Money is where the fun ends: material interests and individuals preference for direct democracy

Transcription

1 Gutenberg School of Management and Economcs & Research Unt Interdscplnary Publc Polcy Dscusson Paper Seres Money s where the fun ends: materal nterests and ndvduals preference for drect democracy Phlpp Harms, Clauda Landwehr September 27, 2018 Dscusson paper number 1815 Johannes Gutenberg Unversty Manz Gutenberg School of Management and Economcs Jakob-Welder-Weg Manz Germany

2 Contact detals Phlpp Harms Char of Internatonal Economcs Johannes Gutenberg Unversty Manz Jakob-Welder-Weg Manz Germany phlpp.harms@un-manz.de Clauda Landwehr Char of Publc Polcy Johannes Gutenberg Unversty Manz Jakob-Welder-Weg Manz Germany landwehr@poltk.un-manz.de All dscusson papers can be downloaded from

3 Money s where the fun ends: materal nterests and ndvduals preference for drect democracy by Phlpp Harms (Johannes Gutenberg Unversty Manz, Germany) and Clauda Landwehr (Johannes Gutenberg Unversty Manz, Germany) 1 Ths verson: September 27, Abstract: Are people s atttudes towards referenda as a decson-makng procedure drven by ther materal self-nterest, or do ndvduals predomnantly value drect democracy as such, regardless of the materal payoffs assocated wth antcpated polcy outcomes? To answer ths queston, we use a survey data set that offers nformaton on respondents support for referenda as a procedure to decde on tax polcy, ther general vews on drect democracy and redstrbuton, ther ncome levels, soco-economc characterstcs, and, most mportantly, ther expectaton about the majorty s support for hgher taxes. Allowng for alternatve motves to welcome or oppose drect democracy, we fnd that ncome per se does not nfluence ndvduals procedural preferences. However, f respondents expect a clear populaton majorty n favor of or aganst more redstrbuton ther atttude towards referenda as a procedure to decde on taxaton s clearly algned wth ther ncome poston. Keywords: consttutonal choce, drect democracy, redstrbuton JEL codes: D72, D78, H20 1 Emal: phlpp.harms@un-manz.de; landwehr@poltk.un-manz.de 2 We are ndebted to semnar partcpants at JGU Manz, at the EPCS 2018 conference n Rome, and at the EWPM3 workshop n Tbls. Specal thanks go to Panu Poutvaara and Zareh Asatryan for very helpful comments. Fnancal support by the Research Unt Interdscplnary Publc Polcy at JGU Manz s gratefully acknowledged. 1

4 1. Introducton There s a well-establshed lterature n poltcal economcs that analyzes the effect of nsttutons and consttutonal rules on economc and socal outcomes (see Persson and Tabelln 2003, Acemoglu et al. 2005, and Vogt 2011 for surveys). More recently, however, researchers n poltcal economcs have started to endogenze the choce of these nsttutons, explanng the emergence of dfferent rules of the game as the outcome of a struggle between dfferent nterests n a heterogeneous socety. Dverse as they may be, these contrbutons share a common logc: ndvduals prefer those procedures whch maxmze the lkelhood that the eventual poltcal-economc equlbrum furthers ther materal nterests. 3 Whle the logc underlyng these contrbutons s consstent wth economc reasonng, t does not go uncontested: n a recent contrbuton, Rodrk (2014) deplores the rather neglgent treatment of deas n the poltcal-economc analyss of polcy nnovatons, argung that much human behavor s drven by abstract deals, sacred values, or conceptons of loyalty that cannot be reduced to economc ends (Rodrk 2014:191). In a smlar sprt, poltcal scence, psychology and behavoral economcs offer a wealth of alternatve motves beyond pure materal nterest that contrbute to understandng the choce of decson-makng procedures. Gven the potental relevance of ntrnsc motves, whch evaluate procedures wth respect to ther transparency, farness, practcablty etc., but not wth respect to ther consequences for ndvdual payoffs, the mportance of nstrumental motves, emphaszed by standard economc analyss, s ultmately an emprcal queston. In ths paper, we put the dea that materal nterests are mportant n shapng ndvduals preferences over alternatve decson-makng procedures to a test: we use data from a self-desgned survey among German resdents that asks ndvduals whether they support the use of referenda as a procedure to decde on redstrbutve taxaton. Along wth the answer to ths queston, the survey provdes us wth nformaton on respondents general support for drect democracy and for government nterventons n favor of lower economc nequalty, as well as on ther ncome and other soco-economc characterstcs. Most mportantly, the survey asks partcpants whether 3 Of course, ths requres that ndvduals can clearly dentfy ther nterests,.e. that the vel of gnorance s lfted. 2

5 they expect a clear majorty of the populaton to be n favor of or aganst hgher taxes. Unlke other emprcal studes of endogenous consttutonal choce, we thus do not have to rely on a mere conjecture that ndvduals are able to correctly antcpate the poltcal-economc equlbrum emergng from alternatve consttutonal choces. Instead, these expectatons are made explct by the survey data, and we can test the smple hypothess that ndvduals wth a lower (hgher) ncome are more (less) lkely to support referenda f they expect a clear majorty to be n favor of hgher taxes. If the support (or rejecton) of drect democracy was predomnantly drven by ntrnsc motves say, the belef that referenda are an optmal procedure n terms of farness, transparency and practcablty nether ndvduals ncome nor ther expectaton about the majorty s poston should play a role for ther procedural preferences. Our emprcal results, however, support the noton that materal self-nterest does matter for ndvduals preferences over decson-makng procedures: whle ncome per se does not affect respondents support for drect democracy, ts margnal effect s negatve for those ndvduals who expect that a clear majorty of the populaton s n favor of hgher taxes. Ths result holds even f we control for other ntrnsc or nstrumental motves that may determne ndvduals procedural preferences. And t demonstrates that money s where the fun ends: however large the enthusasm for drect democracy may be on prncpal grounds, ths support s severely dampened once ndvduals expect ther net ncome to be reduced f referenda are used to decde on tax ssues. The rest of ths paper s structured as follows: Secton 2 surveys the relevant lterature, whle Secton 3 presents a smple model that motvates the subsequent emprcal analyss. Secton 4 ntroduces the structure of our survey data set. In Secton 5, we test whether the desre for redstrbutve taxaton s affected by respondents ncome. Secton 6 turns to ndvduals support for referenda as a mechansm to decde on taxes. We frst analyze the drect effect of ncome (along wth other potental determnants). In a next step, we then explctly consder respondents expectatons on the majorty s poston. Secton 7 offers a summary and some conclusons. Regresson outputs as well as detaled defntons and nformaton on our survey data are gven n the Appendx. 3

