors, belevng such mayors to be of low ablty, but only f the federal government s not perceved as too based. By provdng addtonal nformaton about the federal government s actons, the presence of a rado network causes the federal government to provde ad to more non-algned hgh ablty mayors. Through strengthenng the qualty of the sgnal receved by voters, the federal government s bas s decreasng n the number of non-algned muncpaltes contaned wthn a rado network that also suffer a drought. 3.1 Agents, Actons and Preferences Consder a country comprsng N muncpaltes. Each muncpalty has an electorate normalzed to one that lves for two perods, t {1, 2}. 17 For smplcty, muncpaltes are subject to droughts only n perod 1. The probablty that any muncpalty s ht by a drought s δ (0, 1), and drought realzatons are ndependently and dentcally dstrbuted across the country. Let n denote the number of muncpaltes affected by a drought. The federal government decdes whether to allocate ad to muncpaltes that suffer a drought. 17 The two perod model consdered here s consstent wth the fact that Brazlan mayors can only be re-elected once. 11

12 However, the success of ts assstance depends on the ablty of mayors to effectvely manage those resources. For smplcty, µ t {0, 1} denote whether the mayor of muncpalty at tme t s of hgh ablty, and can therefore manage resources effectvely. Whle the federal government knows the ablty of each mayor, ths s not known to voters. The pror belef of voters that a mayor s of hgh ablty s λ (0, 1). We denote m {0, 1} as an ndcator of effectve management of a drought n muncpalty, where m = f µ 1 and f {0, 1} ndcates that the federal government allocates ad to a muncpalty. There are two types of federal government: based and unbased. An unbased federal government derves dsutlty from the droughts that are unsuccessfully managed, N U u = d (1 m ) ψf. =1 where d {0, 1} ndcates that a muncpalty has a drought and ψ (0, 1) s the cost the federal government faces for allocaton ad to a muncpalty. An unbased federal government thus mnmzes unsuccessful drought management by always provdng hgh ablty mayors wth ad. A based federal government nstead derves utlty from the number of muncpaltes controlled by algned mayors, 2 N U b = a t ψf, t=1 =1 where a t {0, 1} ndcates that muncpalty s algned wth the federal government at tme t. The dfference n the preferences of the federal government types captures, n a reduced form way, the dea that based federal governments care relatvely more about electoral outcomes. 18 Whether a federal government s based or not s unknown to voters and has ex-ante probablty β (0, 1). Some muncpaltes are covered by rado statons that are connected to a rado network. Rado networks dssemnate nformaton about all muncpaltes wthn the network that were affected by 18 A smoother way to capture such preferences could assgn the federal government a utlty functon gven by U f = (1 ω) N =1 d (1 m) + ω 2 N t=1 =1 at ψf, where ω ndcates the relatve weght that a federal government assgns to electoral outcomes. A based federal government would have a strctly larger ω than an unbased one. We consder the extreme case where based and unbased federal governments have weghts 1 and 0, respectvely. 12

13 a drought and receved federal government ad. Voters that experence a drought are then able to observe not just whether the federal government provdes them wth ad, but also ts behavor n the other muncpaltes covered by ther same rado network. For smplcty, we consder the case where a gven muncpalty s covered by a sngle rado network coverng n other muncpaltes. Voters suffer a dsutlty from experencng a drought that s unsuccessfully managed n perod 1 and, f they have a hgh ablty mayor, experence utlty π > 0 n perod 2. Specfcally, voter preferences are gven by U v = d (1 m ) + Θ µ 2 π, where Θ {θ, 1} captures ncumbency bas by dscountng any new mayor by θ (0, 1). 19 After observng the nformaton provded by the rado statons that cover ther muncpaltes, voters decde whether to re-elect ther ncumbent mayor. 20 If the ncumbent s not re-elected, a mayor algned wth the central government s elected wth probablty α (0, 1). 3.2 Tmng Perod 1 1. Nature chooses whch muncpaltes suffer a drought. 2. The federal government decdes whether to allocate ad to the muncpaltes that suffered a drought, {f } n =1. 3. The management of each drought, {m } n =1, and perod 1 payoffs are realzed. 4. Rado networks dssemnate nformaton about the wthn-network muncpaltes that were affected by a drought and the federal government ad. 5. Voters decde whether to re-elect or to replace ther ncumbent mayor. Perod 2 19 If θ = 1, and thus there s no partsan bas, voters never re-elect a mayor that performs a poor management of a drought, even when they are sure that they face a based federal government. For emprcal evdence supportng the exstence of ncumbency bas n Brazl, see Avelno et al. (2012). 20 Although voters receve utlty n the frst perod, ther forward-lookng re-electon decson ultmately reflects only ther belefs about the type of ther ncumbent mayor. 13

14 1. Indvduals receve payoff Θ π f the elected mayor s of hgh ablty. 3.3 Characterzaton We search for a perfect Bayesan equlbrum n undomnated strateges. To smplfy the characterzaton of the equlbrum, we note some domnant strateges for both types of federal governments. No type of federal government provdes assstance to affected muncpaltes wth a low ablty mayor. 21 An unbased federal government supples ad to affected muncpaltes that have a hgh ablty mayor regardless of ther poltcal algnment. A based federal government also helps affected muncpaltes wth a hgh ablty mayor when they are poltcally algned. However, a based federal government supports a hgh ablty opposton mayor wth probablty q [0, 1]. We solve ths game usng backward nducton. Voters decde to re-elect ther mayor f the posteror that ther mayor s of hgh ablty, λ, s larger than the probablty that a new randomly drawn mayor s of hgh ablty, λ, dscounted by θ. To form λ, voters use ther own experence m, together wth nformaton they receve from ther rado network about other wthn-network non-algned affected muncpaltes, I {m j } nna j=1, where nna denotes the number of non-algned muncpaltes sufferng a drought. I only ncludes affected but non-algned muncpaltes because only n such muncpaltes do dfferent federal government types face dfferent ncentves; consequently, all other nformaton s unnformatve. If voters observe successful drought management n ther muncpalty (m = 1), regardless of whether they are from an algned muncpalty or not, ther posteror belef that ther mayor s of hgh ablty s λ (1, I ) = 1 and they re-elect ther ncumbent mayor. Ths occurs n all algned muncpaltes wth a hgh ablty mayor. If algned voters observe unsuccessful drought management n ther muncpalty (m = 0), ther posteror belef that ther mayor s of hgh ablty s λ a (0, I ) = 0 and they elect an alternatve mayor. When non-algned voters observe unsuccessful management drought n ther muncpalty, the votng decson s not as straghtforward. On the one hand, the mayor could have hgh ablty but a based federal government dd not provde assstance; ths occurs wth probablty λ (1 q ) β, 21 A based federal government mght want to provde ad to expose low ablty non-algned mayors, but does not do so to prevent full revelaton of ts type. 14

