Poltcal Competton and Invald Ballots n Mexco: evdence from subnatonal data Irvn Mkhal Soto Zazueta Unversdad Autónoma de Snaloa Facultad de Cencas Económcas y Socales Josefa Ortz de Domínguez S/N. Cudad Unverstara Culacán, Snaloa, Méxco Telephone: +52 (667) 713 38 03 e-mal: rvn.soto@uas.edu.mx "Paper presented at the XXVI Congreso Naconal de Estudos Electorales: Balance de la Reforma Electoral 2014. From 14-16 October 2015, Monterrey, NL, Mexco"
Poltcal Competton and Invald Ballots n Mexco: evdence from subnatonal data Abstract: Ths paper analyzes the determnants of spoled ballots at Mexcan muncpal electons. We use a local electons data base from 2006-2008 to estmate an electoral behavor model n whch local economc varables are ncluded n quadratc form, so that non-lnear effects can be taken nto account. We found that an ncrease n the electoral margn s assocated wth an ncrease n the proporton of spoled ballots Mexcan local electons, the effect s hghly statstcally sgnfcant and stable across specfcatons. Keywords: spoled ballots, poltcal competton, margnalzaton, economc condtons. 1. Introducton An ndspensable condton of democratc regmes s the promoton of free and compettve electons. The consoldaton of democracy, however, requres the partcpaton of ts ctzens n the electon of hs/her representatves. Partcpaton at electons s crucal for well-functonng poltcal systems (De Paola and Scoppa, 2012). Ths process not only represents the most common way of poltcal partcpaton, but also legtmzes and strengthens the poltcal system. On the other hand, not all ctzens choose to partcpate, some of them prefer to abstan and others prefer to spol ther ballot, whether ntentonal or not. Accordng to Power y Garand (2007), despte the large number of studes addressng the ssue of votng, the most of the attenton has been dedcated on voter turnout. Understandng the reasons that lead people to vote n large electons s of major nterest both for poltcal scentsts and for economsts. However, lttle attenton has been gven to nvald ballots, the castng of blank or spoled ballots remans a gapng lacuna n the lterature on comparatve poltcal behavor. Voter abstenton and spoled ballots are part of a system where electons send sgnals to the poltcal system (Damore, Waters and Bowler, 2011). However, those sgnals are not clear due to they may arse from motvatons varyng from a way of protest to even apathy and confuson. More precsely, authors lke Stefbold (1965), Power and Garand (2007), Uggla (2008), Galatas (2008), Obradovc-Wochnk and Wochnk (2014) pont out that unmarked and spoled ballots ballot can be read as ncreasng ctzen apathy towards 1
formal poltcs, also t can be understood as new forms of poltcal partcpaton, especally n poltcal contexts charactersed by low publc trust n key nsttutons, dsappontment after transtons to democracy and lack of formal poltcal partcpaton. Scholars nterpret blank and spoled ballots as resultng from some combnaton of voter ncapacty, where ctzens lack the requste sklls or nformaton to cast a vald ballot, and poltcal motvatons, when voters delberately sgnal ther malcontent (Drscoll and Nelson, 2014). Stefbold (1965) pont out that some voters submt an nvald ballot for reasons of gnorance, ambvalence, or apathy when people nvoluntary spol hs/her ballot. Whle some others delberately nvaldate the ballot as an act of poltcal protest generated by possble mstrust n the poltcal nsttutons, dssatsfacton wth the regme, ncumbents, or the electoral process more generally. Spoled ballots generally represent a small share of total ballots, but ther mportance s ncreasng over tme and an analyss of electoral behavor seems relevant as spoled ballots are probably not random events. Power and Garand (2007), Uggla (2008) and Drscoll and Nelson (2014) observed that s common to fnd electons where many ballots are returned spoled n Latn Amerca. They see to Latn Amerca as an deal settng for a comparatve study of nvald votng; some countres have spoled ballots n the range of 2 e 3%, whle others have averages that are ten tmes hgher. Also, spoled and blank votes have exceeded 15 percent of total votes cast n a thrd of the electons snce 2000. In Mexco, the evoluton of spoled ballots n federal electons follows a welldefned pattern; frst, t s observed a decrease n rates of spoled ballots snce 1970s to 2000, then shows an ncreasng trend mantaned through the frst decade of the present century. Accordng to Csneros-Yescas (2013), spoled ballots n Mexco has become an mportant ndcator of dscontent and smear perceved by ctzens of poltcal nsttutons and democracy. The 2009 federal electon was characterzed by the emergence of a ctzen protest n dfferent parts of the country that called to the electors to cast a null vote to show ther dscontent wth poltcal partes and ther representatves. In the 2009 ntermedate electons, spoled ballots represented 5.41% of the total votes cast, a number greater than votes obtaned by the PT, Convergenca, Nueva Alanza and PSD (poltcal partes). However, ths s not a homogenous behavor among states, the 2