6 2. Related lterature Our study s related to several strands of lterature: the noton that ndvduals preferences over procedures reflect ther materal self-nterest s at the heart of contrbutons that nterpret the emergence of poltcal nsttutons as the result of ratonal agents antcpaton of how alternatve nsttutonal choces wll affect ndvdual (or group-specfc) benefts. In an nfluental study, Acemoglu and Robnson (2000) explan the extenson of the franchse on the bass of a cost-beneft analyss of ncumbent rulers who grant the rght of poltcal partcpaton to the broad populaton n order to reduce the threat of upheaval and revoluton. 4 Aghon et al. (2004) descrbe the optmal dstrbuton of power n a socety as reflectng a trade-off between effcent decsonmakng and the control of a potentally self-servng ruler. In ther analyss, they both derve the consttutonal desgn that s desrable behnd the vel of gnorance, and the desgn that s lkely to emerge once (wealth) heterogenety and ndvduals conflctng nterests are explctly taken nto account. Tcch and Vndgn (2009) relate the choce among alternatve democratc consttutons (majortaran vs. consensual) to the underlyng economc nequalty, whle Robnson and Torvk (2016) descrbe the emergence of presdentalsm as resultng from a struggle between groups that dffer wth respect to ther poltcal orentaton, but also wth respect to ther preferences over the provson of publc goods. Acemoglu et al. (2015) show how the evoluton of coaltons between dfferent groups n socety gves rse to changng patterns of poltcal lberalzaton and represson. Fnally, Mukand and Rodrk (2015) defne lberal democracy as an nsttutonal settng that combnes the protecton of property rghts, electoral rghts, and mnorty rghts, and show that ts emergence s rather an excepton than the rule, relyng on a specfc constellaton of group szes as well as socal and dentty cleavages. What untes the contrbutons mentoned above s the premse that ndvduals (or groups) never value specfc consttutonal choces per se, but that they favor those procedures that are most lkely to further ther own usually materal nterests. From a poltcal scence perspectve, ths logc seems surprsngly poor. For example, preferences for democratc nnovatons,.e. non-representatve forms of decson- 4 Adt and Franck (2015) offer a recent emprcal analyss that supports the preemptve democratzaton hypothess of Acemoglu and Robnson (2000). 4

7 makng such as drect democracy or delberatve ctzen forums are explaned as the result of the more general value shft towards post-materalsm (see Inglehart 1990), referrng to the dea that, n tmes of affluence, non-materal values, ncludng poltcal partcpaton, become more mportant than wealth. Accordng to ths vew, ctzens demand ever more opportuntes to have a drect say n polcy-makng, they become crtcal ctzens (Norrs 1999). A number of contrbutons that study ctzens atttudes towards specfc democratc decson-makng procedures pont out the mportance of dfferng normatve conceptons of representaton and democracy for procedural preferences (Wenzel, Bowler et al. 2000; Bengtsson and Wass 2010; Landwehr and Stener 2017). In these studes, support for specfc procedures or reforms of them s regarded as derved from a more comprehensve understandng of democracy n whch values such as equalty, autonomy, the protecton of lbertes as well as nsttutonal capacty and effectveness are weghed and combned. The dea that ndvduals assessment of alternatve decson-makng procedures does not just depend on ther materal nterests s also underlyng those contrbutons that emphasze the gans n ndvdual well-beng assocated wth the possblty of partcpatng n the poltcal process (Stutzer and Frey 2005, 2006; Pacheco and Lange 2010) as well as theoretcal and emprcal analyses of procedural farness (Tyler and Lnd 2000, Bolton et al. 2005). Moreover, t s reflected n the concept of expressve votng, whch captures the dea that votng may be motvated by concerns other than a concern for the eventual outcome of the electon concerns that are more drectly and mmedately lnked to the act of votng, or of votng for a partcular canddate or opton, tself (Hamln and Jennngs 2011: 645). Fnally, our analyss explores the role of materal nterests n determnng ndvduals support for referenda as a partcular procedure to decde on redstrbutve taxaton. It s thus also related to the volumnous lterature on the determnants and effects of drect democracy, as surveyed, e.g., by Matsusaka (2008), and to contrbutons that analyze how the use of referenda as a decson-makng procedure affects fscal outcomes lke government spendng, publc debt, and taxaton (Feld and Krchgässner 2001, Funk and Gathmann 2011, Asatryan et al. 2017a, Asatryan et al. 2017b). Even closer to our analyss are those studes that use survey or votng data to dentfy the determnants of people s support for drect democracy (Donovan and Karp 2006, Bowler et al. 2007, Dyck and Baldassare 2009, Collngwood 2012, Arnold et al. 2016). None of these contrbutons, however, shares our focus on a specfc polcy 5