15 where q s the equlbrum fracton of affected muncpaltes wth a non-algned hgh ablty mayor that a based federal government provdes assstance to, and β s the posteror belef that the federal government s based. On the other hand, the mayor could actually have low ablty; ths occurs wth probablty (1 λ). When non-algned voters observe unsuccessful drought management n ther muncpalty, the belef that ther mayor s of hgh ablty s then λ na (0, I ) = λ (1 q ) β λ (1 q ) β + (1 λ). These voters then re-elect ther ncumbent mayor as long as λ na (0, I ) θλ, and thus f β β = θ (1 λ) (1 θλ) (1 q ). (1) That s, re-electon only takes place when voters are suffcently convnced that the federal government s based. As reflected n the followng lemma, β s ncreasng n q. Lemma 1 β s ncreasng n q. All proofs are farly straghtforward and contaned n Appendx A. In words, Lemma 1 establshes that the posteror belef regardng the bas of the federal government requred to re-elect the ncumbent, β, ncreases wth the equlbrum fracton q of affected muncpaltes wth a non-algned hgh ablty mayor that a based federal government asssts. The posteror probablty that voters n muncpalty beleve that the federal government s based, based on ther own experence of drought management and ther experence of those n ther network, s then β = (λq ) sna (λq ) sna (λ (1 q ) + (1 λ)) una (λ (1 q ) + (1 λ)) una β + λ sna β (1 λ) una (1 β), (2) where s na and u na are, respectvely, the number of non-algned affected muncpaltes wthn s network where drought management s successful and unsuccessful. Under a based federal government, E[s na ] = λq n na and E[u na ] = (λ (1 q ) + (1 λ)) n na. Lettng these expectatons hold, 15

16 our next lemma then follows. 22 Lemma 2 Under a based federal government, 1. β s decreasng n q. 2. If q < 1/ (1 + λ), β s ncreasng n n na. The frst part of the lemma states that the posteror belef that the federal government s based s decreasng n the equlbrum fracton of affected muncpaltes wth a non-algned hgh ablty mayor that a based federal government provdes assstance to. Intutvely, f a based federal government provdes ad to a larger fracton of non-algned affected muncpaltes wth a hgh ablty mayor, t s less lkely that voters nfer that t s based. The second part of the lemma ndcates that, when a based federal governments provdes lttle ad to non-algned hgh ablty mayors, a voter s belef that the federal government s based ncreases wth the number of nonalgned muncpaltes n ther rado network. Ths s because the probablty of learnng that other non-algned muncpaltes n s network unsuccessfully manages a drought ncreases. Denote n na,h and n na,l respectvely as the number of wthn-network affected muncpaltes under a non-algned hgh ablty and low ablty mayor. A based federal government effectvely solves the followng maxmzaton problem: { max q α[n na,h } (1 q) τ(q) + n na,l τ(q)], where 1 f β(q) β(q) τ (q) = 0 otherwse ndcates that voters n wthn-network muncpaltes do not beleve that the federal government s suffcently based when re-electng a mayor followng a drought that was not well managed. Defne n na as the n na snce β (q = 0) s not a functon of n na such that β (q = 0) = β (q = 0). Importantly, n na s unquely defned, and from Lemma 2, β (q = 0) s ncreasng n n na. Our frst 22 The expectaton provdes an ncreasngly good approxmaton as N becomes large. 16

17 proposton characterzes the optmal choce of q by a based federal government. Proposton 1 Under a based federal government, f n na q s the unque q > 0 such that β = β. n na, q = 0. However, f n na > n na, When a rado network cannot broadcast nformaton about many non-algned affected muncpaltes where there s unsuccessful management of droughts, a based federal government can get away wth provdng no support to non-algned hgh ablty mayors. Otherwse, a based federal government needs to provde support to a large enough fracton of non-algned hgh ablty mayors so that voters do not nfer that t s suffcently based to affect ther re-electon decson. As the next proposton ndcates, such an effect s ncreasng n the number of non-algned affected muncpaltes covered by the rado network. Proposton 2 Under a based federal government, f q < 1/ (1 + λ), q s ncreasng n n na. The more non-algned affected muncpaltes a rado network can expose a based federal government to, the larger the fracton of those muncpaltes that the government has to help to avod that voters nfer that t s suffcently based. Gven mayor ablty s randomly dstrbuted, p λq denotes the probablty that a based federal government provdes ad to a non-algned muncpalty. Our man theorem then follows. Theorem 1 If q < 1/ (1 + λ), 1. p 2. p s larger f muncpalty s covered by a rado network. s ncreasng n the total number of non-algned muncpaltes covered by the rado network that operates n muncpalty. 3. For a gven number of affected non-algned muncpaltes n a rado network, p s ncreasng n the number of other non-algned affected muncpaltes covered by the network that receve federal ad at the tme muncpalty s ht by a drought. 17

