CDDRL WORKING PAPERS. Varieties of Clientelism: Machine Politics During Elections. Number 119 October 2010

Transcription

1 CDDRL WORKING PAPERS Number 119 October 2010 Varetes of Clentelsm: Machne Poltcs Durng Electons Jordan Gans-Morse Unversty of Calforna, Berkeley Sebastan Mazzuca Harvard Unversty Smeon Nchter Stanford Unversty Center on Democracy, Development, and The Rule of Law Freeman Spogl Insttute for Internatonal Studes Addtonal workng papers appear on CDDRL s webste:

2 Center on Democracy, Development, and The Rule of Law Freeman Spogl Insttute for Internatonal Studes Stanford Unversty Encna Hall Stanford, CA Phone: Fax: About the Center on Democracy, Development and the Rule of Law (CDDRL) CDDRL was founded by a generous grant from the Bll and Flora Hewlett Foundaton n October n 2002 as part of the Stanford Insttute for Internatonal Studes at Stanford Unversty. The Center supports analytc studes, polcy relevant research, tranng and outreach actvtes to assst developng countres n the desgn and mplementaton of polces to foster growth, democracy, and the rule of law.

3 Varetes of Clentelsm: Machne Poltcs Durng Electons Jordan Gans-Morse Sebastan Mazzuca Smeon Nchter October 29, 2010 The authors thank the followng people: Davd Coller, Ernesto Dal Bo, Scott Gehlbach, Peter Lorentzen, Al Montero, Robert Powell, Jm Robnson, Mchael Tomz, and Rodrgo Zarazaga. Jordan Gans-Morse and Smeon Nchter acknowledge support from the Natonal Scence Foundaton. Unversty of Calforna, Berkeley; Amercan Bar Foundaton. Harvard Unversty. Stanford Unversty, Center on Democracy, Development and the Rule of Law; Center for Global Development.

4 Abstract Clentelst partes (or poltcal machnes) engage n a varety of strateges durng electons. Most studes focus exclusvely on vote buyng, a strategy that rewards opposng voters for swtchng ther vote choces. Yet n many countres, machnes also adopt other strateges, such as actvatng ther passve consttuences through turnout buyng. What factors explan varaton n patterns of clentelsm durng electons? We develop an analytcal framework and formal model that emphasze the role of ndvdual and contextual factors. Poltcal machnes focus on two key attrbutes of ndvduals poltcal preferences and nclnaton to vote when choosng ther mx of clentelst strateges. Machnes also talor ther mx to at least fve contextual factors: compulsory votng, machne support, poltcal polarzaton, salence of poltcal preferences, and strength of ballot secrecy. Evdence from Argentna, Brazl, and Russa s consstent wth these fndngs.

5 Introducton Durng electons n many countres, clentelst partes (or poltcal machnes) offer selectve benefts to ctzens n exchange for poltcal support. Such partes compete not only on the bass of polcy platforms, but also wth materal nducements gven drectly to ndvduals. These nducements often nclude food, medcne, and other forms of sustenance. In contexts where ctzens are hghly dependent on such handouts, ncludng countres where the state fals to provde a socal safety net, ths pattern of machne poltcs can have partcularly mportant consequences for democratc accountablty and responsveness (Ktschelt & Wlknson 2007). In the past, promnent scholars vewed clentelsm as a pre-ndustral poltcal phenomenon that would wane as socetes modernzed (e.g., Scott 1969). But the evoluton of machne poltcs s often remarkably dfferent than n the U.S., where powerful machnes such as Tammany Hall n New York and the Dawson machne n Chcago lost consderable nfluence over tme. In many advanced democraces, such as Greece, Italy and Span, clentelst partes contnue to attract substantal numbers of votes usng drect materal nducements (Ktschelt & Wlknson 2007; Patton 2001). The nfluence of clentelsm s even more pronounced n many developng countres, where a growng body of evdence reveals the remarkable extent to whch partes engage n machne poltcs. In Brazl, the prevalence of nducements durng campagns motvated over one mllon ctzens to sgn a petton n 1999 for strcter legslaton, leadng to the recent prosecuton of over 660 poltcans (Movmento de Combate à Corrupção Eletoral 2010). Despte the major role of clentelsm n many contemporary socetes, we contnue to lack a thorough understandng of how poltcal machnes dstrbute benefts durng campagns. Most studes focus exclusvely on vote buyng, a strategy that rewards opposng voters for swtchng ther vote choces (e.g., Lehoucq 2007, 33; Stokes 2005, 315). Yet n many countres, machnes also adopt other strateges, such as actvatng ther passve consttuences through turnout buyng (Nchter 2008; Cox 2006). What factors explan varaton n patterns of 1

6 clentelsm durng electons? The present study provdes an analytcal framework and a formal model that hghlght how clentelsm s shaped by both ndvdual and contextual factors. Poltcal machnes focus on two key attrbutes of ndvduals (1) poltcal preferences and (2) nclnaton to vote when decdng how to dstrbute benefts durng electons. They fnd t relatvely expensve to nfluence the vote choces of strongly opposed ctzens, or to nduce turnout of ctzens who are strongly nclned to stay home on Electon Day. Machnes also talor ther mx of clentelst strateges to at least fve characterstcs of poltcal envronments: (1) compulsory votng, (2) machne support, (3) poltcal polarzaton, (4) salence of poltcal preferences, and (5) strength of ballot secrecy. Overall, our analyss suggests that clentelsm s far from a homogeneous poltcal phenomenon, and helps to explan varaton n patterns of machne poltcs. We examne evdence from Argentna, Brazl, and Russa, drawng on feldwork ntervews, local meda accounts and academc studes. Each country exhbts a dstnct mx of clentelst strateges: (1) predomnantly vote buyng n Brazl, (2) predomnantly turnout buyng n Russa, and (3) a relatvely balanced mx of vote buyng and turnout buyng n Argentna. Evdence suggests that factors hghlghted by our study help to explan ths varaton. Understandng varaton n clentelst strateges has mportant normatve mplcatons. Consder, for example, the dstncton between vote buyng and turnout buyng. Vote buyng may be seen as unambguously perncous for democracy, as the strategy nterferes wth free and far electons, and undermnes poltcal equalty by allowng those who have resources to buy the votes of the poor (Stokes 2005, 316; see also Schaffer & Schedler 2007). By contrast, Hasen (2000, , 1370) contends that the normatve mplcatons of turnout buyng are more ambguous because t may ncrease equalty of poltcal partcpaton by nducng the poor to vote. Such normatve questons challenge scholars to deepen ther understandng of how poltcal machnes dstrbute benefts durng campagns. The present study does not clam to provde an exhaustve analyss of all varetes of 2