rates vared between 2.9 and 10.9%. The number of spoled ballots n muncpaltes are hgher than the total votes receved by some poltcal partes, too. In local electons of 2008 n Huchpan, Hdalgo, for example, the rate of spoled ballots reached 14% of total votes cast. Ths percentage s hgher than the percentage of votes receved by 8 of the 10 partes or coaltons that competed n that electon. Even the heterogenety s greater among muncpaltes. Galatas (2008) argue that some blank and spoled ballots result from systemc factors and predctable patterns assocated wth the context of the electon. Thus, spoled ballots, although rare, are not unmportant; studyng spoled ballots provdes a rcher understandng of electoral behavor (Galatas, 2008:448). Moreover, Scholars clam that the percentage of spoled ballots seem to be related to soco-economc and nsttutonal factors, such as educaton, ncome and votng systems. For ths reason, t s mportant to dentfy and evaluate the dfferent factors that nfluence ths electoral behavor. We wll focus our analyss on muncpal electons because of the hgher varaton n rates of spoled ballots at the muncpal level. In ths context we ask what are the determnants of electoral behavor n Mexco? Even further, what factors nfluence dfferences n spoled ballots n muncpal electons? Paola and Scopa (2012) explan that poltcal competton may also affect ths type of voter choces, but the relatonshp could be ether postve or negatve. On the one hand, poltcal competton has a postve effect on voter turnout. As the chances of castng the decdng vote are hgher n more compettve electons, some authors pont out that the lkelhood of castng a blank ballot should be lower. Endersby, Galatas, and Rackaway (2002) and Aldashev and Mastrobuon (2013) argued that spoled ballots may decrease because of the hgher expected benefts dervng from closeness, related both to the hgher probablty of beng pvotal and to the hgher costs of expressng dscontent through a voluntary nvald ballot. On the other hand, De Paola and Scopa (2012) menton that a postve relatonshp may be possble when poltcal competton ncreases socal pressure, leadng to a hgher cost of not gong to the polls, but electors may stll escape from supportng the canddates by castng a blank or nvald ballot. 3
In ths paper, we propose an econometrc model to study how the closeness relates to the fracton of spoled ballots n Mexco. We examne electoral nformaton n 2013 muncpaltes that took place between 2006-2008. Few studes have tred to dentfy the factors that explan the fracton of spoled ballots n Mexco. Most studes focus on descrbng ndvdual electoral behavor through opnon surveys, or through hstorcal, terrtoral and conceptual analyss. Even more, whle there are a number of works analyzng the effect of electoral competton on turnout, only few emprcal studes have looked at the spoled ballots. The present artcle ncludes fve sectons: In the second secton, we revew the man factors nfluencng these varables. The econometrc model, as well as the technque that s used to estmate t, s represented on the thrd secton. The emprcal analyss s done on the fourth secton and we show the man conclusons of the study on the ffth secton. 2. Lterature revew Ths secton presents a theoretcal framework to help us explan the dfferentated behavor observed n the prevous secton. Exstng lterature dentfes a lst of varables that affect electoral turnout and n consequence affects voter abstenton. However, only a few studes have emprcally analyzed the phenomenon of spoled ballots. A small academc lterature assesses ths phenomenon n comparatve perspectve, systematcally consderng blank and spoled votng n countres around the world. Some scholars have studed the effect of some nsttutonal, poltcal envronments and soco-economc condtons on spoled ballots (Power and Garand, 2007). From the nsttutonal approach, academcs dentfy nsttutonal desgns, complexty of the electoral system and ballot structure. Ths approach draws attenton to the legal and formal structures that make votng desrable versus undesrable, or smple versus dffcult (Power and Garand, 2007). McAllster and Makka (1993) noted that blank and spoled ballots are more common n the Senate contests owng to the relatvely complcated sngle-transferable vote procedure. Also, Power and Roberts (1995) reported that Brazlan Senate, whch uses the smple sngle-member dstrct pluralty method, has fewer nvald ballots than the Chamber of Deputes that uses under open-lst proportonal representaton. They concluded that vote nvaldaton s more common n open lst proportonal representaton systems n whch 4
