An indirect approach to map ethnic identities in post-conflict societies

Transcription

1 An ndrect approach to map ethnc denttes n post-conflct socetes Danel Bochsler* NCCR Democracy, Unverstät Zürch Affolternstrasse 56 CH Zurch Swtzerland E-mal: danel.bochsler@uzh.ch web: Tel Fax * correspondng author Basl Schläpfer Centre for Democracy Studes Aarau (ZDA) Küttgerstrasse Aarau Swtzerland E-mal: basl.schlaepfer@uzh.ch Ethncty remans one of the most salent layers of ndvduals socal denttes, and nformaton about the dstrbuton of ethnc denttes turns out to be crucal for many studes that nvestgate poltcal or socal processes n dvded socetes. In the aftermath of cvl wars, however, censuses provdng such data are controversal and often delayed. Where census data s lackng, researchers can make use of the multpler method to nfer the dstrbuton of ethnc denttes based on ndrect observatons from sub-samples of the populaton. However, due to the selectve nature of ther data, the sub-samples mght not be representatve of the populaton. Ths paper proposes a method, whch corrects for such selecton effects. Based on the ethnc dentty of brth-gvng parents, we estmate the dstrbuton of ethnc denttes n the muncpaltes of the Federaton of Bosna and Herzegovna for We correct for possble selecton bases by ncludng economc, demographc and war related varables. Multple tests of valdty show that our estmaton appears to be the most accurate procedure currently avalable for the dstrbuton of ethnc denttes n muncpaltes n Bosna and Herzegovna. Keywords: Ethnc poltcs; denttes; Bosna and Herzegovna; multpler method. 1

2 Introducton In post-war socetes, populaton censuses can be hghly poltcsed ssues. Whether and how ethncty should be subject to census questons s often controversal. 1 Wth good reason: askng ctzens about ther ethncty contrbutes to the consoldaton of hstorcal categores, and can help to foster poltcal dvsons along ethnc lnes (Horowtz, 1985, pp ; Kertzer & Arel, 2001; Szreter et al., 2004; Atken, 2007). In consocatonal settngs, quotas and the poltcal control over the terrtory are sometmes drectly lnked to the offcally recognsed ethnc structure of the populaton (McCulloch, 2014; McCrudden & O'Leary, 2013; Wllams & Husk, 2013, p. 297). Therefore, poltcal eltes mght even be ncentvsed to moblze a suffcently large support for ther ethnc category n the offcal census. In such settngs, census questons about ethncty, language, or relgon are poltcally controversal (Vsoka & Gjevor, 2013), and censuses are often delayed (recently n Fj or Macedona) or not conducted at all (Lebanon). At the same tme, the lack of systematcally collected data about the dstrbuton of ethnc denttes n the ntermedate perod s a major obstacle for academc nvestgatons that would requre ths knd of data (Mtchell et al., 2009; Buhaug & Rød, 2006; Ralegh & Hegre, 2009). For example, ethnc denttes can be related to clentelsm (Franck & Raner, 2012), and studes of poltcal behavour and radcalsaton n post-conflct socetes often requre nformaton on ethnc dversty (Pugh & Cobble, 2001; Bellon, 2007, pp ). The lack of data has consequences for academcs and for polcy makers. The dstrbuton of denttes can affect securty and nterethnc cooperaton (Fearon & Latn, 1996; Varshney, 2001), and t s one man predctors of refugee returns (Internatonal Crss Group, 1999; Nalepa, 2012, pp. 349, note 369). Therefore, research occasonally resorts to ndrect proxes (Caspersen, 2004). Where offcal data s lackng, ndrect estmators, whch do not gan the same legtmacy as offcal censuses, are used to map the dstrbuton of ethnc denttes. There s a possbly unlmted 1 Cf. BBC News, 31 August 2010, Kenya defends trbal census fgures, 2