18 Frst, Theorem 1 demonstrates that non-algned muncpaltes connected to a rado network are more lkely to receve ad from a based federal government. Second, greater coverage of nonalgned muncpaltes wthn a gven rado network ncreases the chances of affected muncpaltes wthn the network recevng ad from a based federal government after sufferng a drought. Thrd, a non-algned affected muncpalty covered by a rado network s more lkely to receve ad from a based federal government when more non-algned affected muncpaltes wthn the network are recevng ad. In the next secton we outlne our emprcal strategy for testng the predctons of Theorem 1. 4 Data and Emprcal Strategy In ths secton, we frst descrbe our data before detalng our emprcal strategy desgned to test the emprcal mplcatons of the model. 4.1 Data We use several data sources to conduct our emprcal analyss. Our man outcome varable comes from the Natonal Secretarat of Cvl Defense, and t s an ndcator varable for whether a muncpalty had a state of emergency declared due to drought n a gven year. Durng our perod of analyss 2002 to 2008 the federal government declared a state of emergency due to drought n over 3,200 muncpaltes, whch represent an average of 8% of muncpaltes per year. 23 We dentfy muncpaltes that have experenced a drought by usng nformaton on the monthly level of ranfall at the meteorologcal staton level between 1961 and INMET, the Brazlan Insttute of Meteorology, provdes ths nformaton for 280 statons, whch are llustrated n Fgure 1a. We nterpolate ths nformaton for the whole of Brazl and then calculate the muncpal levels of ranfall. Our man measure of drought s the muncpal ranfall z-score for the Sprng- Summer season, calculated as the devaton from the hstorcal mean normalzed by the hstorcal 23 Alternatvely, a state of emergency due to drought was declared n 1,655 dfferent muncpaltes between 2002 and On average, affected muncpaltes experenced two epsodes of drought over ths perod. 18

19 standard devaton. 24 Snce we focus on low precptaton events, we set postve z-scores to zero. 25 As ndcated n Table 1, the mean level for ranfall z-score n the perod we study s and the standard devaton s Our electon data s from the Superor Electoral Court (TSE). We defne a muncpalty as algned wth the federal government f t s governed by a mayor from the same party. Ths varable may change every two years snce muncpal and Presdental electons take place every four years and are two years apart. 26 Fgure 2a ndcates all the muncpaltes that were algned at a certan pont of our analyss perod. From 2002 to 2008, 92% of muncpaltes were non-algned to the federal government. 27 We dentfy meda presence n each muncpalty by usng nformaton from the Telecommuncaton Natonal Agency (ANATEL), the Brazlan regulator of communcatons, on all meda outlets n Brazl, ther locaton, the date they were lcensed, and the members of ther boards. 28 To dentfy the rado statons connected to networks and ther afflates, we frst collected nformaton from three webstes: Donos da Mda, TudoRado and Wkpeda. 29 We then collected nformaton on the dates when each rado staton became afflated or dsafflated to a rado network by checkng ther webstes or through telephone calls. The Donos da Mda database also allowed us to dentfy whch televson statons are connected to a network. Accordng to our database, 33% of muncpaltes have a commercal rado staton, and 5% have a rado staton connected to a net- 24 We take nto account only ranfall levels durng Sprng and Summer because ths s the man crop season for the majorty of crops cultvated n Brazl. 25 Results usng an ndcator varable for whether there s a drought are qualtatvely smlar. 26 Whle Presdental electons occurred n 2002, 2006 and 2010, muncpal electons took place n 2000, 2004 and Because the PT governed Brazl from 2003 to 2010, n practce our varaton comes manly from power swtches at the federal level n 2003 and at the muncpal level n 2005 and We also calculated alternatve defntons of poltcal coalton whch take nto account whether the mayor s from the same party that controls the Mnster of Natonal Integraton, whch s responsble for the dsaster relef polcy. Results do not dffer when we also consder as algned muncpaltes those that are from the state from whch the Mnster of Natonal Integraton orgnates and are algned to the party that controls the Mnstry. The specfcaton where we also nclude as algned muncpaltes all muncpaltes under the party that controls the Mnstry provdes nosy estmates. Ths s because the nature of the party that controlled the Mnstry over most of the our tme-perod, the Partdo do Movmento Democratco Braslero (PMDB). Whle the PMDB governs a ffth of the muncpaltes n Brazl, t s a very fragmented party controlled by several ndependent regonal bosses. 28 We dentfed all meda outlets n Brazl by consderng all the outlets that appear n the Sstema de Controle de Radofusao (SCR) on the ANATEL webste. Ths webste also provdes the locaton of each outlet and a lst of all documents that have been ssued to each meda outlet. As a proxy for ther lcensng date, we use the date that the frst document n the name of the outlet was ssued. Informaton on partners comes from ANATEL s Sstema de Acompanhamento de Controle Socetaro (SIACCO). We use a lst of partners of meda outlets from Aprl