8 ssue.e. redstrbutve taxaton and none of them explctly elcts ndvduals expectatons on the majorty opnon. 3. Dstrbutonal nterests, expectatons, and ndvduals support for referenda: A smple model On the followng pages, we wll develop a smple model that allows dsentanglng the varous forces whch potentally determne an ndvdual s choce between alternatve procedures. More specfcally, we wll dfferentate between ndvduals substantal polcy preferences here, on redstrbutve taxaton and ther procedural preferences, whch refer to the way the decson on a gven polcy ssue s taken. These procedural preferences can be ntrnsc,.e. the support for a gven procedure s ndependent of, and unaffected by substantal preferences, or they may be nstrumental,.e. support for a specfc procedure s condtonal on the belef that t wll brng about desred outcomes. 5 We consder an economy that s nhabted by a large number of ndvduals who are ndexed by and receve an exogenous ncome y. The cross-sectonal dstrbuton of ncomes s common knowledge, characterzed by an average ncome y and a medan ncome y, and skewed to the rght,.e. y y. The government leves a lnear ncome tax and unformly redstrbutes tax revenues among all ctzens. For smplcty, we assume that redstrbuton s restrcted to beng ether complete ( 1) or totally absent ( 0). The decson on the tax rate s ether taken through a referendum or by the government tossng a con. We assume that ndvduals preferences over decson-makng procedures have an ntrnsc component, whch assesses dfferent procedures accordng to ther nherent farness, transparency, feasblty etc., and an nstrumental component, whch reflects the utlty an ndvdual derves from the outcome she expects as resultng from a gven procedure. 6 We thus wrte ndvdual s total utlty as (1) V U 5 In Harms and Landwehr (2017), we further dscuss ths dfference and explore the mportance of ntrnsc vs. nstrumental procedural preferences wth respect to a wde array of polcy ssues. 6 As we wll argue below, ths part of total utlty does not only depend on an ndvduals materal payoff. 6

9 In (1), reflects the addtonal utlty that ndvdual derves from the use of a specfc procedure, regardless of her vew on taxaton. We assume that for a referendum and 0 for a con toss. Note that s ndvdual-specfc, and that we do not mpose any restrctons on ts sgn or absolute sze. The term U s the utlty derved from a specfc tax rate, and also depends on ndvdual characterstcs. Fnally, 0 s the relatve weght gven to nstrumental (as opposed to ntrnsc) motves. Note that, for smplcty, we assume that s dentcal across ndvduals. If nstrumental consderatons are completely rrelevant n shapng ndvduals atttudes towards referenda, we have 0. Conversely, f becomes nfntely large, procedural preferences entrely reflect nstrumental motves. The utlty an ndvdual derves from a specfc tax rate U s a lnear functon of her after-tax ncome and of a term that reflects her general atttude towards taxaton: (2) 1 U y y If s postve, ndvdual has a generally crtcal atttude towards taxaton, regardless of her own ncome poston. Ths may be because she emphaszes the potentally detrmental ncentve effects assocated wth hgher taxes. Conversely, a negatve value of reflects a generally postve atttude towards taxaton due, e.g., to nequty averson (Fehr and Schmdt 1999) that nduces an ndvdual to prefer a more even ncome dstrbuton. We assume that the dstrbuton of the parameter s not observable, and that ndvduals assgn subjectve probabltes to all potental realzatons of n the populaton. Ths results n a subjectve expected value, whch potentally dffers across ndvduals,.e. ndvduals may have dfferent expectatons on ctzens average support or rejecton of redstrbutve taxaton. For the sake of smplcty, we assume that all ndvduals agree on the fact that s not correlated wth y. Moreover, ndvdual s own general atttude towards redstrbuton, as reflected by, may, but need not concde wth. 7

10 Gven the government s choce between complete redstrbuton ( 1) and no redstrbuton at all ( 0), the utlty functon n (2) obvously mples that ndvdual prefers a tax rate of 100 percent f y y. Ths expresson has a straghtforward nterpretaton: a hgher general averson aganst taxaton ( 0) has the same effect as a hgher ncome, possbly nducng an ndvdual to reject taxaton even f her ncome s below the mean. Conversely, f an ndvdual s generally postve about redstrbuton ( 0 ), she may support a 100-percent tax although her ncome s above the mean. For a gven dstrbuton of, however, ndvduals wth a hgher ncome are more lkely to reject redstrbutve taxaton. In our emprcal analyss of secton 4, we wll start by testng ths hypothess. If a referendum s used to decde on redstrbutve taxaton, the outcome depends on the dstrbuton of ncomes and the dstrbuton of. From the perspectve of ndvdual, the expected outcome of the referendum s (3) E referendum 0 f y y 1 f y y Note that, snce s based on ndvdual s personal belefs, the expected outcome of the referendum s subjectve, too. The expresson n (3) s a slghtly modfed verson of Meltzer and Rchard s (1981) result: f ndvdual beleves that the general atttude towards taxaton s hostle on average ( 0 ), she expects the referendum to result n complete redstrbuton only f the dfference between average and medan ncome s large enough. Conversely, complete redstrbuton s perceved as a sure thng f ndvdual beleves that the skewness of the ncome dstrbuton s augmented by a postve vew on redstrbuton prevalng, on average, among the populaton ( 0 ). Obvously, all ndvduals are aware that the con toss results n 1 wth a probablty of 0.5, and n 0 wth a probablty of 0.5, hence E con toss 0.5. Confronted wth the queston whether a referendum or the con toss should be used as a procedure to decde on taxaton, the ndvdual compares the utlty specfed n (1) for the two procedures. For the ndvdual to support the referendum, the followng condton has to be satsfed: 8

11 y y y y (4) 1 referendum referendum E E, 2 where referendum E s gven by (3). Addng an error term, whch reflects random factors that determne an ndvdual s support for referenda and whch follows a cumulatve dstrbuton functon F, we can use the above expresson to derve the probablty that ndvdual opts n favor of a referendum: (5) p 1 F E referendum 0.5 y y p,.e. The condton n (5) llustrates that an ndvdual s support for referenda as a procedure to decde on redstrbutve taxaton depends on the followng factors: - The hgher the general appeal of drect democracy for ndvdual ( ), the greater the lkelhood that she chooses a referendum as a procedure to decde on taxaton for gven expectatons about the majorty s opnon and a gven relatve ncome poston. - If = 0, nstrumental motves are rrelevant, and an ndvdual s support for usng a referendum to decde on taxaton only depends on her general atttude towards drect democracy. Nether an ndvdual s ncome, nor her expectaton about the majorty s opnon or her general atttude towards redstrbuton matter. - If > 0, an ndvdual s support for usng a referendum to decde on taxaton depends on her general atttude towards drect democracy ( ), but also on her relatve ncome poston y y, her general atttude towards redstrbuton, and her expectaton about the majorty s support for complete referendum 0.5. Importantly, whether beng rch (poor) reduces (rases) the lkelhood of supportng the use of a referendum crucally depends on whether the ndvdual expects the referendum redstrbuton, as reflected by E to result n 1 or 0. Hence, f 0 the margnal effect of an ndvdual s 9