7 clentelsm. We restrct our analyss to electoral clentelsm; that s, strateges that exclusvely nvolve the dstrbuton of benefts durng electoral campagns. We acknowledge that clentelsm often nvolves a broader set of strateges than just elte payoffs to ctzens before electons. For example, studes such as Scott (1969), Levtsky (2003), and Lawson (2009) dscuss other forms of clentelsm that nvolve ongong relatonshps of mutual support and dependence. Nevertheless, our explct focus on electoral clentelsm facltates analyss of numerous strateges that reman poorly understood. The fndngs of ths study also contrbute to the broader lterature on dstrbutve poltcs. Vgorous scholarly debate contnues over how partes dstrbute targetable goods, such as nfrastructure projects and partcularstc benefts. Two semnal formal studes offer conflctng predctons: whereas Cox & McCubbns (1986) argue that partes wll dstrbute targetable goods to core supporters, Lndbeck & Webull (1987) contend they wll target swng voters. A more recent conceptual paper by Gary Cox (2006) argues that these and other studes focus too narrowly on persuason (changng voters preferences); when strateges such as moblzaton (affectng whether ctzens vote) are consdered, the core-supporter hypothess s substantally strengthened. The present study contrbutes to ths lterature by developng an analytcal framework and a formal model that nvestgate how clentelst partes combne strateges of persuason and moblzaton. The present study also advances formal studes of clentelsm. Prevous models rely on a one-dmensonal voter space, n whch ctzens are arrayed along a spectrum of poltcal preferences as n the classc Downsan spatal model of poltcal competton. We ntroduce a second dmenson, such that ctzen types are defned both by poltcal preferences and votng costs. Ths nnovaton facltates the ntegraton of nonvoters nto our analyses. As a result, the present study addresses a major lmtaton n almost all exstng models of clentelsm they examne only one strategy. For example, Stokes (2005) provdes a model of vote buyng, and Nchter (2008) develops a model of turnout buyng. By contrast, we analyze the tradeoffs that partes face when combnng strateges. One recent study by Morgan & Vardy 3

8 (2009) also begns to tackle the key ssue of how partes combne strateges, but focuses narrowly on the mpact of ntroducng the secret ballot. The present paper offers a more exhaustve analyss of the range of strateges employed by poltcal machnes and, through the model s comparatve statcs, a fuller assessment of the factors that nfluence varaton of clentelst strateges. 1 Analytcal Framework In order to develop an analytcal framework of how poltcal machnes dstrbute benefts durng campagns, we frst buld on a conceptual typology ntroduced by Nchter (2008). Fgure 1 presents fve clentelst strateges, emphaszng the mportance of two key attrbutes of ndvduals: (1) poltcal preferences and (2) nclnaton to vote. Each strategy targets dfferent types of ndvduals and nduces dstnct actons. [Insert Fgure 1 here] The vast majorty of studes focus on vote buyng, a strategy by whch partes reward opposng (or ndfferent) voters for swtchng ther vote choces. Vote buyng s consdered to be prevalent n many countres; for example, over 70 percent of Ngerans n a recent survey beleved that vote buyng occurs all of the tme or most of the tme durng electons, wth nearly 40 percent reportng that a close frend or relatve was offered benefts n exchange for votng for a partcular canddate n the 2003 presdental electon. 2 Recent publcatons on vote buyng focus on countres ncludng Argentna (Stokes 2005), Benn and São Tome (Vcente & Wantchekon 2009), Japan (Nyblade & Reed 2008), Mexco (Daz-Cayeros, Estevez & Magalon forthcomng), and Thaland (Bowe 2008). Such studes typcally assume ether mplctly or explctly that poltcal machnes dstrbute benefts to voters n exchange for votng aganst ther preferences. 1 Dunnng & Stokes (2009), an unpublshed paper on the topc, examnes only two strateges. 2 Survey of 2,410 Ngerans n all 36 states and the federal captal terrtory n February See Ngerans: Vote Buyng a Common Occurrence, Internatonal Foundaton for Electoral Systems,

9 Although scholars frequently assume that machnes reward ctzens for vote-swtchng, n realty machnes adopt a broader set of strateges. For example, partes often engage n turnout buyng, dstrbutng rewards to unmoblzed supporters n exchange for showng up at the polls. Durng the 2004 US electon, fve Democratc Party operatves n East St. Lous were convcted n federal court for offerng cgarettes, beer, medcne, and $5 to $10 rewards to ncrease turnout of the poor (Nchter 2008). One party offcal pleaded gulty and testfed that operatves offered ndvduals rewards because f you ddn t gve them anythng, then they wouldn t come out (cf Nchter 2008, 19). In the case of Argentna, Nchter argues that although both strateges coexst, survey data n Stokes (2005) are more consstent wth turnout buyng than vote buyng. Evdence of turnout buyng has also been found n the case of Venezuela (Rosas & Hawkns 2008), as well as Argentna and Mexco (Dunnng & Stokes 2009). Another strategy nvolvng moblzaton s double persuason, whch targets ndfferent or opposng nonvoters. The broader lterature on clentelsm suggests that many ndvduals have lttle n the way of deologcal preferences or reasons to vote, other than materal rewards offered by clentelst partes (e.g., Chubb 1982, 171). Wth double persuason, machnes dstrbute benefts to such ctzens n order to nduce ther electoral partcpaton and nfluence ther vote choces. Double persuason targets nonvoters, but s dstnct from turnout buyng because recpents do not nherently prefer the machne on deologcal or programmatc grounds. Although studes typcally gnore double persuason, we fnd that machnes optmally devote some resources to ths strategy whenever they dstrbute selectve benefts durng campagns. Partes may also engage n negatve turnout buyng, whch rewards ndfferent or opposng ndvduals for not votng (Cox & Kousser 1981; Morgan & Vardy 2009). 3 Hstorcally, Cox & Kousser (1981) fnd that negatve turnout buyng ncreased substantally after the ntroducton of the secret ballot n the Unted States. Whle evdence shows that negatve 3 Ths strategy s often termed negatve vote buyng, but the term negatve turnout buyng s more precse as the strategy nfluences turnout, not vote choces. 5

