A Water Cooler Theory of Poltcal Knowledge and Votng Torben Iversen Department of Government Harvard Unversty Davd Soskce Department of Poltcal Scence Duke Unversty [Frst prelmnary draft] Prepared for presentaton at the Annual Meetngs of the Amercan Poltcal Scence Assocaton, Boston, August 2831, 2008.
1 Abstract Snce ndvduals cannot affect publc polces they have no ncentves to nvest n costly poltcal knowledge for that that reason, and the assumpton n standard poltcal economy models that people know how publc polces affect ther economc nterests s consequently problematc. We argue n ths paper that the ncentve to acqure poltcal knowledge s a byproduct of other ncentves, both prvate socal, and that these ncentves are unequally dstrbuted across groups n the electorate. In our amended poltcal economy model of votng, the process of poltcal knowledge acquston s endogenous and we show how ths has notable mplcatons our understandng of dstrbutve poltcs across democraces. We test the model on publc opnon data from 18 advanced democraces and show that the couplng between economc nterests and poltcal choce s much stronger n some economc classes, and n some poltcal systems, than n others. 1
2 1. Introducton The lack of poltcal nformaton among ordnary voters has been thoroughly documented snce the publcaton of The Amercan Voter n 1960 (Campbell et al., 1960; Lews Beck et al. 2008), and the problem s succnctly captured by Downs concept of ratonal gnorance (Downs 1957). Yet modern poltcal economy largely gnores the problem and assumes that people are wellnformed about ther nterests and how publc polces affect them. In ths paper we argue that the acquston of poltcal knowledge s not only problematc, but lkely to be based n a manner that has mportant mplcatons for our understandng of dstrbutve poltcs. We propose an amended poltcal economy model of poltcal choce, whch endogenzes the process of poltcal knowledge acquston. We then test ths model on survey data for 18 advanced democraces. The problem of poltcal knowledge would perhaps not be great f errors n poltcal understandng were randomly dstrbuted snce they would tend to be cancelled out through the poltcal process of preference aggregaton. In that stuaton standard poltcal economy models would stll make the correct predctons to the extent that errors were not systematcally based away from the expectaton. But poltcal knowledge s almost certanly not evenly dstrbuted across groups wth dfferent nterests, so some groups wll be less lkely to vote n a matter that s consstent wth ther economc nterests. The problem s amply llustrated n a new book by Larry Bartels (Bartels 2008). Bartels shows that ordnary people routnely vote aganst ther own dstrbutve nterests and fal to punsh (mostly Republcan) governments for poor economc performance. A sold majorty of Amercans favored Bush s tax cuts, for example, even though only the wealthest Amercans benefted, and even though the same majorty beleved that the government can and should reduce nequalty. In Bartels own words, [t]he results of my analyss suggest that most Amercans support tax cuts not because they are ndfferent to economc nequalty, but because they largely fal to connect nequalty and publc polcy (p.??). The concluson Bartels draws s aptly captured by the mage of the delusonal and easly duped workng class cartoon fgure, Homer Smpson, who graces the front page of one of the chapters. As Bartels notes about hs evdence: [T]here s plenty of raw materal here for an account of contemporary poltcs emphaszng ordnary Amercans msperceptons, myopa, and mssng connectons between values and nterests on one hand and polcy preferences and votes on the other (p. 295). And Bartels observatons mply a dstnct partsan bas n the poltcal consequences of gnorance. Even though Democratc governments tend to mprove the lot of a majorty of Amercans, Republcans end up wnnng most of the tme. Ah, f only Homer would get t rght! But Bartels analyss also shows that some voters are better nformed than others and much more lkely to vote for polces that are n lne wth ther economc nterests. Alverez and Brehm (2002) also show that people who are nformed about poltcs vote dfferently than voters who are not, and Duch et al. (2000) argue that the bas n people s assessments of the economy are correlated wth nformaton. What we need, therefore, s 2
3 a theory of mass poltcs that endogenzes the acquston of poltcal knowledge. We buld such a theory on the assumpton that people acqure poltcal nformaton when, and only when, t makes ndvdual sense for them to do so. We argue that there are two such motvatons, both entrely unrelated to any desre to affect publc polces. One s the ncentve that some people have to understand publc polces and regulatons because ths enables them to make better prvate economc decsons or because they need to understand these polces and regulatons as part of ther jobs. We argue that these ncentves tend to be concentrated among those wth hgh ncomes, sgnfcant property holdngs, or n manageral or supervsory postons. For those who do not have strong prvate ncentves to acqure knowledge, the only motvaton to acqure knowledge s socal. Mountng evdence n socal psychology shows that human behavor s n large part motvated by the desre to gan the recognton and respect of others. Ths desre, we argue, also motvates people to acqure knowledge about poltcs when ths s a recurrent topc of dscusson around the water cooler. Of course, poltcs s often not the focal pont for dscusson n groups, but when t s people have a strong socal ncentve to know at least somethng about poltcs n order to contrbute productvely to the dscusson. The more others n the group know, the greater the ncentve to be knowledgeable yourself. Of course, much of the fodder for poltcal dscusson s comes from the publc spectacle of poltcal eltes competng for the support of potental voters (especally around electons). Indeed, eltes have an ncentve to cultvate poltcal dscusson around topcs that they help to frame. If ths leads to clear dfferentaton of polces and focus on party programs as opposed to ndvdual canddates, the process of translatng poltcal dscusson nto an understandng of that relatonshp between publc polces and economc nterests s facltated. In turn, polcy dfferentaton and programmatc orentaton dffer systematcally across poltcal systems for wellunderstood nsttutonal reasons, and ths helps explan crossnatonal dfferences n dstrbutve poltcs n partcular, we argue, why there appears to be a tghter lnkage between ndvdual economc condtons and left party vote n some countres than n others. 