12 ncome on the lkelhood of her supportng a referendum s negatve f the ndvdual expects a majorty to support complete redstrbuton. In what follows, we wll start by analyzng the relatonshp between ndvduals ncome and ther atttude towards hgher taxes. In a next step, we wll explore whether we can reject the hypothess that materal motves are rrelevant n determnng ndvduals support for the use of referenda to decde on redstrbutve taxaton,.e. that = 0. When dong so, we can explot the fact that our data set offers nformaton on the dfferent varables that enter the expresson n (5). 4. Data To dentfy the mportance of materal nterests n shapng ndvduals atttudes towards the use of referenda, we desgned a survey experment that was felded va the GESIS panel n 2016 (GESIS, 2017). The GESIS panel s a mxed-mode access panel started n 2013, representatve of the German-speakng populaton between 18 and 70 n Germany (Bosnjak, Dannwolf et al., 2017). Snce 2013, panelsts have been partcpatng n b-monthly waves of surveys. 7 The GESIS data nclude, besdes specfc survey tems, a wde range of socodemographc questons as well as standard atttudnal constructs. The waves we draw on are wave 10 ( ce, October-December 2015) and 15 ( dd, August-October 2016). 8 The man dependent varable n our survey s the dscrete choce of the procedure referendum over alternatve procedures for a decson about redstrbutve taxaton. More specfcally, partcpants were asked the followng queston: Currently, there s a lot of dscusson about far taxaton and tax polcy. How do you thnk a decson about ths should be taken? a) After a publc debate, a referendum should be held. 7 Due to the expermental desgn of our survey, only panelsts n the onlne-access mode could partcpate. 8 Varable defntons and summary statstcs are reported n Appendx 1. 10

13 b) The Bundestag [German parlament] should decde on the bass of dscussons wthn the poltcal partes. c) An ndependent expert commsson should develop a recommendaton whch s then mplemented. d) Representatves of all affected groups should come together at a table and jontly fnd a soluton. 9 On the subsequent screen, we asked partcpants about ther own substantal atttudes on the matter: Are you n favor or aganst mplementng a hgher tax on hgh ncomes? Respondents reacted by choosng an answer on a fve-pont scale rangng from absolutely n favor (1) to absolutely aganst (5). In our subsequent regressons, we wll turn ths nformaton nto the dummy varable No_hgher_taxes, whch takes a value of one f an ndvdual stated to be aganst or absolutely aganst hgher taxes. On a thrd and fnal screen, we asked partcpants about ther assessment of the majorty opnon: Do you thnk that, n Germany, there s a majorty n favor or aganst hgher taxes on hgh ncomes? Agan, respondents could choose ther answer on a fve-pont scale, rangng from clear majorty n favor of hgher taxes (1) to clear majorty aganst hgher taxes (5). The answer to ths queston allows us to dentfy ndvduals expectatons about the outcome of a referendum: f a respondent expects a clear majorty to be n favor of hgher taxes on the rch, she or he expects these taxes to materalze f tax polcy s decded by means of a referendum. Hence, the answer to ths queston reflects the sgn of the expresson E referendum 0.5 ntroduced n Secton 3. The addtonal nformaton we use n our analyss concerns ndvduals gender, age, ctzenshp, educatonal attanment, and ncome, all of whch potentally affect ndvduals atttude towards taxaton and referenda. Based on the gender-related nformaton, we defne a dummy varable Female. The varable Age s based on the dfference between 2016 and ndvduals brth year. 10 German ctzen s a dummy 9 Ths s, of course, a much rcher menu of alternatves than the choce between a referendum and a con toss modelled n Secton 3. However, we argue that, f an ndvdual expects a clear majorty n favor of or aganst hgher taxes, she fnds the outcome of a referendum easer to predct than the outcome of any of the alternatves b) d). 10 Snce the GESIS panel uses the value of 1943 (1995) for all respondents that were born n 1943 or earler (1995 or later), we orgnally ntroduced the dummy varables Old (Young) for all ndvduals born 11

14 varable meant to control for respondents ctzenshp, Unversty entrance degree s a dummy varable that ndcates respondents educatonal attanment, reflectng an ndvdual s elgblty to enroll at a unversty or a unversty of appled scences. The ncome varables are, of course, of central relevance to our research queston: The GESIS survey records ndvduals ncome both at the personal and at the household level. For reasons detaled below, we wll focus on the nformaton gven on respondents net household ncome. 11 Income may fall nto one of nne brackets, wth the lowest bracket (1) comprsng all ncomes below 900 Euros per month, and the hghest bracket (9) comprsng all ncomes of 6000 Euros per month and above. In the subsequent analyss, we wll start by usng these categores as regressors. However, ths may be problematc, snce the dstance between the brackets boundares ncreases as we move towards hgher ncomes. As an alternatve, we therefore use the medum value wthn each bracket as a regressor. Ths leaves us wth the problem of how to treat the top bracket (net household ncomes per month above 6000 Euros). Fortunately, we can rely on the nformaton provded by German statstcal authortes (Destats 2015), whch report the share of households wthn certan ncome brackets, but also the average ncome of households wthn these brackets. Appendx 2 descrbes how we use ths nformaton n order to assgn the value of 8700 Euros per month to the top ncome bracket n the GESIS panel. Ths, however, leaves us wth the queston how much every ndvdual earner wthn a household actually brngs home: whle a net ncome of 5000 Euros n a snglencome household would put that entty among Germans top earners, the same amount earned by a two-adult household looks less mpressve. Snce GESIS provdes nformaton on household sze and on the number of chldren below 16, we adopt two approaches: The frst one defnes the number of ncome earners as the dfference between the sze of the household and the number of chldren younger than 16 years suggestng that there may be more than two ncome earners n a household. The second approach only admts sngle-earner and double-earner households, wth the n 1943 or earler (1995 or later). However, these varables were never sgnfcant, such that we eventually dropped them. We also abstaned from usng both the level and the squared value of Age snce t turned out that the latter was not sgnfcant n any specfcaton. 11 Of course, nformaton about households gross ncome would allow for an even clearer characterzaton of ther preferences towards redstrbuton. However, snce net ncome s closely related to gross ncome, we consder the former an approprate measure of ndvduals relatve ncome poston. 12