10 turnout buyng exsts n some developng countres, ths demoblzatonal strategy s not possble n many contemporary poltcal envronments. As Schaffer (2007, 188) argues: For a varety of reasons... buyng abstenton s not a wdespread form of vote buyng around the world. Negatve turnout buyng may be consdered normatvely less acceptable, and n contexts such as Brazl nvolves hgher penaltes for convcted poltcans. Gven that machnes engage n negatve turnout buyng only n relatvely few poltcal envronments, we explore ths strategy as an extenson of our base model. Another potental strategy s rewardng loyalsts, n whch clentelst partes offer rewards to supporters who would vote for them anyway. By defnton, such rewards do not nfluence vote choces or nduce turnout durng a contemporaneous electon. Scholars typcally understand such benefts as part of ongong, long-term relatonshps between poltcans and ctzens (e.g., Auyero 2000; Ktschelt & Wlknson 2007). In one explanaton of rewardng loyalsts, Daz-Cayeros, Estevez & Magalon (forthcomng, ch. 4) argue that partes offer selectve benefts to core supporters durng electons n order to prevent the eroson of partsan loyaltes over tme. Gven that we focus on short-term electoral clentelsm, such ongong relatonshps are outsde of the scope of our analyss, and we do not ncorporate rewardng loyalsts n the present paper. Combnng Strateges When dstrbutng benefts durng campagns, poltcal machnes frequently combne several of the strateges n Fgure 1. To provde ntuton and motvate formal analyss of how poltcal machnes combne strateges, we frst present a stylzed example. Assume that a poltcal machne has $75 to dstrbute to ctzens durng a campagn. The machne seeks to maxmze ts electoral prospects by nfluencng vote choces and/or nducng turnout. There are nne ctzens whom the machne can target: Opposng Voters: Veronca ($10), Vctor ($30), Vrgna ($50) Supportng Nonvoters: Tomas ($10), Teresa ($15), Tona ($20) Opposng Nonvoters: Debora ($10), Davd ($20), Dego ($30) 6

11 We focus here on vote buyng (targets opposng voters), turnout buyng (targets supportng nonvoters), and double persuason (targets opposng nonvoters). Observe that dfferent payments (n parentheses) are requred to buy each ctzen usng the relevant strategy. The requred payments vary because ctzens dffer wth respect to two key attrbutes: (1) poltcal preferences and (2) nclnaton to vote. For example, vote buyng s more expensve when a ctzen strongly opposes the machne on deologcal grounds. Lkewse, turnout buyng and double persuason are more costly f the ctzen s strongly nclned not to vote. Gven the dfferent requred payments, how does the machne allocate ts budget? The frst crucal consderaton s that vote buyng benefts the machne more than other strateges. Vote buyng provdes two net votes t adds a vote to the machne s tally, and subtracts one from the opposton. By contrast, turnout buyng and double persuason provde only one net vote because they target nonvoters. To allocate ts budget effcently, the machne should target ctzens who offer the most net votes per dollar spent. Usng ths metrc, the machne should start by vote buyng Veronca. For $10, t earns two net votes (.e., $5 per net vote). To vote buy an addtonal ctzen, the machne would need to pay Vctor $30 ($15 per net vote). Thus, the machne would be better off turnout buyng Tomas and double persuadng Debora, as each provdes one net vote for $10. The machne now has $45 remanng, and consders costler ctzens. It should vote buy Vctor for $30 and turnout buy Teresa for $15. Both optons are equally cost-effectve ($15 per net vote), and preferable to the alternatve of double persuadng Davd ($20 per net vote). Ths stylzed example provdes several nsghts for further nvestgaton: (1) machnes optmally combne clentelst strateges; (2) ther mx ncludes turnout buyng and double persuason; (3) ther mx depends on ctzens poltcal preferences and nclnaton to vote; and (4) machnes are wllng to pay more for vote buyng relatve to other strateges. We now develop a model that confrms the ntuton gleaned above, and also suggests how machnes talor ther mx of clentelst strateges to specfc poltcal envronments. 7

12 Model Setup Consder two poltcal partes, an ncumbent machne party (M) and an opposton party (O). Each party offers a platform, x M and x O, respectvely, on a one-dmensonal deologcal spectrum rangng from X to X. Wthout loss of generalty, let x O < x M, and for smplcty, assume that the partes platforms are symmetrc around zero (that s, x O = x M ). 4 Both partes platforms are fxed for the duraton of our analyss. Ths smplfyng assumpton s consstent wth our focus on electoral clentelsm, and accurately reflects realty durng many electoral campagns: partes may have attrbutes that cannot be credbly transformed n the short run, such as the personal or deologcal characterstcs of ther leaders. Each ctzen s defned by her poltcal preferences x and votng costs c, where x and c are ndependent. The ctzens deal ponts x are dstrbuted over [X, X] accordng to F (x), where F has a strctly postve and contnuously dfferentable densty f over (X, X). Costs of votng c are dstrbuted over [0, C] accordng to G(c), where G has a strctly postve and contnuously dfferentable densty g over (0, C). For ease of explcaton, we focus on the case where the partes platforms are the endponts of the ctzens deologcal spectrum (.e., X = x O and X = x M ), but results are not affected f some ctzens have more extreme poltcal preferences (.e., X < x O and X > x M ). 5 A ctzen s utlty equals the dfference between her expressve value from votng and her votng costs. 6 Formally, a ctzen of type (x, c ) who votes for party P {M, O} receves utlty: U P (x, c ) = x P x c (1) 4 Ths assumpton smplfes the algebra but qualtatvely does not affect our results. 5 A proof s avalable upon request. 6 Morgan & Vardy (2009) offer a formal justfcaton for the assumpton that voters receve only expressve utlty, not nstrumental utlty (.e., utlty derved from affectng the outcome of the electon). Gven reasonable assumptons, a ctzen s probablty of beng pvotal converges to zero as the electorate sze ncreases. 8