2. A water cooler theory of poltcal knowledge 2.1. The mcrolevel logc Because most publc polces are collectve goods, ndvduals cannot nfluence them through votng and they consequently have no reason to acqure costly poltcal nformaton for that reason. We nstead propose two classes of ndvdually ratonal ncentves: one s strategc and socally motvated; the other s nonstrategc and prvately motvated. Prvately motvated knowledge acquston happens when publc polces affect whch economc choces are the most optmal for ndvduals. The decsons of busness people, for example, depend on regulatons and taxes n ways that make t advantageous to understand these. Ths s also true for managers and upperlevel 3
4 professonals who are often charged wth applyng publc regulatons and polces n ther capactes as supervsors and admnstrators (thnk of regulaton of employment, wages, socal benefts, and so on). A smlar logc apples to economc decsons such as buyng a house where one needs to know rules for deductng mortgage nterest payments, and how these may change or nvestng n a prvate penson scheme, an educaton, and so on. As ndvdual economc choces multply, so do the ways n whch publc polcy mpact these. In turn, knowledge acqured to make sound economc choces can n turn be used to decde whch polces and poltcal partes are most lkely to mprove one s economc stuaton. In general, we would expect busness people, people n management, and people wth hgh ncomes and sgnfcant property holdngs, to have stronger prvate ncentves than others to acqure costly nformaton about complex tax codes and regulatons. Ths does not mply that such codes and regulatons are unmportant for the welfare of lowerncome people ndeed, they underpn most efforts of redstrbuton but for most of these publc polces are largely collectve goods wth few consequences for prvate economc decsonmakng. 1 Ths classbas n the economc ncentve to acqure poltcal nformaton may be magnfed by educaton snce t s easer to make sense of sometmes complcated publc polces for people wth hgh educaton, and hence cheaper for these ndvduals to acqure the nformaton. For ths reason socal ncentves for acqurng poltcal nformaton are probably partcularly mportant at the lower end of the ncome dstrbuton. A good startng pont for understandng these ncentves s what Verba, n work wth Brady and Schlozman, calls cvc sklls (Verba, Schlozman and Brady 1995). Although they use cvc sklls as an explanatory varable, they also suggest that such sklls are acqured through nteracton nsde nformal groups and networks, ncludng the famly, the neghborhood, churches, and the workplace. In a recent paper (Abrams, Iversen, and Soskce 2007) we argue that the key for understandng the poltcal role of these nformal networks s the desre of human bengs to belong to groups and to be valued by ther peers (Brennan and Pettt 2004). The desre to earn the respect and esteem of those around us can be a powerful ncentve to act n ways that at frst blush appear unselfsh, even rratonal. In partcular, we argue that many people learn about poltcs because they want to mantan or mprove ther standng n groups where poltcs s a central topc of conversaton. When poltcal knowledge becomes a marker for group standng, the desre for approval can turn nto an engne of poltcal knowledge acquston. Verba, Ne and Km long ago observed that ndvdual resources wll be harnessed for poltcal purposes prmarly when people wth such resources come nto contact wth others who are poltcally actve, and they specfcally ponted to poltcal dscusson as a catalyst for actvsm (Verba, Ne and Km 1978). There s also a long tradton n 1 There are clearly exceptons such as the earned ncome tax credt, whch s advantageous to understand n makng decsons about workng hours and so on. Our clam s smply that those wth more property and ncome tend to have stronger ndvdual ncentves to know about a broader range of publc polces 4
5 socology, begnnng wth the poneerng work of Lazersfeld and Berelson and ther assocates (Lazersfeld et al 1944; Berelson et al 1954) and echoed more recently n the hghly nfluental study by Huckfeld and Sprague (1991, 1995) and an nsghtful recent book by Walsh (2004), that places dscusson of poltcs at the center of an understandng of opnon formaton and votng. It s not hard to spot the ndvdual ncentve to acqure poltcal knowledge n the context of closeknt networks. Smply put, f poltcs s a recurrent topc of dscusson wth frends, famly and colleagues you don t want to be Homer Smpson who never knows what s gong on n the world. The same would be true f the topc s baseball, moves, or you name t. In all cases most people wll develop an nterest n, and knowledge about, the relevant topc for the sole purpose of beng able to contrbute productvely to the dscussons of the group. The topcs groups talk about are not randomly dstrbuted, although t may often appear that way to ndvduals. Dscusson of poltcs s more lkely to emerge n settngs wth repeated nteracton between ndvduals who can be assumed to have wellalgned nterests. When people of the same socoeconomc class nteract repeatedly t s sensble to expect that ssues of publc polcy wll enter the range of topcs that are broached. As we argue n more detal below ths tendency s renforced by poltcal eltes wth an nterest n usng ther local presence to try to shape the content of group dscusson through facetoface nteracton. Unlke poltcal nformaton that s acqured for ndvdual economc gan, poltcal partes can help shape peoples understandng of how ther own nterests are lnked to publc polces. We focus on three specfc socal mechansms that shape ndvduals noneconomc ncentves to acqure poltcal knowledge or to be able to make use of such knowledge. One can be called a varance reducton mechansm and occurs from the smple sharng of nformaton. Indvduals may have some dea about ther nterests and how publc polces affect them, but these are prone to error. For example, a low ncome