many canddates are lsted on the ballot, creatng a comparatvely nosy electoral envronment. Poltcal-protest approach clams that spoled ballots s an ndcator of voter dscontent that may reflect rejecton of ncumbents, condemnaton of the exstng poltcal regme (.e., an ant-system orentaton), or some combnaton of all of these factors (Power and Garand, 2007; Galatas 2008; Drscoll and Nelson, 2014). 1 Stefbold (1965) rejected blank and spoled ballots were smply a result of voter error, reportng that blank and spoled ballots ncreased followng the prohbton of the Communst party n West Germany suggestng the optons presented on the ballot were nsuffcent. Several studes have documented that compulsory votng electoral systems are very effectve to ncrease electoral turnout, but when protest stuatons arse, the votes are translated nto an ncrease n blank and spoled ballots. Power and Garand (2007) argued n countres where mandatory votng s strctly enforced, ctzens who would normally abstan under voluntary votng condtons may turn out to the polls only to cast a blank or spoled ballot. Usng an aggregate-level panel-data analyss from legslatve electons held n 18 democraces n Latn Amercan between 1980 and 2000, Power and Garand (2007) found that compulsory votng, electoral dsproportonalty, and the combnaton of hgh dstrct magntude wth personalzed votng ncrease the number of blank and spoled ballots. Uggla (2008) assess the extent to whch voters spol or annul ther ballots n electons at Western Europe, Australa, New Zealand, and the Amercas n the 1980-2000 perod. He concluded that votng s a conscous act related to the poltcal choces avalable through the electon. Precsely, he found that varaton n the fracton of nvald ballots reflect the voters reacton to the perceved absence of poltcal choce. Obradovc-Wochnk and Wochnk (2014) analyzed Serba s 2012 electons. Demonstratng a deep dssatsfacton wth poltcal partes and ther functonares who embody and enact democratc practces, they argued that some nstances of ballot nvaldaton can be understood as a reacton to the broader crses of democracy. They pont out ballot-nvaldatng voters n Serba are not only dssatsfed wth these actors 1 Ths authors menton that unmarked and spoled votng s a sgnal of protest, especally n systems where votng s compulsory. 5
alleged corrupton and neffectveness, but are also deeply dscontented wth the electoral system, whch encourages a lack of transparency and clentelsm. Poltcal competton may also affect the percentage of nvald votes, but the relatonshp s uncertan. Galatas (2008) argued that castng blank ballots seems ntally rratonal from a ratonal choce perspectve: a voter who shows up on Electon Day to cast a blank or ntentonally spoled ballot ncurs the cost of votng, yet receves no beneft. However, snce Rker and Ordeshook (1968) ncluded cvc duty to the Downsan equaton, 2 Galatas explaned that a voter receves a net beneft from castng a ballot n terms of havng performed a cvc duty, even f no addtonal beneft occurs from havng cast the decdng ballot for a specfc party or canddate, so voters derve benefts from castng a blank or unmarked ballot. Even more, a postve relatonshp may be possble when poltcal competton ncreases socal pressure, leadng to a hgher cost of not gong to the polls, but electors may stll escape from supportng the canddates by castng a blank or nvald ballot (De Paola and Scopa, 2012). Galatas (2008) found that unmarked ballots are reduced n the most compettve electons. Closeness ncreases the lkelhood of nfluencng the outcome of the electon and rases the opportunty cost of express dscontent through voluntary nvaldaton of the ballot; the ncentve to cast a blank ballot should be lower. Therefore, a negatve relatonshp s expected to exst between party competton and blank ballots. Usng a dataset from the Italan parlamentary electons n 1994-2001, Aldashev and Mastrobuon (2013) focus nstead on the relatonshp between the closeness and the number of nvald ballots. They found a strong negatve relatonshp between the margn of vctory of the leadng canddate over the nearest rval and the share of nvald ballots. These authors explan that ths relatonshp holds when electon offcers are unbased because they ncrease ther effort to detect hgher rates of nvald ballots n response to the hgher closeness of the electoral race. Offcers wth bas towards any of the canddates are less 2 Cvc duty provdes postve feedback that one contrbutes to the democratc process and enters the equaton as a net beneft to the voter, apart from other benefts lke helpng a preferred canddate or party be elected. 6