15 latter defned by a number of chldren whch s smaller than the total household sze mnus one. For both approaches, we compute a household s ncome per earner by dvdng net household ncome per month (n thousands of euros) by the number of earners. 12 Fnally, we characterzed hgh- and low-ncome households by defnng the dummy varables Household ncome above 4000 Euros and Household ncome below 1700 Euros. The boundares were defned to roughly capture the bottom and top quartles of the ncome dstrbuton n the GESIS panel. In our survey, 18 percent of all respondents fall nto the low-ncome category, whle 31 percent fall nto the hghncome category percentages that roughly concde wth the shares n the German populaton. 13 Whle the use of these dummy varables neglects the nformaton about households at the center of the ncome dstrbuton, t allows dentfyng those respondents who are lkely to be aware of ther relatve prosperty levels Household ncome and ndvduals support for redstrbutve taxaton Whle our key objectve s to dentfy the determnants of ndvduals support for referenda as a decson-makng procedure, we start by explorng whether materal nterest determnes ctzens atttude towards redstrbutve taxaton. Recall from the model n Secton 3 that an ndvdual rejects hgher taxes f y y. For a gven 12 An obvous alternatve would be to consder respondents personal ncome, as reported n the GESIS panel. However, the focus of offcal German statstcs on household ncomes makes t hard to relably defne the ncome of ndvduals n the top bracket. Moreover, ndvduals lvng n the same household are lkely to defne ther dstrbutonal nterests wth respect to household ncome rather than personal ncome. 13 Accordng to Destats (2015), 15 percent of all households n Germany report to receve an ncome below 1700 Euros per month, 48 percent report to receve an ncome above 3600 Euros, and 27 percent report an ncome above 5000 Euros. Snce the dstrbuton wthn the bracket s lkely to be skewed to the rght, we beleve that the share of households n Germany that receve a net ncome above 4000 Euros s not much hgher than the 31 percent of the GESIS panel. 14 Item cfba067a of the GESIS panel allows assessng whether respondents know ther relatve ncome poston by askng them f they consder ther fnancal wealth to be (far) below average, or (far) above average. It turns out that ths percepton s hghly correlated wth respondents reported (household) ncome level. 13

16 dstrbuton of, ths suggests that respondents wth a hgher ncome are less lkely to be enthusastc about more redstrbutve taxaton. To check ths clam, we start by runnng a regresson that uses the bnary varable No_hgher_taxes as a regressand: as descrbed n the prevous secton, ths varable takes on a value of one f respondents are aganst or strongly aganst hgher taxes on hgh ncomes, and a value of zero otherwse. As regressors, we use the dummy varable Female, respondents Age, a dummy for ther educatonal attanment (Unversty entrance degree) and, most mportantly, a varable reflectng ndvduals Income: (6) No_hgher_taxes 0 1 Income k xk In equaton (6), the vector xk reflects the control varables mentoned above. We estmate ths equaton both by usng the OLS estmator the lnear probablty model and the logt estmator. Table 1 gves coeffcents and t-statstcs (based on a robust covarance matrx) for the OLS estmates, and average margnal effects and robust t- statstcs for the logt estmates. Interestngly, Age has a consstently negatve effect on respondents opposton aganst hgher taxes. Less surprsngly, Income has a postve effect for all varants. Ths supports the clam that ndvduals are well aware of ther materal nterests when defnng ther atttude towards tax polcy. Ceters parbus, a respondent n the top-ncome category (9) s 17 percentage ponts more lkely to reject hgher taxes than a respondent n the bottom ncome category (1). The dummes for hgh- and low household ncome also have the expected effect wth hgh-ncome earners beng more lkely to reject hgher taxes and low-ncome earners beng less lkely to be aganst hgher taxes. However, the t-statstc of the low-ncome dummy narrowly msses the threshold for a 90-percent sgnfcance level. Table 1 near here The results n Table 1 confrm our conjecture that ndvduals rejecton of hgher taxes ncreases n ther ncome. However, whle the model of Secton 3 used the smplfyng assumpton that an ndvdual s general atttude towards taxaton, as reflected by the parameter, was uncorrelated wth her ncome, ths assumpton may not be satsfed n realty. More specfcally, t mght be the case that rch ndvduals hold stronger vews 14