person may see an nterest n redstrbuton but mstakenly conclude that a tax cut s lkely to further that nterest (as Bartels documents). If such errors are randomly dstrbuted around the true effect, they wll tend to cancel each other out n a group where the nterests of members are algned and where nformaton about poltcs s shared through dscusson. It s not really that people become more knowledgeable through ths mechansm, but that they are empowered to make more nformed decsons. Second, beng n socal networks ncentvzes ndvduals to acqure more nformaton about publc polces f these are recurrent topcs of dscusson. Ths s the socal esteem mechansm. There s strong evdence n socal psychology that people care a great deal about what others thnk of them and that ths s a key ndvdual motvator. As Baumester and Leary (1995) note n the hghlycted revew artcle of the lterature: Exstng evdence supports the hypothess that the need to belong s a powerful, fundamental and extremely pervasve motvaton A great deal of human behavor s caused by ths fundamental nterpersonal motvaton. We concur and argue that the 5
6 ablty to contrbute productvely to group dscusson s one of the most mportant sources of ndvdual welfare, and sometmes the focus of dscusson s poltcs. Fnally there s a network externalty mechansm that occurs because people n networks are n fact playng strategc games wth each other. An ndvdual wll acqure poltcal knowledge the hgher the level of dscusson, but dscusson s also a functon of knowledge n the group because t s more nterestng to dscuss topcs on whch there s more nformaton and because there s more to be learned from dscusson when that s the case. As we show, ths leads to unque dscussonknowledge equlbra. 2.2. The macrolevel logc Because people are not perfectly sorted nto groups wth dentcal nterests, and because poltcs, ncludng dstrbutve poltcs, s multdmensonal, arrvng at common knowledge about nterests s problematc. Nor s poltcs necessarly the focal pont for dscusson. Ths helps explan a relatvely weak relatonshp between economc poston and poltcal atttudes. Instead, poltcans, poltcal partes, and organzed groups wth a vested nterest n cultvatng common knowledge around partcular partsan deas are lkely to play an mportant role n agendasettng. Dscusson of poltcs, and common knowledge about nterests, s lkely to be partly a functon of elte moblzaton. Our startng pont here s the conjecture by Aldrch (1993) and by Rosenstone and Hansen (1993) that poltcal eltes and poltcal partes have an nterest n moblzng voters who mght share ther poltcal agenda. As argued by Uhlaner (1989), Morton (1991) and others, ths also apples to large nterest groups who stand to beneft, or suffer, from partcular publc polces. But the mcrologc spelled out above suggests a qute dfferent mechansm by whch eltes affect the behavor of ndvduals: Poltcal partes and groups can use advertsng and meda access to try to shape the ssues people pay attenton to, and they can sometmes use ther local presence to shape the agenda through facetoface contacts and by ntatng dscussons about poltcal ssues. Of course, the ntensty of such behavor wll ncrease around electons, but the hot poltcal ssues of the day can and do fnd ther way onto the agenda between electons. Contactng people n localtes that are lkely to be responsve to the poltcal message s one method. Makng use of party or group members, or sympathzers, who can take on the role of opnon leaders s another. An example of the latter s when shop stewards n companes n some countres often vewed as the most knowledgeable and respected colleagues ntate poltcal dscussons around the lunch table. The mplcaton of ths lne of reasonng s that poltcal nformaton and partcpaton wll vary wth the strength of collectve organzatons such as partes, unons, and churches. But the role of poltcans and groups s also a matter of elte ncentves, whch vary by the type of poltcal system. A key for our argument s to understand the nsttutonal condtons under whch partes are more wllng to dfferentate ther poltcal messages from other partes and target ther core consttuents. Clearly dfferentated poltcal messages help reduce the uncertanty of voters about party postons. Another salent consderaton s the ncentves partes have for focusng ther electoral appeals on 6
7 polces as opposed to ndvdual canddates. Canddate qualtes such as honesty, effectveness, and so on, are not lkely to dffer n a manner that s ted systematcally to the content of poltcal messages, and so canddateorented systems are not conducve to poltcal dscusson that clarfes the relatonshp between polces and economc nterests. In turn we know a great deal about the nsttutonal causes of polcy dfferentaton and programmatc focus. In majortaran twoparty systems, the need to convnce the medan voter that partes are credbly commtted to a centrst platform means that they cannot appear to be too beholden to the nterests of ther core consttuents (Aldrch 1993; Schlesnger 1984; Ktschelt 1994). Partes therefore cannot create strong party organzatons around extreme voters f ths s seen by the mddle class to undermne the credblty of commtments to centrst economc platforms. And precsely because the man partes do not dffer radcally on economy polcy, each party s free to use noneconomc appeals to capture voters that what would otherwse be a natural consttuency for the other party (Ktschelt and Rehm 2005). A centerrght party may appeal to workngclass voters on ssues such as relgon and crme that do not drectly engage these voters core economc nterests. By the same token, centerleft partes may appeal to hgheducated, hghncome voters on ssues such as the envronment or democratc accountablty. Partes become, n the words of Krchhemer, catchall partes. And ths means that voters across economc classes wll fnd t harder to decde whch party s more lkely to advance ther own nterests. Multparty PR s dfferent because representaton does not depend on capturng the medan voter. Partes are therefore free to moblze more narrowlydefned segments of the electorate subject to the constrant that too narrow targetng wll nvte the formaton of new partes and undermne the attractveness of the party as a potental coaltonpartner n government. Multparty PR s thus conducve to a