3 range of sources, whch are and can be used as a proxy of ethncty. Often, studes nfer denttes from the dstrbuton of names n the populaton (Harrs, 2015; Internatonal Crss Group, 2003, pp. 6-7,15; Ellott et al., 2008; Mateos et al., 2011). Other useful ndcators can nclude membershp n relgous communtes (Susewnd, forthcomng), relatonshp to kn communtes, or the consumpton of goods and servces, whch are typcal for dfferent ethnc communtes. Della Vgna and colleagues even used beer brands sold n local bars as proxes of ethncty (DellaVgna et al., 2014, pp ). Such ndrect ndcators mght n many contexts come close to a perfect predctor of ethncty at the ndvdual level. Nevertheless, such ndrect ndcators relate to a sub-sample of the overall populaton. Data on membershp of relgous communtes or the consumpton of consumer goods relate to church members or consumers, not to the entre populaton. Whle they mght help to nfer the dentty of ndvduals, the membershp rate n relgous communtes, or the per capta consumpton mght vary between ethnc groups, and between muncpaltes: e.g., church membershp mght be more frequent for group A than B, and more frequent n rural than n urban muncpaltes. Even regster data (marrage regsters, death or brth regsters) (Ellott et al., 2008; Mateos et al., 2011) only capture a sub-sample of the populaton, e.g. newly marred couples, rather than the entre populaton, and the ncluson nto ths sub-sample vares between groups and muncpaltes. Hence, they are not representatve of the overall populaton. Ths paper takes ths knd of sub-sample bases nto account and proposes a multpler approach, to nfer the dstrbuton of ethnc denttes,.e. t estmates the characterstcs of the overall populaton from a sub-sample. We use nformaton on the dentty of parents of newly born babes n the post-war perod n the Federaton of Bosna and Herzegovna, n order to estmate the dstrbuton of denttes n each of ts 77 muncpaltes. For ths perod, no census data are avalable. In Bosna and Herzegovna, the frst post-war census was only very recently conducted (n 2013), 18 years after the peace agreement of Data from the pre-war census n Bosna and Herzegovna s outdated, due to sgnfcant movements n the populaton and `ethnc cleansng 3

4 durng the war (Melander, 2007). Our approach reles on data from offcal brth regstres (Federaln zavod za statstku, ), and multple tests of valdty. Both the appled methodology and the estmated ethnc composton of the muncpaltes of the Federaton of Bosna and Herzegovna contrbute to the research on ethnc poltcs, and on postwar socetes: frst, both quanttatve and qualtatve studes n the feld often rely on measurements of the ethnc composton of a terrtory, of a town or a regon of nterest. The methodology proposed n ths paper provdes for estmates where other, partcularly offcal data s lackng. Second, and n partcular reference to Bosna and Herzegovna, the lack of data on ethnc denttes s a major challenge to academc research, whether or not the appled method s of qualtatve or quanttatve nature. Ths paper provdes for the frst estmate for the post-war perod, although lmted to the Federaton. As multple methods of valdaton show, our estmates are more plausble than earler attempts by state and nternatonal agences to estmate the dstrbuton of ethnc denttes, whch have never been publshed. At the same tme, the fact that a census has been conducted n 2013 means that we apply our approach to a country n whch decson makers have decded to collect and provde such data, although the data collecton has been subject to rregulartes n the most contested muncpaltes. 2 Our estmaton wll provde a bass of comparson. Addressng socal scence methodology more generally, the paper nnovates by provdng a new methodology, that corrects for uneven dstrbutons of brth rates across ethnc groups and muncpaltes. The method can smlarly be appled to other cases, where socal scentsts nfer data ndrect, from sub-samples of the populaton, and where alternatve estmaton methods are not feasble. When nterpretng and usng such methods, one needs to keep n mnd that the nature of the data generaton process has an mportant mpact on the results (Nagle & Clancy, 2012). In our case, the brth regsters have been compled by state offcals. Parents regster wth only one nomnal 2 See, for nstance, Oslobodjenje, 16 October 2013, (last accessed on 15 February 2015). 4 Results and replcaton data avalable from the authors webste. 4

5 dentty, whch mples that the regstres do not nclude nformaton on multple or mult-layered denttes (Chandra, 2006; Lee, 2009). As denttes are hghly poltcsed n Bosna and Herzegovna, the regstraton through a state offcal mght ncrease the number of persons regstered as members of one of the locally domnant denttes (.e., Serbs, Croats, or Bosnaks). Smlar concerns have been rased for the offcal census, whch s underway. We proceed as follows: After a bref revew of the role of ethncty n Bosna and Herzegovna, the second secton of ths paper develops our estmator of ethnc dversty and apples t to muncpaltes of the Federaton of Bosna and Herzegovna. 4 The thrd secton provdes a seres of tests to examne the valdty of our procedure. The concluson dscusses the broader applcablty of our research. 1. Ethncty n Bosna and Herzegovna In Bosna and Herzegovna, the debate about the ethnc denomnatons s a result of the war of Ths secton provdes for a hstorcal account and an overvew over earler estmatons. For nearly half a century, Bosna and Herzegovna has been defned as a mult-ethnc republc, wth three state people: Serbs, Croats, and Muslms (or `Bosnaks ). Muslms were recognsed n 1968, when the country was stll part of the Socalst Federal Republc of Yugoslava, as a consttuent people, on an equal bass to Serbs and Croats, whch led to an ncrease n number of ctzens dentfyng as Muslms (Burg & Shoup, 2000, pp ). In 1993, the ethnc Muslms were offcally re-named Bosnaks. The Dayton peace agreement of 1995 carred on and renforced the ethncsaton of poltcs n Bosna and Herzegovna. It created a complex federal structure: one entty, whch has been called Republka Srpska, s domnated by the Serbs. The other entty, the Federaton, conssts of ten cantons, eght of whch are ether Bosnak- or Croat-domnated. The ethnc separaton s largely the product of the war, durng whch large-scale dsplacement along ethnc lnes took place. The Dayton agreement would allow dsplaced persons to return, and to re-establsh the pre-war ethnc 5