17 about the dsncentve effects and economy-wde losses assocated wth hgh tax rates. If ths were correct, the estmated effect of ndvduals ncome on the atttude towards hgher taxes would not necessarly reflect ther materal nterests. In order to account for ths possblty, we explctly control for by usng a varable from the October- December 2015 wave of the GESIS panel,.e. the wave that took place roughly one year before ndvduals were asked about ther atttude towards hgher taxes. In that wave, respondents were asked to gve ther general opnon on the government s role n reducng economc nequaltes. More specfcally, they were confronted wth the followng statement: The government should enforce the reducton of dfferences between the poor and the rch. Answers could be gven on a seven-pont scale from 1 (fully agree) to 7 (fully dsagree). Based on ths nformaton, we constructed the varable No_redstrbuton, whch takes a value of 1 f respondents chose 6 or 7,.e. uttered a strongly negatve atttude towards redstrbuton. Includng ths varable as an addtonal regressor n the above equaton allowed controllng for other motves that may nstll skeptcsm towards hgher taxes and that may be correlated wth respondents ncome. Table 2 near here Table 2 demonstrates that, not surprsngly, a generally negatve vew on the government s role n reducng economc nequaltes sgnfcantly rases the probablty that survey partcpants are aganst or strongly aganst hgher taxes mposed on the rch. Interestngly, however, there s an addtonal effect of ncome, no matter whether we smply use the ncome categores defned by the GESIS panel, or consder monetary magntudes, ether for the total household, or on a per-earner bass. 15 We nterpret these results as evdence that materal self-nterest s mportant n defnng partcpants atttude vs-à-vs tax polcy: even f we control for the possblty that partcpants may reject redstrbuton for reasons that have nothng to do wth ther own nterests e.g. because of general concerns about farness or economc effcency 15 Usng wave-c nformaton on respondents age, gender, educatonal attanment and ncome, we found that women are less lkely to reject government nterventon to reduce dfferences between the poor and the rch, whle educatonal attanment and ncome have a sgnfcantly postve effect on No redstrbuton. Hence, the total drect and ndrect effect of the rght-hand-sde varables n equaton (6) on people s rejecton of hgher taxes may actually be hgher than suggested by the coeffcents dsplayed n Table 2. 15

18 we fnd that rcher ndvduals are more lkely to oppose hgher taxes. 16 Endowed wth these nsghts, we wll now turn to the queston whether materal self-nterest also affects ndvduals support of referenda as a procedure to decde on tax polcy. 6. Household ncome and ndvduals support for referenda on taxaton As mentoned above, economc self-nterest does not have to be the only motve that gudes ndvduals atttudes towards varous consttutonal desgns. In fact, materal motves may even be of second-order mportance, and ndvduals may support drect democracy smply because they consder t a democratc, partcpatory and transparent way of arrvng at publc decsons. Or they may reject drect democracy because they antcpate the perl of voters succumbng to populst rhetorcs. In the model of Secton 3, the relatve mportance of ntrnsc (as opposed to nstrumental) motves was captured by the parameter. The goal of ths secton s to test whether we can reject the hypothess that = 0,.e. that procedural choces are completely domnated by ntrnsc consderatons. Ths, of course, requres to specfy the emprcal model such that t properly reflects ndvduals materal nterests. To acheve our goal, we use the nformaton already descrbed n Secton 4: the GESIS panel asks respondents to state whch procedure should be used to decde on the queston whether hgher taxes are mposed on the rch. In a frst step, we explore whether an ndvdual s ncome per se has an effect on the lkelhood that she pcks a referendum. However, we must be aware that an ndvdual s atttude towards referenda as a procedure to decde on taxaton may be drven by a generally postve or crtcal atttude vs-à-vs drect democracy. Ths, n fact, was the role of the parameter n the model of Secton 3. If ths parameter s correlated wth y, a negatve coeffcent of an ndvdual s ncome does not necessarly prove the mportance of materal nterests. 16 As an addtonal check, we tested whether accountng for respondents expectatons about ther future fnancal stuaton changed our results. Usng ndvduals reply to tem debl244a n the GESIS panel ( I expect that my fnancal stuaton wll be sgnfcantly mproved n the near future ), we found that, ceters parbus, partcpants wth an optmstc vew on ther fnancal future were more lkely to reject hgher taxes mposed on the rch. However, ncludng ths varable n the regresson had no mpact on the effect of current ncome. 16

19 To control for, we therefore use a varable from the December-2015 wave of the GESIS panel, whch reflects respondents general vew on referenda. Specfcally, ths tem confronted respondents wth the followng statement: There should be more referenda n Germany. Partcpants could choose on a seven-pont scale, reachng from fully dsagree (1) to fully agree (7). Includng ths varable (Referendum_Preference) n our regresson controls for all motves nstrumental and ntrnsc that may determne an agent s general vew on drect democracy. It thus helps to solate the effect of ndvduals ncome on ther choce of referenda as a procedure to decde on tax polcy. 17 We control for other characterstcs that mght affect ndvduals vew on whether referenda should be used to decde on taxaton by controllng for respondents age (Age) and by usng dummy varables that reflect respondents gender (Female), ther educatonal attanment (Unversty entrance degree), as well as ther ctzenshp (German ctzen). We start by gnorng the role of ndvduals expectatons about the outcome of a referendum and estmate the followng equaton: Referendum Income Referendum_Preference x (7) k k In (7), Referendum s a dummy varable whch equals 1 f respondent chooses a referendum as a procedure to decde on tax polcy, and zero otherwse. As n Secton 5, we estmate ths equaton usng both OLS and the logt estmator. Table 3 gves coeffcents, average margnal effects (for logt estmaton), and t-statstcs based on robust standard errors. Table 3 near here 17 Regressng Referendum_Preference on varous soco-economc characterstcs (as reported n wave c of the GESIS panel) revealed that household ncome, the Female dummy and the Unversty entrance degree dummy had a negatve effect on ndvduals general support for referenda. To check whether the ncluson of Referendum_Preference was drvng our results, we also ran all subsequent regressons droppng ths varable. Whle ths omsson substantally reduced the explanatory power of our model, t dd not affect our key results. 17