segmentaton of the electoral market where each segment becomes exposed to relatvely coherent messages. The dstncton between platform versus canddatecentered competton s also shaped by the nature of the electoral systems. We have argued elsewhere (Iversen and Soskce 2006) that whereas partes n PR systems can represent the nterests of ther consttuents drectly, partes n majortaran systems need to reduce the role of the party organzaton and delegate more power to ther elected leaders. The mplcaton s that electons n PR partes wll tend to be ssue or platform orented, whle electons n majortaran systems wll tend to focus more attenton on the qualtes of poltcal leaders. Ths dfference s related to the dea n Carey and Shugart (1995) that some electoral systems produce canddateorented as opposed to programmatc partes. Carey and Shugart focus on the ncentves canddates n dfferent electoral systems have to cultvate a personal vote. For our set of countres, two features of electoral systems are partcularly salent. One s the sze of electoral dstrcts. At one extreme the Netherlands treats the entre country as a sngle electoral dstrct. At the other, sngle member dstrct (SMD) systems have as many dstrcts as there are members n the legslature. Dstrct sze matters to electoral strateges partes because whereas t makes good sense to feld 7
8 canddates n SMD systems who can cater effectvely to local nterests, f the electoral dstrct s the naton as a whole specalzaton of canddate appeals makes lttle sense. In turn, as the focus shfts from ndvdual canddates to party platforms, voters lose nterests n the attrbutes of the former and focus ther attenton on polces nstead. Another electoral feature that affects the extent to whch voters choose partes accordng to ndvdual canddate qualtes as opposed to party platforms s poolng of votes. If votes for a canddate that exceeds the requred number can be transferred to other canddates from the same party lst, votng for a canddate s also n part a vote for the party. Ths forces voters to pay attenton to the party label n addton to ndvdual canddates. To summarze, the stronger the nsttutonal ncentves for poltcans and partes to target partcular consttuences, and the greater ther ncentves to emphasze polces over canddates, the more poltcal dscusson wll focus on the relatonshp between economc nterests and publc polces, and the more dscusson wll be lnked to educaton. When partes need to present centrst economc polces and shun close assocaton wth extreme consttuences, and when there s strong emphass on canddates as opposed to polcy platforms, poltcal dscusson wll be less ted to economc ssues and more to canddate qualtes, and dscusson may be less dependent on educaton. 2.3. Formalzng the argument We assume that the man dmenson of poltcal competton s economc, and that there s a left and rght block of partes. Whether an ndvdual voter votes for the left, L, depends on hs economc stuaton, y, whch we assume he knows. Partes take polcy postons, y L, whch affect the economc stuaton of each voter but are only partally known and understood. If y s closer to s percepton y L than to R s poston y R, votes left. Hs knowledge of L s poston s y L and hs knowledge of R s poston s y R. Hence n Index functon formulaton 2 2 vl 1[( y yl ) ( y yr ) ] 2 2 2 2 1[ y yl 2 yly y yr 2 yry] 2 2 1[2 yr y 2 yl y yr yl ] 2 2 yr y L yl yr 12 y 1y yr y L 2 Now defne s percepton of the mdpont between the two partes as yl yr ylr, 2 so that votes L when 1 v y y L LR or n terms of probabltes where Pv ( L 1) s the probablty that votes L: 8
9 (1) Pv ( 1) Py ( y ) L LR We now ntroduce poltcal knowledge and assume that ylr, s normally dstrbuted around the true mdpont between the two partes ylr so that s drawng from ths dstrbuton generates where y y LR, LR 2 N 0, 2 and κ s some measure of s access to relevant poltcal knowledge. Thus where We now have Pv ( 1) P y y P ( y y) L LR LR N(0,1). 1 (2) P( v 1) y y y y L LR LR and ( z ) s the standard normal densty functon; note that the varance can only be dentfed up to a constant so that σ s dropped. Thus whle we wll see that there are effcency concerns whch wll need to be tackled econometrcally, the model drectly translates nto a Probt form. 2.3.1 Determnng Our argument s that s determned by three renforcng factors: (a) prvate economc ncentves to acqure poltcal knowledge; (b) utlty attached to group approval resultng from poltcal knowledge; and (c) a varance reducton mechansm resultng from group membershp. As we suggested above, one purpose of nvestng n poltcal nformaton s to be able to make better prvate economc decsons (the prvate economc ncentve); another s to gan the approval and respect of people n nformal socal networks where poltcs s a topc of dscusson (the esteem mechansm). We ncorporate these motves n the model by lettng maxmze 2 UKvK c K gdgk 2 9
10 where v s the margnal economc beneft and gd G the margnal socal beneft to of acqurng poltcal knowledge, K, where D G s the degree of poltcal dscusson n s group. There s a quadratc cost of knowledge acquston, so that the margnal cost s ck. Hence at an optmum v gdg ck v g K D c c These represent the ndvdual and socal benefts from acqurng poltcal knowledge. To see how group membershp reduces the error nvolved n votng, the varance reducton mechansm, we assume that the s poltcal knowledge by tself, K, mples κ = K so 2 2 that the varance of ε s equal to / K. But as a member of a group, has access to the 1 estmates of each of the others, and uses the group average G ng as the new G estmate. Ths mples that, f the ndvdual estmates are randomly drawn, the varance of ε G s var G n G 2 K G If we assume there s an error term n the K equaton so that G 2 v g K DG u c c v g KG DG c c Hence we can rewrte (2) Pv ( L 1) ylr y n K y y 1/2 G G LR v g n D y y c c 1/2 G G LR Fnally t may be useful to analyze the determnants of D G, although they wll not be used drectly n the emprcal secton. A smple way to close the model s to assume that group dscusson s proportonal to group knowledge, smply: 10