20 work. Fgure 3a and Fgure 2b respectvely llustrate all the Brazlan muncpaltes where a local and network-connected rado staton operated at a certan pont of our analyss perod. Televson statons are more wdespread, reachng 55% of muncpaltes. 99% of these televson statons are connected to a network and they smply retransmt content. Fgure 3b llustrates all the Brazlan muncpaltes where a televson staton operated at a certan pont of our analyss perod. We calculated the coverage of each network by consderng the number of muncpaltes n whch each network operates. Addtonally, also usng the poltcal algnment of muncpaltes, we computed each network s coverage to algned and non-algned muncpaltes. 30 We also computed the number of non-algned muncpaltes wthn the network that were ht by drought and had a state of emergency declared n a specfc year. 31 Table 1 ndcates that rado networks cover, on average, non-algned muncpaltes. On average there were 3.9 non-algned muncpaltes wthn the network that had the state of emergency declared. Fnally, we gather nformaton from the 2000 populaton census, conducted by the Brazlan Bureau of Statstcs (IBGE), on muncpal characterstcs such as populaton, urbanzaton, populaton densty, ncome per capta, poverty rate and average years of schoolng. We use these covarates n our statstcal analyss to control for muncpal characterstcs that may correlate wth meda presence and poltcal algnment. The FINBRA database, from the Brazlan Treasury, provdes nformaton on muncpal revenue, whch we use to control for the fnancal capacty of a muncpalty to deal wth natural dsasters. 4.2 Emprcal Strategy We now explan the emprcal strategy used to test the predctons n Theorem 1. Our dentfcaton strategy explots varaton n ranfall, as defned above. Snce the parts of the country that generally experence least ranfall could be correlated wth economc and poltcal varables, we explot essentally random wthn-muncpalty varaton n the level of ranfall. The ncluson of 30 We consder the state captals n the computaton of network coverage but we remove these muncpaltes from our sample n the man emprcal analyss. We want to avod state captals drvng the results because all these muncpaltes have a network-connected rado staton and are poltcally mportant. 31 In ths computaton, we consdered that a muncpalty has been ht by a drought f the z-score of ranfall level was less than -1 n the year. 20

21 perod fxed effects ensures that our results are not drven by natonwde ranfall shocks that could be spurously correlated wth poltcal outcomes. 32 Such varaton n ranfall s the startng pont of our analyss, whch focuses on how the federal government s response to drought vares wth the partsanshp of a muncpalty s mayor and dfferent types of meda coverage. We frst seek to establsh the exstence of bas n the federal government s decson to permt drought relef ad. Accordngly, we frst estmate the followng specfcaton: emergency t = β 0 + β 1 drought t + β 2 na t + β 3 drought t na t + X t Γ + η + φ t + ε t, (3) where emergency t s a dummy varable that ndcates whether a drought-related state of emergency was declared n muncpalty n year t, drought t s our normalzed level of precptaton durng the Sprng-Summer season (multpled by 1, so larger values represent a lack of ranfall), and na t s a dummy varable that ndcates that a muncpalty s not algned wth the federal government. η and φ t respectvely denote muncpalty and year fxed effects. X t are a seres of controls, whch nclude nteractons wth drought t to partally address the potental concern that our estmates of nterest capture the effect of omtted varables related to muncpal economc development and state capacty. 33 Standard errors are clustered by state. We examne whether the federal government s based aganst non-algned muncpaltes by analyzng β 3 n equaton (3). If β 3 < 0, non-algned muncpaltes are less lkely to receve drought relef ad. If β 3 = 0, muncpaltes are equally lkely to receve drought relef regardless of ther poltcal algnment. To test the frst part of Theorem 1, our second specfcaton examnes varaton n the federal government s bas by ncludng nteractons wth bnary measures of meda coverage. Specfcally, 32 Omtted results renforce that our drought measure s orthogonal to poltcal algnment; local rado staton, rado network and televson staton presence; and other tme varyng covarates. 33 The controls nclude baselne ncome per capta, poverty rate, muncpal Gn coeffcent, average years of schoolng, nfant mortalty, share of households wth electrcty, muncpal GDP per capta, muncpal revenue per capta, tax revenue per capta, populaton, area, populaton densty, urbanzaton rate, dstance to state captal, and a dummy for beng a coastal muncpalty. 21

22 we estmate: emergency t = β 0 + β 1 drought t + β 2 na t + β 3 drought t na t + β 4j meda j t + β 5j meda j t drought t + j {lr,rn,tv} + j {lr,rn,tv} j {lr,rn,tv} j {lr,rn,tv} β 6j meda j t na t β 7j meda j t drought t na t + X t Γ + η m + φ t + ε t (4) where meda lr t ndcates the presence of a local commercal rado staton, meda rn t ndcates the presence of a commercal rado staton connected to a network, and meda tv t ndcates the presence of a televson staton. Under a based federal government, the model assumes that β 5rn = 0, and Theorem 1 predcts that β 7rn > 0. Our model s slent about β 5lr = 0 and β 7lr, as well as the relatve magntudes of β 7rn and β 7lr, snce the model was not desgned to capture the role of local rado statons, and thus ts relatve relevance wth respect to network-connected rado statons. Ths sad, we expect β 5lr = 0 and β 7lr > 0; and followng the logc of Theorem 1, we also expect β 7rn > β 7lr. That s, we expect that, whle both local and network-connected rado statons compensate for the bas that non-algned muncpaltes suffer when t comes to recevng drought relef, rado statons connected to a network are more effectve at combatng ths bas. Addtonally, we also expect that the model predcts that β 5tv = β 7tv = 0 snce televson statons rarely broadcast local content and do not dssemnate local nformaton through ther network. Our fnal man specfcaton tests the second and thrd parts of Theorem 1 by examnng varaton n the extent of rado network coverage. Accordngly, we estmate, emergency t = β 0 + β 1 drought t + β 2 na t + β 3 drought t na t + β 4 cov rn t +β 5 cov rn t drought t + β 6 cov rn na t +β 7 cov rn t drought t na t + X t Γ + η m + φ t + ε t, (5) where cov rn t represents the coverage of a rado network operatng n muncpalty. To test the second and thrd parts of Theorem 1, we consder two versons of cov rn t. Frst, we quantfy 22