20 The coeffcents n the frst row of Table 3 ndcate that, ndeed, ndvduals general enthusasm for drect democracy has a strong effect on ther vew whether referenda should be used as a procedure to decde on tax polcy. In addton, older and bettereducated ndvduals are less lkely to support the use of referenda for ths polcy ssue. However, there does not seem to be a separate effect of ncome that goes beyond ts nfluence on agents general atttude towards drect democracy. 18 In the lght of the model presented n Secton 3, the rrelevance of ndvduals ncome for ther support of referenda on taxaton does not come as a surprse. As we have demonstrated n Secton 3, the margnal effect of ncome depends on the sgn of E referendum 0.5,.e. on ndvduals expectaton on whether a referendum wll result n hgher taxes or not. To account for the role of expectatons, we use the GESIS panel s nformaton on respondents assessment of the majorty s opnon (see Secton 4). More specfcally, we defne two varables: Majorty atttude on taxaton takes a value of one f a respondent expects a clear majorty n favor of hgher taxaton, a value of mnus one f she or he expects a clear majorty aganst taxaton, and zero otherwse. Interactng ths varable wth agents ncome allows testng the hypothess that the margnal effect of ncome on the support for referenda depends on the expected majorty opnon: a hgher ncome should make ndvduals more skeptcal about referenda f they expect a clear majorty n favor of hgher taxaton. Conversely, agents wth a hgher ncome should be more enthusastc about referenda f they expect a clear majorty aganst hgher taxaton. The second varable we use s the dummy varable Majorty pro tax, whch takes on a value of one f a respondent expects a clear majorty n favor of hgher taxes and zero otherwse. Agan, we nteract ths varable wth respondents ncome n order to test the hypothess that ndvduals wth a hgher ncome are more lkely to reject a referendum f they expect a clear majorty n favor of hgher taxes. Unlke Majorty atttude on taxaton, the dummy varable Majorty pro tax 18 The analyss by Meya et al. (2015) suggests that ndvduals are more lkely to abstan from votng f partcpaton would confront them wth a conflct between ther socal preferences and ther dstrbutonal nterests. Such a conflct could also reduce rch ndvduals overall support for drect democracy. To test whether ths constellaton nfluences our results, we created a dummy varable that equaled one f a respondent supported more redstrbuton n general, but belonged to the group of hgh-ncome earners wth a net household ncome above 4000 Euros per month. Usng ths dummy varable as an addtonal control varable n equaton (7) yelded a negatve, but nsgnfcant coeffcent. 18

21 does not take any stand on how ndvduals ncome affects ther atttude towards referenda f they expect a clear majorty aganst hgher taxes. However, even a negatve sgn of the nteractve terms mentoned above does not necessarly prove the mportance of ndvduals materal self-nterest n defnng ther preferences over alternatve procedures. As demonstrated n Secton 3, respondents may be reluctant to support referenda f they expect a clear majorty n favor of hgher taxaton, but not necessarly for materal reasons: they mght just have stronger vews on the economy-wde effects of redstrbuton and therefore oppose the antcpated polcy outcome. If these vews, as reflected by a postve value of, are correlated wth ncome, ths may result n an erroneous nterpretaton of whatever effect we estmate. To solve ths problem and to account for nstrumental, but not necessarly materal motves to support or reject referenda, we ntroduce two addtonal dummy varables: the dummy varable Atttudes_algned assumes a value of one f respondents are n favor (or strongly n favor) of hgher taxes and expect a clear majorty of the populaton to share ths vew, or f they are aganst (or strongly aganst) hgher taxes and expect a clear majorty of the populaton to oppose hgher taxes as well. Otherwse, the dummy varable takes on a value of zero ether f respondents do not utter an opnon on taxaton, or f they do not expect a clear majorty n ether drecton. Conversely, the dummy varable Atttudes_contrastng s one f a respondent sees herself or hmself n opposton to the expected majorty, and zero otherwse. Whle both dummy varables allow dentfyng an nstrumental motve behnd agents support or rejecton of referenda, ths nstrumental motve does not necessarly reflect materal self-nterest. Includng Atttudes_algned and Atttudes_contrastng as control varables thus further contrbutes to solatng the purely materal consderatons behnd ndvduals atttude towards referenda. The regresson equaton we estmate looks as follows: Referendum Referendum_Preference Atttudes_algned (8) Atttudes_contrastng Majorty Income Majorty Income k x k k Our key regressor of nterest s the nteractve term between a respondent s expectaton on the majorty s poston (Majorty ) captured ether by the varable 19

22 Majorty atttude on taxaton or by the dummy varable Majorty pro tax and one of the (household) ncome varables ntroduced n Secton 4. The clam that materal consderatons play a role n determnng ndvduals atttude towards drect democracy s reflected by the hypothess that the coeffcent s sgnfcantly negatve. Identfyng non-lnear margnal effects n dscrete choce models s far from trval (see A and Norton 2003, and Greene 2010 for a dscusson). The core of the problem s that not even the sgn of the estmated coeffcent let alone ts absolute value and standard error s nformatve about the margnal effect of a varable, wth the latter strongly dependng on the propertes of the underlyng (logstc or normal) dstrbuton and on the pont where t s measured. In order to avod such problems and snce the results n Tables 1 to 3 suggest that the OLS coeffcents don t dffer substantally from the average margnal effects for the logt estmator we decded to estmate equaton (8) by OLS. Table 4 near here The results presented n Table 4 ndcate that the negatve nfluence of Age and Unversty entrance degree on the support for referenda over taxaton can stll be observed for the new specfcaton. Moreover, Atttudes_algned has a strctly postve mpact on ndvduals support for referenda whle Atttudes_contrastng does not seem to matter: apparently a respondent s enthusasm for drect democracy s sgnfcantly enhanced f the majorty s expected to share her or hs opnon. Conversely, the expectaton that a clear majorty holds a vew whch s n contrast to one s own opnon does not seem to affect ndvduals atttude towards referenda. Most mportantly, whle respondents ncome and expecatons about the majorty per se do not have a sgnfcant mpact on ther atttude towards referenda, the coeffcent of the nteracton term s sgnfcantly negatve for columns (1) to (8). Ths suggests that regardless of her/hs general atttude towards drect democracy a hgher (lower) ncome reduces (rases) an ndvdual s support for referenda as a procedure to decde on taxaton f she or he expects a clear majorty n favor of more redstrbuton. Columns (9) and (10) whch use the dummy varables for hgh and low household ncome as well as ther nteractons wth the Majorty varable suggest that the effect of ncome s predomnantly drven by the low-ncome recpents: ceters parbus.e. regardless of agents ncome poston the expectaton of a clear majorty n favor of hgher taxes 20