13 The frst term, x P x, captures the noton that the closer the ctzen s deal pont to the platform of the party for whch she votes, the more utlty she receves from castng a ballot. The second term, c, represents votng costs, such as transportaton, lost wages, or chld care needed to reach the polls. A ctzen who chooses not to vote receves no expressve utlty from votng and also ncurs no drect votng costs. However, n most socetes ctzens who fal to vote face abstenton costs. Such abstenton costs range from socal dsapprobaton to fnes and penaltes n countres wth compulsory votng laws. We thus assume that a non-voter ncurs a cost a > 0. 7 The objectve of the machne s to maxmze ts net votes the number of votes t receves mnus the number of votes the opposton party receves. Snce the machne cannot adjust ts platform durng the campagn, ts task s to acqure addtonal votes by dstrbutng selectve benefts. We assume the machne cannot afford to buy all ctzens, because t has lmted resources gven by a budget B. Thus, the machne must decde how to allocate ts budget optmally across dfferent types of ctzens. 8 We assume that the machne observes ctzens poltcal preferences and votng costs. To llustrate the basc logc of our model, we ntally gnore the rsk of opportunstc defecton by ctzens (e.g., a ctzen receves a vote-buyng payment and stll votes aganst the machne). An extenson of the model then consders opportunstc defecton. In addton, gven that n many contexts partes cannot pay ctzens to stay home on Electon Day (e.g., Schaffer 2007: 188), we ntally assume that machnes cannot engage n negatve turnout buyng. We later relax ths assumpton to analyze how the machne s optmal allocaton of resources changes when negatve turnout buyng s allowed. Fnally, the model assumes that only the machne, and not the opposton party, has the 7 We make two realstc assumptons that ensure an nteror soluton to the machne s optmzaton problem and monotoncty of comparatve statcs: (1) some ndfferent ctzens vote (formally, ths requres a > x M ); and (2) even wth electoral clentelsm, there exst strong supporters who do not vote (formally, ths requres C a > b, where b s defned below as the most-expensve payment to nonvoters). 8 Formally, the machne s problem s to maxmze ts net votes by assgnng a reward b 0 to every ctzen, such that total expendtures, N b g(c)f(x)dc dx, are less than or equal to budget B, where N s the total number of ctzens. 9

14 capacty to offer rewards to ctzens. Ths assumpton, whch follows models of clentelsm such as Stokes (2005) and Nchter (2008), reflects the realty n many contexts where only one party has the nfrastructure, access to state resources, and socal networks necessary to engage n clentelsm. Stokes (2009, 12, 20) offers two explanatons for what she calls the sngle-machne assumpton: (1) the ncumbent party has exclusve access to publc coffers, from whch clentelst payments are made; and (2) only one party has nvested n the dense organzatonal structure and socal proxmty that defne a machne. In many contexts where only the domnant machne has party operatves embedded n neghborhoods, other partes fnd t exceedngly dffcult to collect nformaton about ctzens preferences and votng costs, as well as to enforce clentelst exchanges. Classfyng Ctzens Gven ts knowledge of preferences and votng costs, the machne can classfy ctzens. If a ctzen shows up at the polls, she wll vote for the machne f dong so provdes (weakly) greater utlty than votng for the opposton. That s, a ctzen votes for the machne f U M U O, or equvalently, f x 0. 9 Thus, ctzens wth poltcal preferences x 0 are supporters of the machne, whle those wth poltcal preferences x < 0 are opposers. If a ctzen chooses not to vote, she receves no expressve utlty from votng and faces no votng costs. However, she ncurs abstenton costs. Hence, a ctzen wll choose to vote f she receves (weakly) greater utlty from votng than from abstanng. That s, she votes f max [U M, U O ] a, or equvalently, f max [ x M x c, x O x c ] a. Overall, the machne can classfy the populaton nto four groups of ctzens: Supportng Voters: Ctzens wth x 0 and x M x c a Supportng Nonvoters: Ctzens wth x 0 and x M x c < a Opposng Voters: Ctzens wth x < 0 and x O x c a Opposng Nonvoters: Ctzens wth x < 0 and x O x c < a 9 To ensure that the party s optmzaton problem s well-defned, we assume that ctzens who are ndfferent between the two partes vote for the machne and that ctzens who are ndfferent between abstanng and votng come to the polls. 10

15 Fgure 2a presents a graphcal depcton of these four groups of ctzens (from the perspectve of the machne). Poltcal preferences are represented on the horzontal axs, whle votng costs are represented on the vertcal axs. The vertex lnes represent ctzens who are ndfferent between votng and not votng, because they receve the same utlty from votng as they do from abstanng. 10 All ctzen types on or below lne l 1 vote for the machne; those on or below lne l 2 vote for the opposton. All ctzen types above l 1 and l 2 are nonvoters. [Insert Fgure 2 here] The vertex shape of the cutoff lne between voters and nonvoters reflects the fact that ctzens wth ntense poltcal preferences (.e., voters for whom x approaches ether x M or x O ) receve greater expressve utlty from votng, as can be observed n the utlty functon (Equaton 1). They are thus more nclned to ncur votng costs and turn out to support ther favored party. By contrast, ctzens who have weak poltcal preferences (.e., ctzens for whom x approaches 0) receve lower expressve utlty from votng, and thus are less nclned to ncur votng costs. Addtonally, the vertex ntercepts the vertcal axs above the orgn, whch reflects the fact that some ndfferent ctzens vote. Payments In order to determne the machne s optmal mx of clentelst strateges, we frst dentfy how much the machne would need to pay to buy each ctzen type. For each strategy, the requred payments (b ) are as follows: Vote Buyng: Vote buyng targets opposng voters, who have a reservaton utlty of U O. To nduce an opposng voter of type t = (x, c ) to swtch her vote, the machne must therefore pay b V B such that U M + b V B U O. In an optmal allocaton, the machne sets payments equal to a ctzen s reservaton value, because t wll not overpay (pay a ctzen more than her reservaton value) or underpay (pay a ctzen less than her reservaton value). 11 Thus, 10 Formally, these are supporters for whom x M x c = a and opponents for whom x O x c = a. It thus follows that l 1 = x x M + a and l 2 = x + x O + a = x x M + a, where the second equaton follows from the assumpton of symmetrc party platforms, x M = x O. 11 A proof s avalable upon request. 11