lkely to nvaldate ballots favorng ts preferred canddate, whle t would be more lkely to nvaldate ballots aganst ther canddate, n ths case the ambguous serous effect. De Paola and Scopa (2012) used data from Italan muncpal electons from 1993 to 2011 to nvestgate whether poltcal competton affects vald, blank and nvald ballot, controllng for muncpaltes and canddates characterstcs and for muncpal fxed effects. They found that electoral margn produces a postve hghly statstcally sgnfcant effect on the percentage of blank ballots, that s, when the race s closer ndvduals are less nduced to cast a blank ballot. They mentoned that ther fndngs support the dea that the behavor of electors castng blank ballots s not very dfferent from the behavor of electors who abstan from gong to the polls, so they react n the same way when the electoral race becomes closer..e., that the expected benefts of votng ncrease n closer compettons. Invald ballots also show a smlar pattern, but the effects are statstcally nsgnfcant. In ths case, the arguments of Aldashev and Mastrobuon (2013) about a potental hgher effort provded by unbased electon offcers n checkng ballots n response to closeness, whch would lead to a negatve correlaton between nvald ballots and the electoral margn does not seem to be suffcent to counterbalance the effects dervng from electors behavor (De Paola and Scopa, 2012). Fnally, some emprcal studes have found evdence of the effect of urbanzaton, ncome nequalty and educaton on the percentage of nvald votes. Power and Garand (2007) mentoned that socoeconomc development and urbanzaton favor wder crculaton of and more democratc access to the poltcal nformaton necessary to complete a ballot n natonal electons. Educaton and lteracy contrbute to the poltcal skll levels of ndvdual voters. Emprcal research verfes ths assumpton, reportng that blank and delberately nullfed ballots are more common n dstrcts where llteracy s hgh. Knack and Kropf (2003), McAllster and Makka (1993) and Snclar and Alvarez (2004) found that nvald votng occurred most frequently n precncts where ballot comprehenson s lmted due to lngustc or ethnc dversty. Power and Garand (2007) observed large numbers of voters n Latn Amerca reman llterate or sem-lterate, makng t dffcult for them to complete a ballot paper. 7
Galatas (2008) noted that spoled votes tend to be postvely assocated wth low levels of educaton n provncal electons n Ontaro. He argued that less educated voters, lackng requste knowledge of poltcs, poltcal ssues, partes, and canddates, may be more lkely to cast a spoled ballot or leave a ballot unmarked. However, he sad that educaton may ncrease spoled ballots when t represents a form of protest that rejects the exstng partes and canddates n contenton. In the 2009 federal electons, Csneros-Yescas (2013) found a postve effect of educaton on spoled ballots n Mexcan muncpaltes: the hghest levels of muncpal schoolng, the hghest percentage of spoled ballots. Ths fndng contradcted the common pattern presented n prevous federal electons. Furthermore, the presence of moblzaton n favor of nvald vote caused that the postve magntude between schoolng and nvald vote was stronger. 3. Methodology 3.1.Econometrc model Gven the foregong dscusson, we suggest that spoled ballots are a functon of a varety of nsttutonal, socoeconomc, and protest-democracy varables. In ths secton, a model for estmaton s presented, as well as the econometrc approach used to establsh determnants of spoled ballots n Mexcan muncpal electons. We propose a cross-secton analyss, usng data from 2013 muncpaltes. Consderng the prevous lterature revewed and followng to Soto and Cortez (2014), who studed the determnants of turnout n Méxco, the econometrc model we wll estmate could be expressed as follow: y MV IM IM DEI c 8 o 1 j t 2 3 2 e 4 2 5 pob.rural 6 1,...,2013 pob.nd y represents the proporton of spoled ballots n the muncpalty, that s obtaned from local electons celebrated between 2006 and 2008. We control for economc adversty ncorporatng n the model a muncpal margnalzaton ndex (IM) created by the Naton Populaton Councl (CONAPO): e s the average of years schoolng, 7 [1] pob. rural s the proporton of the populaton that lves n rural zones, pob. nd s the percentage of the 8
populaton n the muncpalty that s ndgenous, DEI s a dummy varable that takes a value of 1 f the local electon s ntermedate or zero otherwse. Fnally, we nclude dummy varables to control the specfcatons for each state and year n whch the muncpal c j electon was held, expressed as and v t, 2008). As well, represents the term for error. n the same order (j= 1, 2,, 32; t= 2006, 2007, Because of the nature of the varables ncluded n the emprcal analyss, potental problems exst that should be resolved frst wth the object of obtanng unbased and effcent coeffcents. One of the defcences comes from the nature of our dependent varable, whch s expressed as a proporton. Papke and Woolrdge (2008) pont out that when a varable that s beng explaned has values between zero and one, t s necessary to realze a transformaton log-odd, log y /(1 y), to be able to do an estmaton usng MCO. Even f ths does not happen, the results are dffcult to nterpret once the transformaton s done. To solve ths problem, the authors developed a method called fractonal probt. One advantage of ths technque s that t allows you to have n mnd the non-lneal relatonshp of the varables. We can estmate ncreasng or decreasng partal effects