23 cov rn t as the number of non-algned muncpaltes covered by the rado networks that operate n muncpalty. Second, cov rn t s also calculated usng the number of non-algned muncpaltes covered by the rado networks operatng n muncpalty recevng ad from the federal government. In ths last specfcaton, followng the thrd part of Theorem 1, we also nclude as a control the number of non-algned muncpaltes covered by the rado networks that cover muncpalty that are subject to a drought nteracted wth the same nteractons of cov rn t. Theorem 1 mples that n both specfcatons β 7 > 0. 5 Results We frst report the results of our man specfcatons, and fnd strong support for the predctons n Theorem 1. In partcular, we show that federal government drought relef s based aganst muncpaltes wth opposton mayors, but that the presence of a rado network almost exactly compensates for ths bas. Furthermore, the lkelhood that a muncpalty receves drought relef ncreases wth the number of non-algned muncpaltes n a rado network, but only when these muncpaltes actually receve drought relef. We then show that our results are robust to a varety of plausble alternatve explanatons. 5.1 Man Results Table 2 reports the results of our emprcal specfcatons n equatons (3) and (4). Column (1) presents the basc specfcaton n equaton (3), whle columns (2)-(5) present dfferent versons of equaton (4): column (2) nteracts the regressors n specfcaton (4) wth an ndcator for muncpaltes contanng a rado staton connected to a network, column (3) adds the nteracton wth an ndcator for a local rado staton, column (4) adds nstead the nteractons wth an ndcator for a televson staton n the muncpalty, and column (5) adds both the local rado and televson staton nteractons. Supportng the exstence of a based federal government, column (1) demonstrates that nonalgned muncpaltes are 5% less lkely to receve drought relef from the federal government. However, column (2) shows the presence of a network-connected rado staton substantally ncreases the 23

24 probablty that a non-algned muncpalty receves drought assstance. Ths statstcally sgnfcant result fully compensates for the federal government s bas aganst non-algned muncpaltes. 34 Furthermore, the results across columns (2)-(5) suggest that, whle rado statons connected to a network contrbute to the lkelhood that non-algned muncpaltes receve drought relef from the federal government, such an effect s absent n algned muncpaltes. Columns (3) to (5) of Table 2 also show that local rado statons and televson statons have no sgnfcant effect. The estmates on the effect of local rado statons and televson statons are consstent wth the fact that ther content does not reflect events n other muncpaltes. Thus, as expected, the effect of network-connected rado statons s sgnfcantly larger than the effect of local rado statons. In the model, the extent of meda coverage from non-algned muncpaltes wthn rado networks s essental n compensatng for federal government bas. To further test the mplcatons of the model, we test dfferences at the ntensve margn hghlghted by the second and the thrd parts of Theorem 1 usng equaton (5). Table 3 reports these results, where columns (1) and (2) use the number of non-algned muncpaltes covered by the rado networks operatng n a gven muncpalty to measure coverage, whle columns (3)-(6) nstead measure coverage usng the number of these muncpaltes that receve drought relef ad. The results n columns (1) and (2) provde suggestve evdence supportng the clam n Theorem 1, showng that the bas s only reduced when meda network covered non-algned muncpaltes affected by drought. In partcular, column (1) shows that the presence of a network-connected rado staton ceases to be relevant becomng negatve and statstcally nsgnfcant when we control for ncludng the coverage of the rado network. Ths ndcates that t s the coverage of the rado networks to non-algned muncpaltes, and not smply ther presence, that drves our results n Table 2. Ths sad, whle the trple nteracton wth coverage s statstcally sgnfcant n column (2), we lack suffcent statstcal power to reject a null effect when the rado network ndcators are ncluded n column (1). However, nterpreted n terms of our model, ths lack of power s not surprsng. Ths s because coverage of non-algned muncpaltes s not enough on ts own to re- 34 The sum of all the coeffcents nteracted wth Network-connected rado add up to.6, and thus fully offset the federal government s bas (-.59). 24

25 duce bas a crtcal mass of such muncpaltes must also be smultaneously affected by a drought, whch happens relatvely rregularly. Measurng coverage by the number of non-algned muncpaltes that receve ad, the results n columns (3)-(6) provde defntve evdence of the last part of Theorem 1. Columns (3) and (4) ndcate that coverage of non-algned and affected muncpaltes drves the results n Table 2. We thus fnd strong evdence that network-connected rado statons reduce poltcal bas because broadcastng the stuaton n those muncpaltes can reveal the federal government bas. Columns (5) and (6) further support such a clam, showng that rado networks not coverng affected nonalgned muncpaltes that receve drought relef ad do not ncrease the probablty that the federal government provdes drought relef to a non-algned affected muncpalty. 5.2 Alternatve Explanatons The results n Table 2 and Table 3 strongly suggest that network-connected rado statons play a sgnfcant role n compensatng for federal government bas n the dstrbuton of drought relef ad. Despte the nuance of the precedng analyss, t stll remans possble that the presence of rado networks partcularly n non-algned muncpaltes mght capture the effects of omtted muncpal network-connected rado staton characterstcs. To address ths, we frst conduct a placebo analyss. Our placebo measure of connecton to a rado network consders that a muncpalty has a rado staton connected to a network f the staton s not currently connected to a network but has been connected or wll be durng our perod of analyss. Usng ths measure, we followed the same procedure descrbed above to calculate the placebo network coverage. Table 1 ndcates that the placebo rado networks cover on average 27 non-algned muncpaltes. Table 4 presents the results of our placebo tests. Column (1) replcates our estmates of equaton (4) from n Table 2, whle columns (2)-(4) replcate our estmates of equaton (4) reported n Table 3. Across all specfcatons, the placebo measure of a network-connected rado staton does not compensate for the bas of the federal government n drought ad aganst non-algned muncpaltes. We now turn to two potental ways n whch the effect of a network-connected rado staton 25