16 the nequalty bnds. Substtutng the denttes of U M and U O from Equaton 1 yelds: x M x c + b V B = x O x c. Then, solvng for b V B : 12 b V B = 2x (2) Wth vote buyng, the machne must compensate a ctzen for castng a vote aganst her poltcal preferences. As shown n Equaton 2, the machne can vote buy all opposng voters wth a gven deal pont for the same prce, even f they have dfferent costs of votng. Because they already show up at the polls, opposng voters only need to be compensated for votng aganst ther poltcal preferences. Turnout Buyng: Turnout buyng targets supportng nonvoters, who have a reservaton utlty of a (where a s the cost of abstenton). To nduce turnout of a supportng nonvoter of type t = (x, c ), the machne must pay b T B such that U M + b T B = a. Substtutng the dentty of U M from Equaton 1 yelds: x M x c + b T B = a. Then, solvng for b T B : b T B = c x + x M a (3) Supportng nonvoters receve more utlty from abstanng than from votng. Thus, wth turnout buyng, the machne must compensate such ctzens for the dfference between the utlty receved from stayng home and the utlty receved from votng for the machne. Double Persuason: Double persuason targets opposng nonvoters, who nether partcpate n electons nor support the machne. Ther reservaton utlty s a. To nduce an opposng nonvoter of type t = (x, c ) to turn out and vote for the machne, the party must therefore pay b DP such that U M +b DP = a. Substtutng the dentty of U M from Equaton 1 yelds: x M x c + b DP = a. Then, solvng for b DP : b DP = c x + x M a (4) Observe Equatons 3 and 4 are dentcal, except that double persuason targets opposng nonvoters (x < 0), whle turnout buyng targets supportng nonvoters (x 0). Wth double persuason, the machne must compensate opposng nonvoters for: (1) votng aganst ther poltcal preferences; and (2) ther dsutlty from votng relatve to abstanng. 12 Recall that by the assumpton of symmetrc party platforms, x M = x O. 12

17 Optmal Mx of Clentelst Strateges Gven ths nformaton about requred payments, we now determne the optmal mx of clentelst strateges. Ths secton provdes ntuton about how a machne optmally allocates resources across vote buyng, turnout buyng, and double persuason n order to maxmze ts electoral prospects. The appendx provdes proofs of each proposton. The machne condtons the sze of rewards on ctzens deal ponts and votng costs (n accordance wth Equatons 2 4), and targets those ctzens who delver net votes most cheaply. Otherwse, the machne would be better off shftng resources to obtan addtonal electoral support. Observe that the machne s wllng to pay twce as much to the most-expensve vote-buyng recpent (a payment of b V B ) as t s wllng to pay to the most-expensve turnoutbuyng and double-persuason recpents (payments of b T B and b DP, respectvely). After all, vote buyng delvers twce as many net votes as the other two strateges. By the same logc, the machne s wllng to pay the most expensve turnout-buyng recpent exactly as much as t pays the most expensve double-persuason recpent, because they both yeld one net vote. In sum, as shown formally n the appendx: Proposton 1: In an optmal allocaton of resources, the machne sets b V B = 2b T B = 2b DP. For notatonal smplcty, analyss below drops the subscrpts, lettng b V B = b and b T B = b DP = b. An mportant fndng follows mmedately. Observe n Proposton 1 that f b V B, b T B, or b DP s greater than 0, then all three terms must be greater than 0. Therefore: Proposton 2: If a machne engages n electoral clentelsm, then optmally t allocates resources across all three strateges of vote buyng, turnout buyng, and double persuason. Proposton 2 reveals that moblzaton s fundamental to understandng the logc of how machnes dstrbute selectve benefts durng electons. Whereas most studes focus exclusvely on vote buyng, the model suggests that machnes never optmally expend all ther resources on one strategy. Beyond vote buyng, machnes should also seek nonvoters who can be nduced to delver votes n exchange for small rewards. Another mportant mplcaton pertans to double persuason. Ths strategy mght not 13

18 seem ntutve why dstrbute benefts to ctzens who nether vote nor support the machne? Indeed, Dunnng & Stokes (2009) even call double persuason a perverse strategy. Yet our model suggests that t s always optmal for machnes to engage n double persuason. When operatves dstrbute rewards, they fnd that targetng weakly opposng nonvoters through double persuason s often more cost-effectve than buyng votes of strongly opposed voters, or buyng turnout of supportng nonvoters wth hgh votng costs. Gven that the machne optmally combnes all three strateges, how does t determne whch ctzens to buy? Fgure 2b provdes ntuton about whom the machne optmally buys, buldng on the vertex shown n Fgure 2a. Frst, consder who wll receve the most-expensve vote-buyng payment (b ). Gven that the machne nether overpays nor underpays, t delvers b to opposng voters who requre exactly that level of benefts to swtch ther vote choces. In accordance wth Equaton 2, these are opposng voters of type t j = (x j, c j ) for whom b = 2x j. Such voters are located on lne segment l 3 n Fgure 2b. 13 For turnout buyng, the machne delvers the most-expensve payment (b ) to supportng nonvoters who requre exactly that level of benefts to come to the polls. In accordance wth Equaton 3, these are supportng nonvoters of type t k = (x k, c k ) for whom b = c k x k + x M a. Such supportng nonvoters are located on lne segment l 4, to the rght of the vertcal axs. 14 Observe that l 4 s parallel to l 1, and the vertcal dstance between the two lne segments s b. In other words, all voters along l 4 receve the same payment, because the hgher votng costs of some ctzens on ths lne segment are balanced by ther stronger preferences for the machne s platform. For double persuason, the machne delvers the most-expensve double-persuason payment (b ) to opposng nonvoters who requre exactly that level of benefts to turn out and vote for the machne. In accordance wth Equaton 4, these are opposng nonvoters of type t l = (x l, c l ) for whom b = c l x l + x M a. Such opposng nonvoters are located on lne 13 The lne segment l 3 s gven by the equaton x = b 2, on the range from the horzontal axs to the pont where l 3 ntersects wth l The lne segment l 4 s gven by the equaton c = x x M + a + b, on the doman from the vertcal axs to X M. 14