accordng to the stuaton. Ths way, through estmaton wth quas-maxmum lkelhood (QMLE), Papke and Woolrdge obtan robust estmators that satsfy the propertes of effcency when the dependent varable les between zero and one. Also, utlzng a model of robust regresson benefts the unobserved heterogenety. We express the fractonal probt model n the followng way: E ( y x, c ) G( x c ) Where G( ) represents a normal standard accumulatve functon of dstrbuton, the dependent varable refers to spoled ballot, c s the unobserved effect of each muncpalty, x s the conjuncton of explcatory varables consdered n the prevous secton and represents the parameters of the model that wll be estmated. To know the margnal effect of some explcatory varables on nvald ballots, we calculate the partal dervatve of the objectve functon n relaton to the condtoned varance: [2] 9
E( y x) k g( 0 x ) x k varable Gven that G( ) s strctly monotonc, the drecton of the partal effect of the x on spoled ballots s the same as the sgn of. Ths way, consderng the k possble functonal relatons between dscussed varables n secton 4 and the electoral turnout, the model proposed wth ths technque s expressed the followng way: k [3] 2 o 1MV 2IM 3IM 4 E( y x, c ) G 8mg 9DEI c j t 2 e 1,...,2013 5 pob.rural 6 pob.nd 7 [4] 3.2.Data The key varable of nterest s the rate of spoled ballots, n our study s measured as spoled ballots as a percentage of all ballots cast n local electons for each muncpalty. In order to estmate the parameters of our model; we collected data on the poltcal varables n the model for 2013 muncpaltes n Mexco n each local electon held from 2006 to 2008. Muncpaltes created after the year 2005 were left out of the analyss, as well as those that selected mayors through a system of uses and customs. In addton, we use the 2005 Mexcan Census of Populaton and Natonal Councl of Evaluaton of Polcy for Socal Development to obtan data at the muncpal margnalzaton ndex, educaton, rural and ndgenous populaton, ncome nequalty. In graphc 1 we present the hstogram and the kernel densty plot for the dstrbuton of spoled ballots for the 2013 muncpaltes for local electons. As one can see, the overall dstrbuton s skewed heavly to the rght. The percentage of spoled ballots represent on average, 2.65% of the votes cast n the electons for mayor n a typcal muncpalty, the standard devaton s 1.84%. Most of the muncpaltes have percentages of spoled ballots below 5%; n fact, only 15% of observatons have spoled ballot percentages of 5% or greater, though there are a few cases n whch the percentage of nvald ballots exceeds 10%. Clearly, n some muncpaltes spoled ballots are a commonplace occurrence, whle n other countres spoled ballots are relatvely rare. 10
0 Densty 10 20 30 40 Graphc 1. Dstrbuton of spoled ballots cast n local electons n Mexcan muncpaltes, 2006-2008. 0.05.1.15 Por. Nulos Source: Own elaboraton wth data from local electoral nsttutes. In table 2 we present some descrptve statstcs and we descrbe the varables consdered n the econometrc model. Voter abstenton s measured as the number of people nscrbed n the nomnal lst that dd not vote as a percentage the total number of cctzens n the nomnal lst for each muncpalty. The varable ranges from zero to one, the source of nformaton were the electoral nsttutes of each state. In a typcal muncpalty the rate of voter abstenton on local electons s 42.5%. t s sgnfcantly greater than spoled ballots, wth a standard devaton of 14.8%. To measure the poltcal competton we use the electoral margn (MV) n muncpal electons,.e. the dfference of the percentage of votes cast for the wnner n the electon mnus the votes obtaned by the second party. The source of nformaton for the constructon of ths varable was the data base of local electons publshed by the Center of Investgaton for Development AC (CIDAC) on ther webste. The average electoral margn n muncpaltes s equal to 16.3%. In the closest electons n our data shows a vote 11
dfference of 0.6% between the two leadng canddates, whle n electon wth the hghest margn s 100%. Ths dfferences are statstcally possble for two reasons: the frst s that n table 1 we can observe that at least n one muncpalty there was only one compettor. The second s that n some muncpaltes, the PRI stll mantans hegemonc power wth a practcally nonexstent competton. The heterogenety of the muncpaltes s manfested n the dsperson of the varables. For example, the lowest rate of electoral turnout was 11.6% (hghest abstenton s 88.4%), whle the hghest rate was of 92.3% (lowest abstenton s 7%). Varables Unt of measure Obs. Table 1. Descrptve Statstcs Average Std. Devaton Mn Max Perod Source Spoled Ballots % 2027 0.0265 0.0184 0 0.1626 2006-2008 Electoral Insttutes Voter Abstenton % 2022 0.4251 0.1477 0.0704 0.8839 2006-2008 Electoral Insttutes Electoral Margn % 2027 0.1630 0.1114 0.0060 1.00 2006-2008 Margnalzaton Index Unts 2454 0.3446 0.1457 0 1 2005 Educaton Unts 