11 D G G K v g DG ng ng DG c c v ng c v DG g c 1 ng g c n Hence D G decreases n c (the margnal cost of poltcal knowledge), and ncreases n g (the mportance of esteem), n G (the sze of the group) and v ( the margnal economc beneft of poltcal knowledge). Groups are economcally homoflc members of an nformal socal network group n whch n effect members choose each other are lkely to have smlar economc characterstcs. So snce v (and probably also N G ) s lkely to be postvely related to an ndvdual s economc characterstcs, and snce v s lkely to be broadly smlar wthn nformal groups, welltodo groups are lkely to have more poltcal dscusson than poorer groups. 2.3.2 Country varatons n the left/rght dvde. Our ntenton s to explan left vote across countres, whch ntroduces the complcaton that whle ylr s a constant for all voters n a partcular country, ylr wll be dfferent across countres (for example lower n Swtzerland than n Sweden). If s a member of J() J country J, let ylr 1 and otherwse zero; and defne ylr JCDJ where CD J s the country dummy for country J (droppng the ndcator). So we can wrte N N J LR LR J J J J y y CD And as has been argued above y depends on varous economc factors, ncome, skll specfcty, prvate/publc sector, etc, whch we can refer to as x, we have y K x j j j G Snce s economc propensty to vote left (EPVL ) ncreases the greater s ylr above y we have N K EPVL ylr y JCDJ jxj J j hence the model we estmate takes the Probt form 11
12 (3) Pv n D y y 1/2 ( L 1) G ( G() ) LR 1/2 ng ( DG() ) EPVL n D CD x N K 1/2 G ( G() ) J J j j J j 2.4. The role of electoral systems. As noted n the macro dscusson n secton 2 t seems plausble that electoral systems set up dfferent ncentves for poltcans to mpact on poltcal knowledge. In partcular t mght be thought that under PR systems more mportance would attach to poltcal knowledge as opposed to personaltes. A very smple way to addng ths to the model s to assume that knowledge s greater wthn PR systems so that N K Pv n D El CD x (5) ( 1) 1/2 ( ()) L G G J J j j J j 3. An emprcal test We use data from the 2004 ISSP survey of ctzenshp to test our argument. The survey ncludes 18 establshed democraces for whch we have complete data on the key varables. 2 The man model to be estmated s Equaton (5), and true to the EITM standard we do so drectly usng a combnaton nonlnear least square regresson and multlevel Probt for survey data. The only smplfyng assumpton s that the effect of group sze s captured by the extent of group dscusson. In addton to (5), assumng all the economc varables have the correct (negatve) sgn, we can also estmate the smplfed model: (6) ( 1) 1/2 ( ()) P vl n G DG El EPVL The reason we cannot use lnear regresson to estmate ths model s that the model mples a partcular (nonlnear) relatonshp between the coeffcents on the economc varables and poltcal dscusson (and PR). The NL procedure n Stata enables the estmaton of such nonlneartes, but t does not allow Probt estmaton. We therefore use a twostep procedure where we frst estmate the parameters on poltcal dscusson and the economc varables n NL, and then reestmate these n Probt through an teratve routne that yelds the correct nonlnear Probt estmates. We explan the exact procedure n Appendx A. The regressons we end up estmatng are exactly those n (5) and (6). 2 The countres are Austra, Australa, Brtan, Canada, Denmark, Fnland, France, Germany (wth separate surveys for east and west Germany), Ireland, Japan, the Netherlands, New Zealand, Norway, Portugal, Span, Sweden, Swtzerland, and the US. 12
13 We use fve economc varables to capture the economc propensty to vote left, EPVL: Famly ncome, prvate sector employment, skll portablty, selfemployment, and supervsory poston. 3 All can be reasonably expected to be negatvely related to support for left partes, our dependent varable, based on standard poltcal economy arguments about redstrbuton (eg. Meltzer and Rchard 1981), sector and occupaton (eg. Ktschelt 1984; Garrett and Way 1999), class (eg. Ercson and Goldthorpe 1992; Evans 1993), and nsurance (eg. Iversen and Soskce 2001 and Rehm 2008). Except for the logged ncome varable (whch ranges between 0 and about 14), all economc varables are scaled so that they range between 0 and 1 (n the case of the dummy varables, these are the only possble values). Hence: Hypothess (): δ j < 0, where j = (ncome, prvate sector, portable sklls, selfemployment, supervsory poston) When economc condtons predspose an ndvdual to vote left (EPVL) that s, when ncome s low, employment s n the publc sector, sklls are specfc, and so on poltcal dscusson s expected to ncrease the probablty of votng left. Ths mples that the nteracton of poltcal dscusson and the country fxed effects s postve for all countres, assumng the economc varables are zero (or very low): Hypothess (): γ j > 0, for all j = (Australa, Austra, Brtan,, US) As economc condtons become more favorable for a rght vote, the effect of poltcal dscusson should dmnsh and potentally become negatve. Ths mples that the followng nequalty must hold f all varables are set to ther observed maxmum: Hypothess (): n 1 * K j j 0 j Fnally, wth respect to poltcal nsttutons, the argument mples that PR, as well as electoral systems that emphasze party programs as opposed to ndvdual canddates, strengthens the assocaton between economc nterests and left vote. Specfcally, the estmated parameter, η, n Equaton (4) must be postve: Hypothess (v): η > 0 We have two measures of the electoral system. One smply assgns the code 1 to countres wth more or less pure PR electoral rules and 0 to all other countres (ncludng those wth mxed systems). The other s based on the dstncton between canddatecentered versus partycentered electoral systems and s a mean of the two prevously dscussed measures: the sze of electoral dstrcts (standardzed by dvdng by the 3 Famly ncome s n US dollars usng 2004 PPP exchange rates. Portablty of sklls s the n nverse of the skll specfcty measure (s_1) used n Iversen and Soskce (2001) and avalable at www.people.fas.harvard.edu/~versen/data. Indvduals are coded to be n prvate sector employment f they work for a prvate frm or are selfemployed (coded 1; otherwse 0). Supervsors are those who say they supervse others at work (coded 1; otherwse 0) 13