26 mght nstead capture the fact that network-connected rado statons are owned by a dfferent type of ndvduals. Frst, network-connected rado statons mght be more lkely to be owned by poltcans, and thus manpulated to pursue poltcal goals. 35 Second, the owners of networkconnected rado statons mght be more powerful, and thus better able to exert pressure over the federal government. To address these concerns, we estmate the followng specfcaton: emergency t = β 0 + β 1 drought t + β 2 na t + β 3 meda rn t + β 4 drought t na t +β 5 meda rn t drought t + β 6 meda rn t na t +β 7 meda rn t drought t na t + β 8 meda alt t +β 9 meda alt t drought t + β 10 meda alt t na t +β 11 meda alt t drought t na t + X t Γ + η m + φ t + ε t (6) where meda alt t s defned n two dfferent ways, dependng on the concern that we address. In the frst case, meda alt t s an ndcator varable that a poltcan (mayor, local counclor, or state and federal congressman) owns a network-connected rado staton n muncpalty at tme t, or the poltcan s assocated wth the network through a famly member. 36 In the second case, meda alt t represents an ndcator for a rado staton belongng to a rado network accordng to an alternatve rado network defnton that consders a rado staton part of a network f one of ts board members owns or shares a famly name wth someone on the board of a meda outlet located n another muncpalty. Table 1 ndcates that there are on average 12% of muncpaltes have the presence of rado-statons connected through ownershp. Appendx B explans how ths data was assembled. 35 For nstance, Boas and Hdalgo (2011) fnd that meda control facltates the entrenchment of local poltcans n Brazl. Poltcans have a hgher chance to obtan a lcense of a communty rado staton, and the ownershp of a rado staton substantally ncreases the probablty of wnnng local electons. Stadnk (1991) ponts out that 79 out of 503 Congressmen owned, drectly or ndrectly, a TV or rado staton n In addton, Motter (1994) documents that half of the concessons for televson and rado statons ssued n sx decades were dstrbuted by former Presdent Sarney between 1985 and 1998 and dsproportonally favored poltcans who voted on key legslaton, such as amendments to the 1988 consttuton. 36 Our results do not dffer when we nstead consder separate specfcatons for local poltcans (mayors and local counclors) and congressmen. 26

27 Table 5 presents the results of these tests. The results n columns (1) and (2) show that the fndng that rado networks compensate for the federal governments bas aganst non-algned muncpaltes s not drven by local poltcan s ownershp or assocaton to network-connected rado statons. Smlarly, columns (3) and (4) ndcate that connectvty to rado statons n other muncpaltes through ownershp does not contrbute to the lkelhood that non-algned muncpaltes receve federal drought relef support. Ths fndng s consstent wth our model snce rado statons connected through ownershp do not possess systematc mechansms through whch nformaton can be collected and shared on a frequent bass. Fnally, n Appendx C we corroborate n Table 6 that baselne covarates that correlate wth state capacty and economc development do not predct the dfferental presence of networkconnected rados n non-algned muncpaltes. Moreover, we show n Table 7 that lagged populaton, populaton densty, and GDP per capta do not predct changes n the presence of networkconnected rado statons. Addtonally, unreported results show that droughts do not affect agrcultural output dfferently n muncpaltes wth network-connected rado statons. In sum, whle we lack randomness on the locaton of rado statons connected to a network, these results suggest that t s unlkely that the presence of a network-connected rado staton n a non-algned muncpalty captures the effect of other omtted muncpal attrbutes. Furthermore, these results underscore the predcton of the model that network-connected rado statons are mportant precsely because they dsclose nformaton on dsasters to non-affected places. 6 Concluson In ths paper, we provde evdence that the structure of the meda nfluences how meda affects poltcal accountablty. In partcular, we analyze how the presence and geographcal coverage of rado networks mpact the federal government s ncentve to provde drought relef ad to nonalgned muncpaltes n Brazl. Our central argument s that the structure of rado networks facltates the nformaton flow among afflated rado statons, whch ncreases the salence of the federal government s response to natural dsasters n other muncpaltes covered by rado networks. Rado networks thereby hold poltcans accountable because the federal government cannot gnore 27

28 non-algned affected muncpaltes wthout revealng ts poltcal bas. We develop a model that captures these dynamcs and provde emprcal evdence supportng our argument. Frst, whle the presence of network-connected rado staton compensates for the poltcal bas, nether the presence of local rado statons nor the presence of televson statons nfluence the federal government s decsons. Second, the effect of the rado network ncreases wth ts geographcal coverage of non-algned muncpaltes. Thrd, the rado network presence only plays a role when there s a crtcal number of wthn-network non-algned affected muncpaltes. Fourth, a placebo exercse whch consders as connected only those muncpaltes that had prevously or subsequently became connected to a rado network shows no effect on the federal government s behavor. Fnally, we show that our results are not explaned by the concentraton n the ownershp of network-connected rado statons or by ther manpulaton by local poltcans. Our fndngs ndcate that meda networks play a central role n supportng poltcal accountablty, and consequently have mportant polcy mplcatons. The results hghlght key ngredents for the development of meda regulaton to enhance poltcal accountablty. Addtonally, they suggest mportant gans from promotng the mplementaton of technologes that contrbute to the exchange of nformaton among meda outlets that operate n dfferent poltcal markets. 28

29 References Acemoglu, D. and J. A. Robnson (2001). Ineffcent redstrbuton. Amercan Poltcal Scence Revew 95(3), Achen, C. and L. Bartels (2004). Muscal chars: Pocketbook votng and the lmts of democratc accountablty. Presented at the 2004 annual meetng of the Amercan Poltcal Scence Assocaton. Amaral, O. (2011). Anda conectado: o PT e seus vínculos com a socedade. Opnão Públca 17(1). Ames, B. (1994). The reverse coattals effect: Local party organzaton n the 1989 brazlan presdental electon. The Amercan Poltcal Scence Revew 88(1), Ansolabehere, S. and J. M. Snyder (2006). Party control of state government and the dstrbuton of publc expendtures. Scandnavan Journal of Economcs 108, Arulampalam, W., S. Dasgupta, A. Dhllon, and B. Dutta (2009). Electoral goals and centerstate transfers: a theoretcal model and emprcal evdence from Inda. Journal of Development Economcs 88(1), Avelno, G., C. Bderman, and L. S. Barone (2012). Artculações ntrapartdáras e desempenho eletoral no Brasl. Workng Paper. Barro, R. J. (1973). The control of poltcans: An economc model. Publc Choce 14(1), Bechtel, M. M. and J. Hanmueller (2011). How lastng s voter grattude? An analyss of the shortand long-term electoral returns to benefcal polcy. Amercan Journal of Poltcal Scence 55(4), Berry, C. R., B. C. Burden, and W. G. Howell (2010). The presdent and the dstrbuton of federal spendng. Amercan Poltcal Scence Revew 104(4), Besley, T. and R. Burgess (2002). The poltcal economy of government responsveness: Theory and evdence from Inda. Quarterly Journal of Economcs 117(4),