2456 6.0611 1.5587 1.41 12.89 2005 Rural populaton % 2456 0.6079 0.3616 0.0000 1.00 2005 Indgenous populaton Income nequalty (Gn) Center of Investgaton for Development AC (CIDAC) Natonal Councl of Evaluaton of Polcy for Socal Development (CONEVAL) Populaton and household count of 2005 (INEGI) Populaton and household count of 2005 (INEGI) % 2456 0.2491 0.3574 0.0000 1.00 2005 INEGI Unts 2453 0.4160.0482.2469 0.6895 2005 Natonal Councl of Evaluaton of Polcy for Socal Development (CONEVAL) Intermedate Unt 2027 0.7618-0 1 2006-2008 Electoral Insttutes electons Source: Own elaboraton wth data from dverse sources. Concernng socoeconomc condtons, the dfferences are enormous. We used the ndex of muncpal margnalzaton of 2005, elaborated by the Natonal Councl of 12
Populaton (CONAPO) to measure the margnalzaton of the muncpaltes we used. In ths case, we normalze the varable so that s takes values from zero to one. Also, educaton (e) s measured by the average number of years of schoolng n the muncpalty. The source of nformaton used s the populaton and household count of 2005. The average years of educaton n the populaton are 6.06 n the muncpaltes, the hghest value s 12.89 years of educaton whle the lowest s 1.4 years. To prove the hypothess that urbanzaton takes to a debltaton of the nterpersonal lnks, we nclude the proporton of the populaton that lves n rural zones (Rural populaton). In effect, there are muncpaltes wth 100% of ther populaton lvng n rural locatons, whle n other muncpaltes nobody lves n rural zones. Although on average 60.8% of populaton n muncpaltes lves n rural zones wth a standard devaton of 36%. The same occurs wth the ndgenous populaton, n average s of 24.9%, wth a standard devaton of 35%. To prove the argument of socal homogenety we use the percentage of ndgenous populaton n the muncpalty. The source of nformaton s the populaton and household count of 2005 done by the INEGI. We used the Gn Index at a muncpal level estmated by the Natonal Councl of Evaluaton of Polcy for Socal Development (CONEVAL) to control the dfferences 2 caused by ncome nequalty ( ). Intermedate electons (DEI) s a dummy varable that takes values of 1 f the local electon s ntermedate or zero otherwse. The source of nformaton was the electoral nsttute of each federal entty. 4. Emprcal Analyss Earler, the theoretcal expectatons of the relatonshps between the dynamcs of the party system and the castng of spoled ballots were dscussed. As explaned n secton 2, the relatonshp between closeness and spoled ballots could be ether postve or negatve. To test ths relaton, frstly a smple correlatons (Pearson s r) explanatory analyss was made among ndependent varables of the equaton to avod multcollnearty problems. We found that the dfferent varables that could be used to measure the economc adversty such as poverty, margnalzaton and per capta ncome are hghly correlated. In the same sense, these varables are correlated wth the average years of educaton, for ths reason we decded not to nclude them n the same model. 13
In table 2 we report the margnal effects estmated for our model of spoled ballots n Mexcan muncpaltes from 2006-2008, gven the medan values of each varable. 3 We present four models accordng to varables ncluded n the regresson analyss. In the columns [1] and [2] are reported estmaton results obtaned replcatng the specfcatons estmated by Soto and Cortez (2014), but consderng as dependent varable the percentage of spoled ballots, also educaton and ncome nequalty are excluded from the model to avod multcollnearty problems, whle they are ncluded n columns [3]. In the last column we try to solate the margnal effect of electoral margn on spoled ballots. Table 2. Margnal effects on spoled ballots n Mexco Varables [1] [2] [3] [4] Medan Values Constant Electoral Margn 0.0055 *** (0.0020) 0055***.0020 Gn - - Margnalzaton Index Margnalzaton Index squared 0.0197 *** (0.0057) -0.0019 (0.0072) -0.0018 ** (0.0007) 0.0017 ** (0.0009).0182***.0023 0.0060 *** (0.0022) 0.0260 *** (0.0042).0052**.0022 0.2000 0.4240-0.3211 - - 0.1222 Rural populaton -.0018*** -0.0004.0007 (0.0007) Indgenous populaton.0017** 0.0038 ***.0008 (0.0007) Educaton - (0.0002) -0.0011 *** Intermedate electon 0.0088 ***.0088*** 0.0085 ***.0066*** (0.0008).0008 (0.0008).0008 Fx effects yes yes yes yes Obs 2013 2026 2013 Source: Own elaboraton *** Level of sgnfcance at 1%,** Level of sgnfcance at 5%, *Level of sgnfcance at 10%. Standard error n parentheses. 0.5449 0.1874 6.2865 0.7137 Our fndngs provde evdence that patterns of spoled ballots are explaned n part by varaton n the poltcal competton. In all specfcatons we fnd that Electoral Margn produces a postve effect on the percentage of spoled ballots, the effect s hghly 3 Because of the non-lneal structure of the model, the coeffcents obtaned through ths technque cannot be nterpreted drectly. Therefore, we do a transformaton accordng to the establshed crtera n the equaton (3) and those are reported n the chart. 14