14 number of seats n the natonal assembly), and the poolng of votes across canddates (coded 1 f votes are pooled across all canddates n a dstrct, 0 f no poolng s allowed, and ½ f poolng s across subsets of canddates). 4 For both measures, f there s more than one ter n the electoral system, we use an average across ters weghted by the share of seats elected from each ter. Poltcal dscusson s based on a survey queston that specfcally asks respondents to assess the frequency of dscusson n nformal groups: When you get together wth your frends, relatves or fellow workers how often do you dscuss poltcs? The respondent could answer never (coded 1), rarely (1), sometmes (2), and often (3). The percentage dstrbuton of answers across the four values s 14, 31, 42, and 13. The dependent varable s a dummy ndcatng whether an ndvdual voter s afflated wth, or voted for, a leftofcenter poltcal party. In the case of the US, respondents are also asked about the strength of ther dentty wth a party. The dvson of partes nto left, center, and rght s based on the ISSP s classfcaton, whch s derved from the natonal party codes. 5 3.1. Results Table 1 shows the results of estmatng Equaton (4) wth and wthout the ndcators for electoral system. Model (1) shows that those wth hgh ncome, n prvate employment, n supervsory postons or selfemployed, and wth sklls that are portable have a very small probablty of votng left (no more than half a percent or so). As the personal economc stuaton becomes more proptous to a left vote (when ncome s low, employment s n the publc sector, and so on), the probablty of votng left rses. Ths pattern s predcted by standard poltcal economy models and confrms Hypothess (). But as we have argued poltcal dscusson condtons these relatonshps n mportant ways. 4 Ths follows Carey and Shugart (1995), and the mplementaton of ther codng scheme by Johnson and Wallack (2007). 5 Ths s varable 258 n the ISSP data set. 14
15 Table 1. The probablty of votng left as a functon of economc varables, poltcal dscusson, and electoral nsttutons (tscores n parentheses) Poltcal dscusson x [] 1) Programmatc partes x [] PR x [] (1) (2) (3) (4) (5) (6) 1.331 0.300 (0.489) (0.041) 0.346 (0.050) 0.245 4) (0.024) 0.371 (0.188) [Income Portablty of sklls Prvate sector employment Supervsor Selfemployed Mean country dummy 2) ] 0.012 (0.004) 0.045 (0.009) 0.046 (0.005) 0.030 (0.005) 0.053 (0.008) 0.435 (0.046) 0.009 (0.002) 0.029 (0.006) 0.030 (0.003) 0.020 (0.003) 0.036 (0.005) 0.298 (0.031) 0.010 (0.003) 0.037 (0.007) 0.039 (0.004) 0.025 (0.004) 0.046 (0.006) 0.612 (0.053) EPVL (econ. propensty to vote left) 3) 3.823 (0.150) 0.934 (0.188) 0.698 (0.163) EPVL x Poltcal Dscusson 0.621 (0.045) 0.869 (0.064) 0.364 (0.026) EPVL x Electoral System (programmatc) 4.632 (0.245) EPVL x Electoral System (PR) 0.678 (0.048) Constant Pseudo Rsquared N Number of countres 1.808 (0.104) 0.088 15809 18 1.743 (0.038) 0.088 15809 18 1.737 (0.100) 0.088 15809 18 1.746 (0.038) 0.088 15809 18 3.200 (0.210) 0.088 15809 18 3.188 (0.080) 0.088 15809 18 Notes: 1) The varables n brackets are nonlnearly nteracted wth 1 plus the varables outsde brackets accordng to model?? n the text; 2) The effects of ndvdual country dummes are not shown. They vary between.18 (Japan) and.55 (France); 3) The EPVL varable s created as a lnear combnaton of all the economc varables plus the country dummes. It ndcates whether an ndvdual s to the left or rght of the poltcal center, and far away. 4) All parameters are sgnfcant at a 0.01 level or better. 15
16 Consstent wth Hypothess (), among the natural consttuents of the left those wth low ncome, publc sector employment, and so on the probablty of votng left rses notably wth poltcal dscusson. An average voter to the left of the poltcal center has a slghtly greater than 20 percent probablty of votng for a left party when he or she does not engage n poltcal dscusson, but an almost 40 percent probablty of dong so when he or she engages n frequent dscusson. Apparently socal ncentves are essental n causng lowncome voters to vote ther economc nterests. Among those wth hgh ncome and other economc condtons pontng away from the left, dscusson matters much less (Hypothess ). These ndvdual apparently have lttle dffculty recognzng that ther nterests are not well served by the left, and they vote accordngly. We attrbute ths fact to the prvate economc ncentves people wth hgh ncome, property ownershp, and manageral responsbltes have n understandng the effects of fscal and regulatory polces on ther own economc choces. Such knowledge can be used to make poltcal choces even n the absence of strong socal ncentves. Ths central fndng can be presented n a very parsmonous way usng Equaton (6). We frst create a varable for the economc propensty to vote left EPVL n the theoretcal model based on the estmated parameters for the economc varables and the country dummes, assumng poltcal dscusson s zero. The EPVL varable s postve f an ndvdual s the left of the natonal poltcal center, and t s negatve f an ndvdual s to the rght of the center. We then nteract the varable wth poltcal dscusson (or electoral system) so we capture the condtonng effects of dscusson (or electoral system) on the economc propensty to vote left. The results are llustrated n Fgure 1. The fgure shows that economc condtons clearly matter for how people vote, as expected by standard poltcal economy models (the lower black lne). When there s no dscusson of poltcs, the estmated probablty of votng left ncreases from a lttle above zero to about 40 percent as we go from those wth the lowest to those wth the hghest economc propensty to vote left. In the presence of poltcal dscusson, however, ths effect s notably larger (the upper black lne). In average, the probablty of votng left when objectve economc condtons ( nterests ) are proptous for such a vote (hgh EPVL) s nearly 80 percent. Poltcal dscusson s thus a crtcal factor n lnkng objectve economc condtons to poltcal choce, especally among the natural consttuents of the poltcal left. On the bass of these results t s not surprsng that left partes have always been more proactve than rght partes n establshng organzatons of mass moblzaton and organzng socal lfe of ther wouldbe supporters around poltcs. Yes, the ncentves of eltes to do so vary across countres accordng to wellunderstood macronsttutonal ncentves. Although much research needs to be done to dentfy the mechansms lnkng these ncentves to poltcal dscusson n socal groups theoretcally as well as emprcally we have focused on what would appear to be an mportant nsttuton n shapng elte behavor: electoral systems. Models (3) (6) of Table 1 test the mportance of ths varable, usng dfferent measures of electoral systems. 16