30 Besley, T. and A. Prat (2006). Handcuffs for the grabbng hand? Meda capture and government accountablty. Amercan Economc Revew 96(3), Boas, T. C. and F. D. Hdalgo (2011). Controllng the arwaves: Incumbency advantage and communty rado n Brazl. Amercan Journal of Poltcal Scence 55(4), Brollo, F. and T. Nanncn (2012). Tyng your enemy s hands n close races: The poltcs of federal transfers n Brazl. Amercan Poltcal Scence Revew 106, CEPED (2012). Atlas braslero de desastres naturas Techncal report, Centro Unverstáro de Estudos e Pesqusas sobre Desastres, UFSC. Chen, J. (2013). Voter partsanshp and the effect of dstrbutve spendng on poltcal partcpaton. Amercan Journal of Poltcal Scence 57(1), Cole, S., A. Healey, and E. Werker (2012). Do voters demand responsve governments? evdence from Indan dsaster relef. Journal of Development Economcs 97, Comasseto, L. R. (2006). O rádo local e a nformação global. Estudos em Jornalsmo e Mída 3(1). Costas-Pérez, E., A. Solé-Ollé, and P. Sorrbas-Navarro (2011). Corrupton scandals, press reportng, and accountablty: Evdence from Spansh mayors. Workng Paper. Cox, G. W. and M. D. McCubbns (1986). Electoral poltcs as a redstrbutve game. Journal of Poltcs 48(2), Dxt, A. and J. Londregan (1996). The determnants of success of specal nterests n redstrbutve poltcs. Journal of Poltcs 58(4), Djankov, S., C. McLesh, T. Nenova, and A. Shlefer (2003). Who owns the meda? Journal of Law and Economcs 46(2), Fearon, J. (1999). Electoral accountablty and the control of poltcans: Selectng good types versus sanctonng poor performance. In Przeworsk, A., Susan C. Stokes and Bernard Mann (ed.) Democracy, accountablty, and representaton, Ch

31 Ferejohn, J. (1986). Incumbent performance and electoral control. Publc Choce 50(1-3), Ferraz, C. and F. Fnan (2008). Exposng corrupt poltcans: The effects of Brazl s publcly released audts on electoral outcomes. The Quarterly Journal of Economcs 123(2), Franquet, R. (2003). Haca un nuevo sstema mundal de comuncacón, Chapter La rado ante la dgtalzacón: Renovarse en la ncertdumbre. Barcelona: Edtoral Gedsa. Gentzkow, M. (2006). Televson and voter turnout. The Quarterly Journal of Economcs 121(3), Gentzkow, M., E. L. Glaeser, and C. Goldn (2006). Corrupton and Reform: Lessons from Amerca s Economc Hstory, Chapter The Rse of the Fourth Estate. How Newspapers Became Informatve and Why It Mattered, pp Unversty of Chcago Press. Gentzkow, M., J. M. Shapro, and M. Snknson (2011). The effect of newspaper entry and ext on electoral poltcs. Amercan Economc Revew 101(7), Gorgen, J. (2002). Sstema central de mída: Proposta de um modelo sobre os conglomerados de comuncação no Brasl. Master s thess, Faculdade de Bbloteconoma e Comuncação, UFRGS. Hamlton, J. (2006). All the News Thats Ft to Sell: How the Market Transforms Informaton nto News. Prnceton, NJ: Prnceton Unversty Press. Healy, A. and N. Malhotra (2009). Myopc voters and natural dsaster polcy. Amercan Poltcal Scence Revew 103(3), Holmstrom, B. (1999). Manageral ncentve problems: A dynamc perspectve. Revew of Economc Studes 66(1), Lma, V. (2006). Mída e Crse Polítca no Brasl. São Paulo: Fundação Perseu Abramo. Lndbeck, A. and J. W. Webull (1987). Balanced-budget redstrbuton as the outcome of poltcal competton. Publc Choce 52(3),

32 Lzzer, A. and N. Persco (2001). The provson of publc goods under alternatve electoral ncentves. Amercan Economc Revew 91(1), Motter, P. (1994). O uso polítco das concessões das emssoras de rádo e televsão no governo Sarney. Comuncação e Polítca 1(1), Prat, A. and D. Stromberg (2013). Advances n Economcs and Econometrcs, Tenth World Congress, Volume 2: Appled Economcs, Chapter The Poltcal Economy of Mass Meda. New York: Cambrdge Unversty Press. Rbero, P. F. (2010). Dos sndcatos ao governo: a organzação naconal do PT de 1980 a São Paulo: EdUFSCar/FAPESP. Robnson, J. A. and T. Verder (2013). The poltcal economy of clentelsm. Scandnavan Journal of Economcs 115(2), Snyder, J. M. and D. Stromberg (2010). Press coverage and poltcal accountablty. Journal of Poltcal Economy 118(2), Stadnk, C. (1991). A hpótese do fenômeno do coronelsmo eletrônco e as lgações dos parlamentares federas e governadores com os meos de comuncação de Massa no Brasl. Master s thess, Faculdade dos Meos de Comuncação Socal, PUC-RS, Porto Alegre. Stromberg, D. (2004). Rado s mpact on publc spendng. The Quarterly Journal of Economcs 119(1), Valente, J. C. (2009). Produção regonal na tv aberta Braslera. Techncal report, Observatóro do Dreto á Comuncação. Vega, L. G. and M. M. Pnho (2007). The poltcal economy of ntergovernmental grants: Evdence from a maturng democracy. Publc Choce 133,