statstcally sgnfcant (the t-stat s typcally greater than 6 n absolute value) and stable across specfcatons. An ncrease n 10% on the electoral margn s assocated wth an ncrease n the proporton of spoled ballots of 0.6%. That s, when the race s closer ndvduals are less lkely to null ther vote. These fndngs are consstent wth the dea that spolng ther ballot has a greater cost of opportunty when the electon s very compettve. In ths sense, the behavor of electors castng null ballots s not very dfferent from the behavor of electors who abstan from gong to the polls, so they react n the same way when the electoral race becomes closer. Aldashev and Mastrobuon (2011) shown that margn at electoral unt level has a negatve effect on nvald ballots, whch s the opposte to the predcton of explanaton based on voter protest. They found that f the leadng margn ncreases from one electon to the other wthn the same electoral unt, the fracton of nvald ballots reported at ths unt decreases. The margnal effect for ncome nequalty s postvely related wth the percentage of null votes. Power and Garant (2007) contend that ncome nequalty wll be negatvely related to nvald votng, snce n countres wth hgh ncome nequalty the electorate wll be more compact and wll lkely have a hgher proporton of the hgh-ses, hgh competency voters who are less prone to castng nvald ballots. Accordng to them, our emprcal s not consstent wth ths hypothess. On the other hand, squared margnalzaton ndex shows no statstcal sgnfcance on spoled ballots, suggestng the exstence of a lneal relatonshp between the percentage of spoled ballots and the grade of margnalzaton. 4 Margnalzaton represents a wde range of characterstcs of the electorate that would be related to voter competency, and hence we hypothesze that voters n relatvely wealther muncpaltes wll exhbt lower levels of spoled ballots. Our results n are consstent wth ths hypothess, the coeffcent for margnalzaton ndex s n the expected drecton, and t shows statstcal sgnfcance at conventonal levels. An ncrease of 10 decmals of a unt of the margnalzaton ndex causes an ncrease of 0.2% n spoled ballots. 4 Ths result s mantaned usng other varables of economc adversty such as poverty, the IDH and the ndex of socal margnalzaton. 15
The average years of educaton are also negatvely assocated wth the percentage of spoled ballots n Mexco. Ths fndng s consstent wth the assumpton that educaton and lteracy contrbute to the poltcal skll levels of ndvdual voters. Ths fndng s also consstent wth emprcal research that report blank and spoled ballots are more common n dstrcts where llteracy s hgh. Galatas (2008) suggest that people wth hgh educaton levels are less lkely to make a mstake whle markng the ballot, choosng voter abstenton as a mechansm of protest. Accordng to Csneros-Yescas (2013), educaton had a negatve relaton wth spoled ballots n the federal electons of 2000, 2003 and 2006 n Mexco,.e. n these electons the null vote ncreased n muncpaltes wth hgher average schoolng. It s common to see n lterature arguments that stablsh that urbanzaton wll depress nvald ballots, snce ctzens n urban areas are more lkely to be exposed to the knds of poltcal organzaton and easly-accessble nformaton that wll reduce nvald ballots. We found that rural muncpaltes had a negatve assocaton wth spoled ballots, whch makes sense snce, Power and Roberts s study (1995) from Brazl found that urbanzaton appeared to have a postve effect on nvald votng for the upper house. Data also ndcates that an ncrease of 10% n the rural populaton reduces the rate of spoled ballots approxmately 0.18%. The percentage of ndgenous populaton s assocated wth a hgher percentage of null votes because of the propensty people have to make a mstake; an ncrease of 10% n the ndgenous populaton ncreases the rate spoled ballots by approxmately 0.17%. 5. Conclusons Before presentng the conclusons, t s convenent to revew some lmtaton of ths paper, partcularly those from the characterstcs of avalable nformaton. One of the clear lmtatons of ths research s temporalty, t s necessary to elaborate a panel data econometrc model to study the consstent of the mpacts on the temporal horzon. We used statstcal technques to solve the problem of the dependent varable and the heterogenety. Usng data from muncpal electons, we estmated the margnal effects of poltcal, economc and demographc varables used n nternatonal studes. The results obtaned helped us to corroborate some emprcal rregulartes found n other works. In 16