17 Fgure 1. Economc propensty to vote left and actual probablty to vote left, by level of dscusson and electoral system Probablty of votng left 0.2.4.6.8.2 0.2.4 Economc Propensty to Vote Left (EPVL) The postve effects of PR and programmatc party systems confrm hypothess (v), and s llustrated n Fgure 1. Electoral systems that encourage programmatc partes produce a notably stronger relatonshp between economc condtons and left vote, especally when dscusson s hgh (the red lne n Fgure 1). By contrast, electoral systems that encourage canddatecentered partes tend to be assocated wth a weaker relatonshp between economc condtons and left vote, especally when dscusson s low (the blue lne). The effect of PR s smlar. Ths suggests to us that the structure of party competton n PR systems wth strong programmatc partes generates messages, and focal ponts for poltcal dscusson, that are more closely ted to underlyng economc nterests as compared to majortaran, canddatecentered systems. There s however a countervalng effect that s not captured by these results. If we use poltcal dscusson as a dependent varable, t turns out that dscusson s more closely assocated wth hgh educaton n PR systems than n majortaran and canddatecentered systems (the results are shown n Appendx B). We speculate that ths s because the latter focus voter attenton on canddate qualtes nstead of party programs, and that dscusson of personal qualtes (honesty, effectveness, and so on) depends much less on educaton than s true for dscusson of the economc consequences of publc polces. Majortaran, canddatecentered systems, t would appear, generate a comparatvely hgh level of dscusson among the natural consttuences of the left, but much of ths 17
18 dscusson s nosy n terms of establshng a clear lnkage between economc nterests and party choce. 4. Concluson We have argued n ths paper that standard ratonal choce poltcal economy models of mass poltcs need to be amended to take account of the process by whch people accumulate nformaton about publc polces and how these affect ther economc nterests. Ratonal choce theory tself mples that poltcal knowledge acquston s problematc and gnorng ths nsght reduces the explanatory power of standard models. We have argued that there are prvate ncentves to nvest n specfc types of poltcal nformaton, especally among the betteroff, but much nformaton s acqured for socal reasons n order to partcpate n group dscusson and earn the recognton and respect of others. Especally at the lower end of the ncome dstrbuton, people have few reasons to understand poltcs except f such understandng affects ther standng n the socal socal groups to whch they belong. The fact that socal ncentves for poltcal knowledge acquston are partcularly mportant at the lower end of the ncome dstrbuton means that the poltcal role of socal groups s mportant for dstrbutve poltcs. Knowledge s power, even f people do not acqure knowledge to become powerful. Ths shows up clearly n our results because among those who dscuss poltcs frequently, and whose economc crcumstances make them natural consttuents for the left, the probablty of votng for the left s twce as hgh. Among those whose economc crcumstances predspose them to vote for the rght also tend to do so, regardless of the level of poltcal dscusson. But these ndvduals often have prvate ncentves to acqure nformaton about polces (especally the tax code), and they often have the educaton to make the acquston of poltcal nformaton less costly. Understandng the level and dstrbuton of poltcal dscusson across groups s thus an mportant task for future research. We have provded a framework through whch such research can be done. But the task s not smply to explan the level of dscusson, but also the content. Assumng that much poltcal dscusson s nduced by poltcal eltes, we have suggested one macronsttutonal mechansm that may help explan the dstrbuton and content of poltcal knowledge: electoral systems. Because poltcal partes under PR can target the economc concerns of ther core consttuents wthout the fear of losng representaton, poltcal dscusson s lkely produce a better ft between votng and ndvdual economc nterests (measured by standard economc varables). Because partes n majortaran systems need to present themselves as centrst, they may send less clear sgnals to ther consttuents, and open these up to moblzaton on other dmensons. The result s lkely to be poltcal dscusson that s less clearly focused on economc nterests. Another factor affectng the structure of poltcal dscusson s what Carey and Shugart have called the ncentve to cultvate a personal vote. In canddatecentered systems dscusson of poltcs tends to become focused on personal qualtes as opposed to 18
19 polces, whereas n partycentered systems the reverse s true. As n the case of PR ths shows up as a stronger effect of dscusson n partyorented systems. Interestngly enough, however, canddateorented systems appear to be more conducve to dscusson among loweducated people, perhaps because dscusson of canddate qualtes requres less understandng of causal relatonshps than publc polces. The concluson seems to be that majortaran and canddateorented systems are conducve to poltcal dscusson among the natural consttuences of the left, but such dscusson s less ted to economc crcumstances. Needless to say these conjectures have to be corroborated through future research. 19