33 Appendx A: Proofs Proof of Lemma 1 We show that β s ncreasng n q. β q = θ(1 λ) (1 θλ)(1 q ) 2 > 0. Proof of Lemma 2 We frst show that f q < 1/ (1 + λ), β s ncreasng n n na. Operatng the expresson of β n equaton (2) yelds, β = 1/ ( 1 + q sna (1 λ) una (λ (1 q) + (1 λ)) una ) (1 β). β Usng that under a based federal government s na λqn na and u na (λ (1 q) + (1 λ)) n na, β = 1/ 1 + ( (1 λ) (λ(1 q)+(1 λ)) q λq (λ (1 q) + (1 λ)) (λ(1 q)+(1 λ)) ) n na (1 β). (7) β Denote c (1 λ) (λ(1 q)+(1 λ)) q λq (λ(1 q)+(1 λ)) (λ(1 q)+(1 λ)). It s clear that β s ncreasng n n na, as long as c < 1. Operatng such an expresson and usng the approxmaton that log (1 + x) x, t s straghtforward to see that such a condtons holds f q < 1/ (1 + λ). Snce ths must hold n equlbrum, we requre q < 1/ (1 + λ). Second, we show that β s decreasng n q. From equaton (7), t s clear that β s decreasng n q as long as c s ncreasng n q. Usng the approxmaton that log (1 + x) x, t s straghtforward to see that c q = cλ (1 + λ) (1 q ) > 0. Proof of Proposton 1 We frst show that f n na n na, q = 0. By the defnton of n na, f n na n, snce β (q = 0) s not a functon of n na and (from Lemma 2) β (q = 0) s ncreasng n n na, β β and thus τ (0) = 1. It s then optmal for a based federal government to set q = 0. Second we show that, f n na > n na, q s the unque q > 0 such that β = β. If n na > n na, under q = 0, we have that β > β and thus τ (q) = 0, whch yelds a zero payoff to the federal government. The government then has ncentves to ncrease q the mnmum possble requred to ensure that τ (q) = 1. Snce (from Lemma 1) β s ncreasng n q, and (from Lemma 2) β s decreasng n q, q s unquely defned by the q > 0 so that β = β. Proof of Proposton 2 To show that, f q < 1/ (1 + λ), q s ncreasng n n na, we splt the 33

34 proof n two. Frst, the larger ñ, the more lkely that n na > n na. Second, for n na > n na, from Proposton 1, q > 0 s such that β = β. Usng, the mplct functon theorem, where (from Lemma 1) β q q n na = β n na < 0; and f q < 1/ (1 + λ), ( / β q β ) q > 0, β n na > 0; and (from Lemma 2) β q > 0. Proof of Theorem 1 Denote na the number of non-algned muncpaltes covered by the network that operates n muncpalty. The frst and second part of the theorem follow trvally from the fact that n na that q s ncreasng n n na s ncreasng n na, snce n na δna, and the result from Proposton 2 f q < 1/(1 + λ). To show the thrd part of the theorem, note that the number of other non-algned affected muncpaltes covered by the rado network a non-algned muncpalty belongs to s n na. The number of these muncpaltes that receve ad from the federal government s q λn na. Fxng n na, ths number s ncreasng q, whch s the probablty that a non-algned affected muncpalty receves ad from the federal government. B: Meda Ownershp Data We tracked the poltcal connecton of each partner that s on the board of a meda outlet by matchng hs or her name to the names of poltcans. We use two dfferent matchng procedures dependng on the poltcans rank. For local poltcans, we dentfed the names of mayors and local counclors elected n 2000, 2004 and 2008, and consder that poltcans own a meda outlet f ther name s on the board of a meda outlet n the same muncpalty they were elected. We also consder that poltcans are assocated to a meda outlet f they share a famly name wth someone on the board of a meda outlet located n the muncpalty where they were elected. 37 For state and federal congressmen, we use the lst of elected poltcans n 1998, 2002 and 2006, and consder they own a meda outlet f ther name appears on the board of any meda outlet n 37 We were very conservatve n ths matchng and dd not consder that ndvduals are relatves f they share very common famly names such as Slva, Costa and Santos. In partcular, we dd not match names whose frequency s greater than 5% n the state. 34

35 2012, regardless of the locaton of the meda outlet and ther poltcal base. Ths assgnment of meda to congressmen follows the pattern we found n the data whch ndcates that many poltcans own meda n muncpaltes that are not ther poltcal strongholds or even n states that they do not represent. We dentfed that n 2% of the muncpaltes there s a commercal rado staton drectly owned by a congressmen. We acknowledge that our method may underestmate the number of meda outlets controlled by poltcans snce research has shown that poltcans ndrectly control meda by assgnng relatves or frends to the board of meda outlets (see Lma 2006 and Gorgen 2002). 38 We dentfy that 1% of Brazlan muncpaltes have a network-connected rado staton assocated wth a local poltcan or owned by a state and federal congressmen. Appendx C: Fgures and Tables 38 Unfortunately we do not have the muncpalty of orgn for each congressman, and thus we are unable to perform the same matchng procedure we use wth local poltcans. 35

36 36 (a) Ranfall statons (b) Emergency Declaraton Fgure 1: The fgure on the left shows all the ranfall statons from whch we used data on ranfall. The fgure on the rght ndcates the muncpaltes where there was an emergency declaraton at some pont of our perod of analyss.

37 37 (a) Never algned muncpaltes (b) Network-connected rado statons. Fgure 2: The fgure on the left shows the muncpaltes that were algned wth the federal government at some pont n tme durng our perod of analyss and those that were never algned. Out of the former muncpaltes, only 5 muncpaltes were algned durng our entre perod of analyss. The fgure on the rght ndcates the muncpaltes where there was a rado network operatng at some pont of our perod of analyss.