addtons, we detected some varables n Mexco that do not behave the same as n other countres. The man queston of ths research s: what factors nfluence dfferences n spoled ballots n muncpal electons n Mexco? Even more, does poltcal competton reduce the percentage of spoled ballots? Internatonal lterature suggests that the cost of expressng dscontent through a voluntary nvald ballot s hgher when the race s close, for ths reason. Our hypothess s that patterns of spoled ballots n Mexco are explaned n part by varaton n the poltcal competton. We found a postve mpact of Electoral Margn on the percentage of spoled ballots n all the specfcaton of the econometrc model; the effect s always hghly statstcally sgnfcant. That s, when the race s closer ndvduals are less lkely to null ther vote. These fndngs are consstent wth the dea that spolng ther ballot has a greater cost of opportunty when the electon s very compettve. In ths sense, the behavor of electors castng null ballots s not very dfferent from the behavor of electors who abstan from gong to the polls, so they react n the same way when the electoral race becomes closer. Ths analyss shed some consderatons that should serve as gudelnes for electoral poltcs. An essental part of democracy conssts n guaranteeng ctzens true optons n electoral competton and these optons are not measured by the number of canddates, but rather the level or representatveness that each canddate offers. Ths way, one of the problems of the Mexcan electoral system s the lack of a real mechansm to express the unconformtes durng electons. In present day, when ctzens do not feel represented by any of the canddates or they do not represent a real votng opton, the only form of protest s through voter abstenton or spolng the ballot. A measure that mght solve ths problem s creatng an opton of none of above. Wth ths mechansm, we could separate people who smply make a mstake whle votng from ctzens that smply do not feel represented by any of the extng poltcal partes. Second, people who use voter abstenton as a mechansm of protest would be reduced, therefore decreasng the analytcal problems assocated wth the electoral data base. Addtonal research s needed, for example, to nvestgate the consequences and motvatons people have when they spol ther ballot. A second lne of nvestgaton that derves from ths work s the analyss of the mpact of poltcal competton on the 17
democratzaton process n Mexco. A thrd aspect s the relatonshp between subjectvewelfare and partcpaton n Mexco. References Aldashev, Gan, and Govann Mastrobuon. 2013. Invald Ballots and Electoral Competton. Workng Paper. Csneros-Yescas, G. Isaac. 2013. Movlzacón, escolardad y voto nulo: La eleccón federal de 2009 en Méxco. Polítca y Goberno 20(1): 39-78. Damore, Davd, Mallory M. Waters, and Shaun Bowler. 2011. Unhappy, Unnformed, or Unnterested?: Understandng None of the Above Votng. Poltcal Research Quarterly 65(4), 895 907 De Paola, Mara, and Vncenzo Scoppa. 2012. The causal mpact of closeness on Electoral partcpaton explotng the Italan dual ballot system. Unverstà della Calabra, Dpartmento d Economa e Statstca, Workng Paper n. 03. Downs, Anthony (1957), An Economc Theory of Democracy, Nueva York: Harper and Row. Drscoll, Amanda, and Mchael J. Nelson. 2014. Ignorance or Opposton? Blank and Spoled Votes n Low-Informaton, Hghly Poltczed Envronments. Poltcal Research Quarterly. Endersby, James W., Steven E. Galatas, and Chapman B. Rackaway. 2002. Closeness Counts n Canada: Voter Partcpaton n the 1993 and 1997 Federal Electons. Journal of Poltcs 64 (2): 610-31. Galatas, Steven. 2008. None of the Above? Castng Blank Ballots n Ontaro Provncal Electons. Poltcs & Polcy, 36(3): 448 473. Knack, Stephen, and Martha Kropf. 2003. Voded Ballots n the 1996 Presdental Electon: A County Level Analyss. Journal of Poltcs 65 (3): 881 97. McAllster, Ian, and Ton Makka. 1993. Insttutons, Socety or Protest? Explanng Invald Votes n Australan Electons. Electoral Studes 12 (1): 23 40. Mllgan, Kevn, Enrco Morett, and Phlp Oreopoulos (2004). Does Educaton Improve Ctzenshp? Evdence from the Unted States and the Unted Kngdom, Journal of Publc Economcs, 88:1667-95. 18
Obradovć-Wochnk, Jelena, and alexander Wochnk. 2014. East European Poltcs and Socetes and Cultures 28 (4): 808 835. Papke, Lesle, and Jeffrey Wooldrdge. 1996. Econometrc methods for fractonal responsevarables wth an applcaton to 401 (K) plan partcpaton rates. Journal of Appled Econometrcs 11: 619-632. Power, Tmothy J., and James C. Garand. 2007. Determnants of Invald Votng n Latn Amerca. Electoral Studes 26 (2): 432 44. Rker, Wllam, and Peter Ordershook. 1968. A Theory of the Calculus of Votng. Amercan Poltcal Scence Revew 62:25-42. Snclar, D. E. Betsy, and R. Mchael Alvarez. 2004. Who Overvotes, Who Undervotes, Usng Punchcards: Evdence from Los Angeles County. Poltcal Research Quarterly 57 (1): 15 25. Stefbold, Rodney P. 1965. The Sgnfcance of Vod Ballots n West German Electons. Amercan Poltcal Scence Revew 59 (2): 391 407. Soto, Irvn M., and Wlly W. Cortez. 2014. Determnates de la partcpacón electoral en Méxco. Estudos Socologcos 32(2):323-353. Uggla, Fredrk. 2008. Incompetence, Alenaton, or Calculaton: Explanng Levels of Invald Ballots and Extra-Parlamentary Votes. Comparatve Poltcal Studes 41 (8): 1141 64. 19