20 Appendx A: Dervng nonlnear Probt estmates Stata only has a nonlnear least squares estmaton program, so ths doesn t work for the probt form. However we can use nonlnear OLS to approxmate the parameters on the dfferent varables and then use an teratve reestmaton procedure n Probt to get the correct (probt) estmates: (1) Assume for smplcty that there s only one country and two economc varables (the method generalzes straghtforwardly to more countres and varables). The model s now Pv ( L 1) ( D )( x z ) EPVL x z (2) Defne EPVL,(0) (0) (0) x (0) z, where γ (0), etc. are derved from the Stata s nonlnear (NL) OLS procedure. (3) Run Probt wth 2 explanatory varables EPVL,(0), D* EPVL,(0). Recuperate estmates (1), (1) from Probt estmaton and form,(1) (1) (1) D (4) Run Probt wth 3 explanatory varables,(1),,(1) * x,,(1) * z Recuperate estmates (2), (2), (2) from Probt estmaton and form EPVL x z,(2) (2) (2) (2) (5) Run Probt wth 2 explanatory varables EPVL,(2), D* EPVL,(2). Recuperate estmates (3), (3) from Probt estmaton and form,(3) (3) (3) D (6) Repeat untl convergence s acheved (n practce about 34 teratons). Ths procedure produces correct nonlnear Probt estmates, but n prncple there s a more drect method, whch s to dfferentate the log lkelhood by each of the 5 parameters. The log lkelhood s log L log ( D)( x z ) log 1 ( D)( x z ) y 1 y 1 so the 5 estmatng equatons take the form: log L log L log L 0 0 0 log L log L 0 0 We do not have a program to estmate these. 20
21 Appendx B: The determnants of poltcal dscusson The Probt regresson uses a 01 verson of the poltcal dscusson varable, where 1 ndcates a hgh level of dscusson. The model s fxed effects (so PR cannot be entered separately), but the results of the country dummes are omtted. Standard errors are n parentheses. Educaton PR * Educaton Income Portablty of sklls Prvate sector employment Supervsor Selfemployed Number of observatons Number of countres Dependent varable: Poltcal dscusson 0.08*** (0.01) 0.10*** (0.03) 0.14*** (0.04) 0.02 (0.02) 0.10*** (0.04) 0.18*** (0.03) 0.18*** (0.04) 15804 18 21
22 Bblography Abrams, Iversen, and Soskce 2007. Ratonal Votng wth Socally Embedded Indvduals. Typescrpt, Department of Government, Harvard Unversty. Aldrch, John. 1993. Ratonal Choce and Turnout. Amercan Journal of Poltcal Scence. 37: 24678. Alvarez, Mchael, and John Rehm. 2002. Hard Choces, Easy Answers: Values, Informaton, and Amercan Publc Opnon. Prnceton: Prnceton Unversty Press. Bartels, Larry. 2008. Unequal Democracy: The Poltcal Economy of the New Glded Age. Prnceton: Prnceton Unversty Press. Baumester, R.F., and M.R. Leary. 1995. The Need to Belong: Desre for Interpersonal Attachments as a Fundamental Human Motvaton. Psychologcal Bulletn, 117: 497 529. Berelson, Bernard R., Paul F. Lazarsfeld, and Wllam N. McPhee. 1954. Votng. Chcago: Unversty of Chcago Press. Brennan, Geoffrey, and Phllp Pettt. 2004. The Economy of Esteem: an Essay on Cvl and Poltcal Socety. Oxford: Oxford Unversty Press. Campbell, Angus, Phlp E. Converse, Warren E. Mller, and Donald E. Stokes. 1960. The Amercan Voter. Unversty of Chcago Press. Carey John M. and M. S. Shugart. 1995. Incentves to Cultvate a Personal Vote: A Rank Orderng of Electoral Formulas Electoral Studes 14 (4) 417439. Cusack, Thomas, Torben Iversen, and Phlpp Rehm. 2006. Rsk at Work: The Demand and Supply of Government Redstrbuton Oxford Revew of Economc Polcy 2006, pp. 36589. Duch, Raymond M., Harvey D. Palmer, and Chrstopher J. Anderson. 2000. Heterogenety n Perceptons of Natonal Economc Condtons. Amercan Poltcal Scence Revew 44 (4), 63552. Erkson, Robert and John H. Goldthorpe. 1992. The Constant Flux: A Study of Class Moblty n Industral Socetes. Oxford: Oxford Unversty Press Evans Geoffrey. 1993. Class, Prospects and the Lfecycle: Explanng the Assocaton Between Class Poston and Poltcal Preferences. Acta Socologca. 36:263 76. Evans Geoffrey, ed. 1999. The End of Class Poltcs? Class Votng n Comparatve Context. Oxford, UK: Oxford Unversty Press. 22
23 Evans Geoffrey. 2000. The Contnued Sgnfcance of Class Votng. Annual Revew of Poltcal Scence 3: 40117. Garrett, Geoffrey, Chrs Way. 1999. Publc sector unons, corporatsm, and macroeconomc performance. Comparatve Poltcal Studes 32 (4): 41134. Grofman, Bernard. 1993. "Is Turnout the Paradox That Ate Ratonal Choce Theory?" In Bernhard Grofman (ed.), Informaton, Partcpaton, and Choce: An Economc Theory of Democracy n Perspectve. Ann Arbor: Unversty of Mchgan. Huckfeldt, Robert, and John Sprague. 1991. Dscussant Effects on Vote Choce: Intmacy, Structure, and Interdependence. Journal of Poltcs 53:12258. Huckfeldt, Robert and John Sprague. 1995. Ctzens, Poltcs, and Socal Communcaton: Informaton and Influence n an Electon Campagn. Cambrdge Unversty Press Iversen, Torben, and Davd Soskce. 2001. An Asset Theory of Socal Polcy Preferences Amercan Poltcal Scence Revew, December 2001, pp. 87593. Iversen, Torben, and Davd Soskce. 2006. Electoral Insttutons and the Poltcs of Coaltons: Why Some Democraces Redstrbute More than Others. Amercan Poltcal Scence Revew 100 (2), May 2006, pp. 16582. Johnson JW, Wallack JS. 2007. Electoral systems and the personal vote. http:dss.ucsd.edu~jwjohnsoespv.htm Ktschelt, Herbert. 1994. The Transformaton of European Socal Democracy. Cambrdge: Cambrdge Unversty Press. Ktschelt, Herbert, and Phlpp Rehm. 2005. "Who Supports the Left n the 21th Century? Patterns of Party Competton and Electoral Coaltons." Presented at the Conference "Socal Justce and the Future of European Democracy" at the Mnda de Gunzburg Center for European Studes. Cambrdge (MA), March, 2005. Lazersfeld, Paul F., Bernard Berelson, and Hazel Gaudet. 1944. The People s Choce. New York: Duell, Sloan, and Pearce. LewsBeck, Mchael, Helmut Norpoth, Wllam Jacoby, and Herbert Wesberg. 2008. The Amercan Voter Revsted. Mchgan Unversty Press. Morton, Rebecca B. 1991. Groups n Ratonal Turnout Models. Amercan Journal of Poltcal Scence. Vol. 35, No. 3: 758776. Rehm, Phlpp. Forthcomng. "Rsks and Redstrbuton. An IndvdualLevel Analyss." Comparatve Poltcal Studes. 23
24 Rosenstone, Steven J. and John Mark Hansen. 1993. Moblzaton, Partcpaton and Democracy n Amerca. New York : Macmllan Publshng Company. Schlesnger, James. 1984. "On the Theory of Party Organzaton." Journal of Poltcs, 46: 369400. Uhlaner, Carole J. 1989. Ratonal Turnout: The Neglected Role of Groups. Amercan Journal of Poltcal Scence. 33: 390 422. Verba, Sdney, Norman H, Ne and JaeOn Km. 1978. Partcpaton and Poltcal Equalty: A CrossNatonal Comparson. New York: Cambrdge Unversty Press. Verba, Sdney, Henry E. Brady, and Kay Lehman Schlozman. 1995. Voce and Equalty: Cvc Voluntarsm and Amercan Poltcs. Cambrdge: Harvard Unversty Press. Walsh, Katherne C. 2004. Talkng about Poltcs: Informal Groups and Socal Identty n Amercan Lfe. Chcago: Unversty of Chcago Press. 24