19 segment l 5, to the left of the vertcal axs. 15 Observe that l 4 and l 5 ntercept the vertcal axs at the same pont, because the most-expensve payments for double persuason and turnout buyng are the same. Thus far, graphcal analyss suggests whom the machne buys wth ts most-expensve payments (b and b ): ctzens on l 3 receve vote-buyng payments of b, ctzens on l 4 receve turnout-buyng payments of b, and ctzens on l 5 receve double-persuason payments of b. Another key nsght s that the machne optmally buys all ctzens whose requred payments are less than or equal to the most-expensve payments for each respectve strategy. That s, the machne buys all ctzens n the shaded areas n Fgure 2b. For further ntuton, assume that a voter X weakly opposes the machne and requres a vote-buyng payment b, whch s smaller than b. If the machne vote buys an opposng voter Y for b, then t must also vote buy X, because she provdes the same number of net votes for a smaller payment. Otherwse, the machne would be better off buyng X nstead of Y, and reallocatng the savngs. Note that the machne optmally pays X exactly her requred payment, as t does not overpay n equlbrum. Such logc also apples for turnout buyng and double persuason. The model also provdes nsght about whom the machne does not buy. In an optmal allocaton of resources, the machne dstrbutes no benefts to opposng voters who requre payments greater than b, or to nonvoters who requre payments greater than b. That s, the machne buys no ctzens outsde the shaded areas n Fgure 2b. For further ntuton, assume that a voter Z strongly opposes the machne and requres a vote-buyng payment b, whch s greater than b. Observe that even the most-expensve vote-buyng payment b underpays Z and s not enough to persuade her to swtch her vote. Thus, t cannot be optmal for the machne to expend resources on ctzens requrng vote-buyng payments larger than b. The logc s analogous for turnout buyng and double persuason. Taken together, these fndngs suggest the optmal mx of clentelst strateges: 15 The lne segment l 5 s gven by the equaton c = x x M + a + b, on the doman from the pont where l 5 ntersects wth l 2 to the vertcal axs. 15

20 Proposton 3 Vote Buyng: If b V B Turnout Buyng: If b T B Double Persuason: If b DP b, the machne pays an opposng voter b V B b, the machne pays a supportng nonvoter b T B b, the machne pays an opposng nonvoter b DP No Payment: The machne makes no payment to all other ctzens The appendx provdes a formal dervaton of these equlbrum condtons, and shows how the machne determnes b and b. In order to explore why ths optmal mx dffers across electoral contexts, we now examne comparatve statcs. Comparatve Statcs Formal analyss reveals how contextual factors shape patterns of clentelsm durng electons. Machnes optmally talor ther mx of clentelst strateges to at least fve characterstcs of poltcal envronments: (1) compulsory votng, (2) machne support, (3) poltcal polarzaton, (4) salence of poltcal preferences, and (5) strength of ballot secrecy. Ths secton provdes ntuton about how each factor nfluences the optmal mx, based on analytcal solutons derved n the appendx. More specfcally, the formal analyss ndcates how machnes optmally change the quantty of ctzens bought wth each strategy n response to parameter shfts n the model. In response to such changes, machnes alter whch ctzens they buy by reallocatng resources across and wthn strateges of electoral clentelsm. Changes n the poltcal envronment affect the number of cheap targets that the machne can buy wth each strategy. Thus, machnes reallocate resources towards strateges that now offer addtonal cheap targets. In addton, machnes reallocate resources wthn a gven strategy to ensure that they contnue to buy the cheapest ctzens. For tractablty, comparatve statcs examne the case where x and c are dstrbuted unformly. [Insert Fgure 3 here] The ntroducton of compulsory votng ncreases vote buyng ( V B a buyng ( T B a < 0), and decreases double persuason ( DP a 16 > 0), decreases turnout < 0). Wthn the model, the pa-

21 rameter through whch compulsory votng affects electoral clentelsm s ncreased abstenton costs (a). Hgher abstenton costs boost turnout and shft the vertex upwards (compare Fgure 3a to Fgure 2b). Ths upward shft ncreases the number of cheap vote-buyng targets, who are weak opposng voters clustered along the vertcal axs under the vertex. In order to buy these newly ntroduced cheap targets for vote buyng, the machne: (1) reallocates resources from turnout buyng and double persuason towards vote buyng, and (2) reallocates resources wthn vote buyng from the most-expensve recpents towards the newly ntroduced cheap targets. An mportant substantve mplcaton s that efforts to boost electoral partcpaton through compulsory votng may well lead to the unntended consequence of ncreased vote buyng, whch nduces ctzens to vote aganst ther partsan preferences. An ncrease n machne support decreases vote buyng ( V B x buyng ( T B x DP > 0), and decreases double persuason ( x < 0), ncreases turnout < 0). We conceptualze machne support as the proporton of ctzens who prefer the machne s platform over the opposton party s platform. To analyze ths comparatve statc, we unpack ctzens poltcal preferences such that x = x + ɛ, where x represents the poltcal preferences of the medan voter, and ɛ captures ndvdual-specfc devaton from the medan voter. 16 A rse n support for the machne s platform ncreases x and shfts the vertex left (see Fgure 3b). Ths leftward shft ncreases the number of cheap turnout-buyng targets, who are supportng nonvoters clustered just above l 1. In order to buy these newly ntroduced cheap targets for turnout buyng, the machne: (1) reallocates resources from vote buyng and double persuason towards turnout buyng, and (2) reallocates resources wthn turnout buyng from the most-expensve recpents towards the newly ntroduced cheap targets. Substantvely, ths comparatve statc suggests that a machne operatng n several poltcal dstrcts wll optmally talor ts clentelst mx accordng to poltcal support. When dstrbutng benefts n dstrcts wth many loyalsts, the machne employs relatvely more turnout buyng. But 16 The utlty functon for machne supporters (Equaton 1) thus becomes: U M = x M (x + ɛ ) c. Observe that Equaton 1 s a specal case of ths setup, n whch x = 0 (.e., n the orgnal setup, the machne party and opposton party have equal levels of poltcal support). 17