23 dampens ndvduals support for drect democracy (see the fourth and ffth row n columns (9) and (10)). Enthusasm for referenda s further reduced for partcpants who receve a low ncome. However, f these partcpants expect a clear majorty n favor of taxaton, the lkelhood of supportng a referendum as a procedure to decde on tax polcy ncreases substantally. More specfcally, the results dsplayed n column (10) suggest that partcpants who expect a clear majorty to support taxaton are 12 percentage ponts less lkely to choose a referendum than (otherwse dentcal) persons who ether have no clear vew on the majorty poston or expect the majorty to oppose hgher taxes. By contrast, a low-ncome ndvdual who expects the majorty to be n favor of hgher taxes s 23 percentage ponts more lkely to pck referendum as her preferred procedure than a low-ncome ndvdual wthout such expectatons, and 27 percentage ponts more lkely to pck a referendum than an ndvdual who does not receve a low ncome, but shares these expectatons. Whle t s surprsng that we do not fnd a partcularly strong opposton aganst referenda by those hgh-ncome ndvduals who expect a majorty n favor of hgher taxes, our results support the noton that materal nterest does play an mportant role for ndvduals preferences over decson-makng procedures. So far, our analyss has focused on respondents choce of referenda as a procedure to decde on taxaton, lumpng together all other procedures parlamentary decsons, expert decsons, decsons after delberaton among all affected groups nto one (composte) alternatve choce. In a last step, we test whether ndvduals ncome combned wth ther expectatons about the majorty s poston also affects the support for these other procedures. If the pattern dscernble n Table 4 were also dscernble for alternatve ways of arrvng at a decson, our fndngs would lose a lot of ther bte, snce we argued that the clear relatonshp between majorty expectatons and antcpated polcy outcomes s a specfc property of referenda. We thus re-ran regresson (8), subsequently replacng the dependent varable Referendum by dummy varables (Parlament, Experts, Delberaton) that took a value of one whenever a respondent chose one of these procedures. Once more, we used the OLS regressor. And to save space, we restrcted our attenton to two partcular versons of equaton (8) the one that used the level of net household ncome (n mllon Euros) as a proxy for respondents prosperty, and the one that used the hgh- and low-ncome dummes. 21

24 Moreover, we only show the results for the specfcatons that used the Majorty pro tax dummy to represent respondents expectaton about the majorty s opnon. 19 Table 5 near here For the sake of comparablty, the frst two columns of Table 5 replcate columns (4) and (10) of Table 4,.e. the effect of agents characterstcs, ncome, and expectatons on the lkelhood of preferrng a referendum. The coeffcents and sgnfcance levels shown n the other columns support our conjecture that respondents household ncome, combned wth ther expectaton on the majorty opnon, are partcularly relevant for ther choce of referenda as a decson-makng procedure. By contrast, these varables do not seem to matter for respondents support or rejecton of parlamentary decsons. Interestngly, however, respondents tend to support expert decsons when they expect a majorty to be n favor of hgher taxes, but the enthusasm for ths procedure s substantally muted for low-ncome respondents who share these expectatons. Understandably, ths group of ndvduals prefers the use of referenda to decde on taxaton. Fnally, a hgher ncome per se seems to reduce ndvduals support for delberatve procedures that gve a voce to all affected groups, but ths result s nether drven by low-ncome nor by hgh-ncome ndvduals. There s thus no other procedure for whch ndvdual support, ceters parbus, depends on the specfc combnaton of ncome and majorty expectatons as t s the case for referenda. We nterpret these results as further evdence that materal motves affect ndvduals preferences over procedures. 7. Summary and conclusons Whle few economsts would contest the dea that ndvduals atttudes towards specfc polcy ssues e.g. redstrbutve taxaton reflect ther materal self-nterest, t s far less obvous that such consderatons should also play a role at the consttutonal stage,.e. when the rules of the poltcal process are defned. It could be argued that values 19 Runnng these regressons for the full set of ncome varables and usng the varable Majorty atttude on taxaton as an alternatve to the Majorty pro tax dummy yelded smlar results. Moreover, whle the ncluson of Referendum_Preference enhances the explanatory power of our model, t does not affect the other qualtatve results dsplayed n Table 5. 22

25 such as democratc autonomy, procedural farness, ctzen partcpaton and transparency are far more mportant than materal self-nterest n determnng agents preferences over decson-makng procedures. Focusng on ndvduals support for referenda as a procedure to decde on taxaton, our analyss has tred to shed lght on ths ssue. Usng a survey that we desgned wthn the German GESIS panel, we frst demonstrated the postve relatonshp between respondents ncome and the lkelhood that they reject redstrbutve taxaton. We showed that ths relatonshp emerges even f we control for other, not necessarly materalst motves to oppose more redstrbuton. Turnng to ndvduals support for referenda, we then demonstrated that ncome per se does not have a sgnfcant effect on ndvduals procedural choces. However, once we accounted for respondents expectatons about the majorty opnon, we could demonstrate that the support for drect democracy decreases wth ncome f ndvduals expect a clear majorty n favor of taxaton. Contrary to other contrbutons that merely assume agents ablty to antcpate the poltcal-economc equlbrum, our analyss makes these expectatons explct and thus documents the mportance of materal nterests n shapng ndvduals preferences over alternatve democratc decsonmakng procedures. Whle our results ndcate that the tradtonal poltcal-economc, nterest-based explanaton of consttutonal choces s alve and well, we also have to stress that our fndngs do not rule out the relevance of ntrnsc motves for procedural preferences n general, and for supportng referenda and drect democracy n partcular: as we have argued above, the hghly sgnfcant varable Referendum preference s lkely to reflect a mx of ntrnsc and nstrumental consderatons. Moreover, the generally low R 2 values, whch our study shares wth most survey-based emprcal analyses, sgnals that we are stll far away from completely understandng procedural choces. Nevertheless, our fndngs demonstrate that materal self-nterest s a factor to be taken serously when t comes to explanng ndvduals atttudes towards referenda. 8. References Acemoglu, D. and J. A. Robnson (2000). "Why dd the West extend the franchse? Democracy, nequalty, and growth n hstorcal perspectve." The Quarterly Journal of Economcs 115(4):