22 n opposton balwcks, t employs relatvely more vote buyng. An ncrease n poltcal polarzaton decreases vote buyng ( turnout buyng ( T B (x M x O ) > 0), and ncreases double persuason ( DP V B (x M x O ) (x M x O ) < 0), ncreases > 0). We conceptualze poltcal polarzaton as the deologcal dstance between partes (formally, x M x O ). Observe that as polarzaton ncreases, voters wth moderate deologcal preferences receve less expressve utlty from votng, because the deologcal dstance from ther preferred party grows. As a result, some voters no longer come to the polls, and the vertex shfts down (see Fgure 3c). Ths downward shft decreases the number of cheap vote-buyng targets, who are weak opposng voters clustered along the vertcal axs under the vertex. As the number of cheap vote-buyng targets decreases, the machne: (1) reallocates resources from vote buyng to turnout buyng and double persuason, and (2) reallocates resources wthn vote buyng from the lost cheap targets towards costler opposng voters. Overall, the model suggests that machnes rely relatvely more on moblzatonal strateges where poltcal polarzaton s hgh, and rely relatvely more on vote buyng where poltcal polarzaton s low. An ncrease n the salence of poltcal preferences decreases vote buyng ( V B κ < 0), ncreases turnout buyng ( T B κ DP > 0), and ncreases double persuason ( κ > 0). To analyze ths factor, we ntroduce a parameter κ > 0 to the utlty functon of ctzens (Equaton 1): U M = κ x M x c. 17 The parameter κ represents the mportance of expressng one s poltcal preferences, relatve to the cost of votng. As the salence of poltcal preferences rses (.e., κ ncreases), the vertex becomes steeper and shfts down (see Fgure 3d). Ths downward shft reduces the number of cheap vote-buyng targets, who are weak opposng voters clustered along the vertcal axs under the vertex. Gven that the number of cheap vote-buyng targets declnes, the machne: (1) reallocates resources from vote buyng to turnout buyng and double persuason, and (2) reallocates resources wthn vote buyng from the lost cheap targets towards costler opposng voters. Overall, when poltcal preferences are more salent, t s relatvely more expensve to nduce ctzens to vote aganst ther 17 Observe that Equaton 1 s a specal case of ths setup, n whch κ = 1. 18

23 preferences, and thus machnes shft resources away from vote buyng. Ballot Secrecy We also examne how machnes talor ther mx of clentelst strateges to a ffth contextual factor the strength of ballot secrecy. Ballot secrecy affects the rsk of opportunstc defecton when machnes reward ctzens for votng aganst ther poltcal preferences. To nvestgate the effects of ballot secrecy on electoral clentelsm, we extend the base model by relaxng the assumpton that transactons are fully enforceable. Machnes employ a varety of tactcs to compromse the secret ballot. For example, partes n the Phlppnes gve out carbon paper so voters can copy ther ballots, and Italan partes lend moble phones wth cameras so reward recpents can photograph how they vote (Schaffer & Schedler 2007, 30 31). The strength of ballot secrecy affects the machne s ablty to montor vote-buyng and double-persuason agreements. Both of these strateges requre some ablty to volate ballot secrecy, n order to ensure that reward recpents comply by votng aganst ther preferences. By contrast, the strength of ballot secrecy does not affect the machne s ablty to montor turnout-buyng agreements. As Nchter (2008) emphaszes, turnout buyng nvolves only montorng whether not for whom a supporter votes. To capture the effect of ballot secrecy, we buld on Stokes (2005) and Nchter (2008), adoptng a smpler setup that provdes the same analytcal leverage. If a ctzen does not comply wth an agreement to vote aganst her preferences, wth probablty p [0, 1] the machne montors her vote choce and rescnds the reward. Stronger ballot secrecy reduces p. Employng ths setup, we now determne the payments requred for vote buyng and double persuason when contracts are not fully enforceable. To prevent defecton when engagng n vote buyng, the followng ncentve-compatblty condton must hold: U M + b V B (x, c ) pu O + (1 p)[u O + b V B (x, c )] bv B (x, c ) 1 p [ 2x ] = 1 p bv B (5) In words, vote buyng wll be effectve when the payoff from complyng wth the agreement s 19

24 greater than or equal to the expected value of defectng and votng aganst the machne. As above, the machne wll not overpay or underpay, so the nequalty bnds. Observe that b V B s the requred vote-buyng payment when contracts are fully enforceable (the base model), whereas b V B s the requred vote-buyng payment when contracts are not fully enforceable. When opportunstc defecton s possble, the machne pays a premum on every dollar spent on vote buyng (that s, 1 p 1). Analogous logc s employed to determne the requred payment for double persuason when contracts are not fully enforceable: b DP = 1 p bdp. These payment equatons ndcate the party s optmal allocaton strategy under the condton of ballot secrecy. The logc underlyng Propostons 1-3 contnues to apply, and the form of the optmal allocaton s unchanged except that the rato of most-expensve payments reflects the premum payments for vote buyng and double persuason: 1 p b V B = 2 p b DP = 2b T B.18 To examne how ballot secrecy affects the optmal mx, we analyze comparatve statcs. Enhanced ballot secrecy decreases vote buyng ( V B p ( T B p < 0), and decreases double persuason ( T B p > 0), ncreases turnout buyng > 0). As ballot secrecy ncreases, the machne s probablty p of catchng ctzens who defect on vote-buyng or double-persuason agreements declnes, so t must pay a larger premum to ensure complance when usng these two strateges. Thus, the party optmally shfts resources away from vote buyng and double persuason, and reles relatvely more heavly on turnout buyng. Hstorcal studes suggest that the ntroducton of the secret ballot reduced vote buyng, as predcted by our model. Proponents of the secret ballot argued that t would reduce montorng of vote choces and thereby decrease vote buyng, whch was relatvely common wth open votng (Campbell 2005, 97; Lehoucq 2002, 6). For example, a US newspaper commented n 1888 that f the act of votng were performed n secret, no brbed voter could or would be trusted to carry out hs bargan when left to hmself (cf Campbell 2005, 97). The broad consensus s that vote buyng dd n fact decrease wth secret ballot (e.g., Cox 2006, 5; Hasen 2000: 1328). As Hasen (2000, 1328) explans, wth the rse of the secret 18 The proof s dentcal to the proof for Proposton 1, substtutng b V B and b DP for b V B and b DP. 20

25 ballot and the concomtant ncrease n the cost of verfyng that vote buyers were gettng what they pad for, vote buyng almost certanly has declned. Of course, machnes stll buy votes n many contexts, usng varous (albet more costly) methods to montor vote choces. Negatve Turnout Buyng In some contexts, machnes also engage n negatve turnout buyng, whch rewards opposng voters for not comng to the polls. To examne how the ncluson of negatve turnout buyng affects the optmal mx of clentelst strateges, we adopt the same approach as above. Frst, the machne determnes requred payments for negatve turnout buyng (b NT B ). The strategy targets opposng voters, who have a reservaton utlty of U O. If such ctzens do not vote, they receve payoff a of a nonvoter. In order to convnce an opposng voter of type t = (x, c ) to stay home, the machne must offer a reward b NT B such that: b NT B a U O. The machne optmally nether overpays nor underpays, so ths nequalty bnds. Substtutng U O from Equaton 1 yelds: b NT B = x O x c + a. Then, solvng for b NT B : 19 b NT B = x x M c + a (6) Wth negatve turnout buyng, the machne must compensate opposng voters for: (1) the forgone utlty of votng for ther preferred party; and (2) the cost they ncur by abstanng. Gven these requred payments, we now determne the optmal mx of clentelst strateges. Observe that whereas vote buyng yelds two net votes, negatve turnout buyng (as wth turnout buyng and double persuason) yelds only one net vote. As a result, the machne s wllng to pay the most-expensve vote-buyng recpent twce as much as t s wllng to pay the most-expensve ctzens purchased wth negatve turnout buyng, turnout buyng, and double persuason. Thus, when negatve turnout buyng s a vable strategy, the machne agan optmally sets b V B = 2b NT B = 2b T B = 2b DP. An mportant fndng follows mmedately for contexts where negatve turnout buyng s possble. Observe that f the most-expensve payment for any strategy s greater than zero, then the most-expensve payments for all four strateges must be greater than zero. 19 Recall that by the assumpton of symmetrc party platforms, x M = x O. 21

26 Therefore, n such contexts, machnes dstrbutng benefts durng campagns wll optmally engage n all four strateges of vote buyng, turnout buyng, double persuason, and negatve turnout buyng. [Insert Fgure 4 here] To provde ntuton about how the machne optmally combnes all four strateges, Fgure 4 bulds on the vertex shown n Fgure 2b. The logc for turnout buyng and double persuason, whch target nonvoters, remans the same as n the base model. By contrast, the machne now faces a trple choce wth each opposng voter reward her for stayng home (negatve turnout buyng), reward her for votng aganst her preferences (vote buyng), or provde her wth no reward. The machne s decson for each opposng voter can be analyzed n two steps: (1) dentfy whether vote buyng or negatve turnout buyng s more cost-effectve; and then (2) dentfy whether the more cost-effectve strategy s preferable to provdng no reward. To examne the frst step, assume the machne rewards an opposng voter W of type t w = (x w, c w ). Intutvely, vote buyng yelds double the net votes, so t s more attractve to pay W to swtch her vote, unless dong so s more than twce as expensve as payng W to stay at home. Therefore, gven the requred payments for each strategy (Equatons 2 and 6), the machne chooses negatve turnout buyng under the followng condton: 20 2xw > 2[ xw x M cw + a] c w > x O + a (7) Ths condton s shown n Fgure 4 as horzontal lne segment l If the machne rewards an opposng voter located above l 6, negatve turnout buyng s more cost-effectve. If the machne rewards an opposng voter located on or below l 6, vote buyng s more cost-effectve. The next step s to determne whether the more cost-effectve strategy s preferable to provdng no reward. To ths end, we consder the most-expensve payments for each strategy. 20 We assume that f both strateges are equally cost-effectve, the machne engages n vote buyng. 21 The lne segment l 6 s gven by the equaton c = x O + a, on the range from X O to the vertcal ntercept. 22

27 The machne pays the most-expensve vote-buyng payment b to opposng voters who are located on or below l 6, and who requre exactly b to swtch ther votes. Smlar to Fgure 2b, such voters are located on the vertcal lne segment l 3 n Fgure 4. Wth respect to negatve turnout buyng, the machne pays the most-expensve payment b to opposng voters who are located above l 6, and who requre exactly b to stay at home. In accordance wth Equaton 6, these are opposng voters of type t h = (x h, c h ) for whom b = x h x M c h + a. Such opposng voters are located on lne segment l 7, whch extends from X O to the pont where l 7 ntercepts wth l The shaded areas n Fgure 4 reflect the same logc as the base model: the machne buys ctzens f and only f ther requred payments are less than or equal to the most-expensve payment. Taken together, these fndngs suggest how the machne optmally allocates resources across clentelst strateges, when negatve turnout buyng s vable: Vote Buyng: Pay b V B to opposng voters f b V B Negatve Turnout Buyng: Pay b NT B Turnout Buyng: Pay b T B Double Persuason: Pay b DP b and c x O + a to opposng voters f b NT B to supportng nonvoters f b T B to opposng nonvoters f b DP No Payment: Make no payment to all other ctzens Emprcal Evdence b b b and c > x O + a The model of electoral clentelsm developed above offers nsghts nto how characterstcs of poltcal envronments affect the strateges that machnes employ. Emprcal evdence from Brazl, Russa, and Argentna s consstent wth the model s predctons. Gven space constrants, we focus specfcally on how two factors compulsory votng and machne support affect the relatve prevalence of vote buyng and turnout buyng n each country. The model predcts, ceters parbus, that whereas machnes vote buy more ctzens when votng s compulsory, they turnout buy more ctzens when votng s optonal. Compulsory votng ncreases overall turnout, ncludng that of weakly opposed voters, who are the cheap- 22 The lne segment l 7 s gven by the equaton c = x x M + a b, on the range from X O to the pont where l 7 ntercepts wth l 6. 23