6 dversty, but the process has not been completed. However, a more nuanced pcture of the ethnc dstrbuton after the war s lackng. In ths stuaton, conductng a census s poltcally controversal: whle some poltcans understand t as a prelmnary snapshot of the achevements of the poltcs of returns, t can also serve to offcally recognse the post-war terrtoral dvsons. Aganst ths backdrop, no census has been conducted that nformed about the dstrbuton of ethnc denttes up untl As a result, for more than two decades, data on the terrtoral dstrbuton of ethnc denttes s lackng. In 2005, the Federal Offce of Statstcs has estmated the ethnc structure of all 79 muncpaltes of the Federaton of Bosna and Herzegovna, but t s possble that ther estmates mght partally stll rely on the pre-war census (Federaln zavod za statstku, 2005). Based on more recent data, the OSCE has estmated the dstrbuton of ethnc denttes n 65 muncpaltes, but ther methodology has not been publshed. Toal and Dahlman (2011) provde estmates for fve muncpaltes of the Federaton, the Internatonal Crss Group (2003, pp. 6-7,15) has estmated the dstrbuton of ethnc denttes n Mostar from the entres n the electoral regster, Caspersen {, 2004 #5310} reles on ndrect proxes from electoral results. They seem to be straghtforward, as the poltcal landscape of Bosna and Herzegovna s hghly ethncsed. We refran from usng such a strategy, however, as t would not allow the clear dentfcaton of voters of mult-ethnc partes, and because data derved from electon results would be hghly endogenous for the explanaton of poltcal behavour or poltcal cooperaton. 2. A two stage procedure for the estmaton of the dstrbuton of ethnc denttes We propose an approach that s based on a sub-sample of the populaton n order to estmate the dstrbuton of ethnc denttes for some of these years. There are a few common methods to estmate populatons based on ncomplete data, ncludng capture-recapture, enumeraton, and multpler technques. The capture-recapture technque uses 6

7 two or more lsts of ndvdually dentfable subjects to estmate the total number of subjects belongng to a gven populaton (e.g. Bshop et al., 1975, pp ; Ball & Asher, 2002). Enumeraton methods are smlar to census methods, but lmted to pre-defned segments of the target populaton. Fnally, multpler methods work wth two ndependent records of data that relate to each other and extrapolate certan characterstcs from a sub-sample of the populaton to the overall populaton, assumng that the sub-sample of the populaton s representatve of the overall populaton (UNAIDS/WHO 2003, pp ). Approaches based on ndrect observatons rely on the multpler method, even though ther sub-samples are usually not representatve, and therefore, they volate the underlyng assumptons. Indrect approaches are based on observable manfestatons of ethncty, such as the presence of languages or symbols n the publc space, ndvdual (consumer or poltcal) behavour, whch s related to certan ethnc groups n a specfc context (DellaVgna et al., 2014). They can also be related to drect observatons of ethncty or closely related denttes (relgous afflaton, languages, names) for a sub-sample of the entre populaton (Abrahamse et al., 1994; Internatonal Crss Group, 2003, pp. 6-7,15; Ellott et al., 2008; Mateos et al., 2011). Our estmaton of the dstrbuton of ethnc denttes n Bosna and Herzegovna s based on the multpler method, but corrects for the selectvty of the sub-sample. We explot the fact that whle no representatve survey or census s avalable to measure the dstrbuton of ethnc denttes at the muncpal level, nformaton on ethnc denttes s collected upon the regstraton of newly born chldren. Every brth regstraton ncludes the ethnc dentty of both parents of newborns. From the parents denttes, we extrapolate to the overall muncpal populatons. Parents of newly born chldren are, of course, not a random or representatve sample of the full populaton. Brth rates vary geographcally and across ethnc and socal groups, so that the number of parents cannot drectly nform us about the overall sze of the group they belong to. In other words, parents of newborns are not only a selectve sub-sample of the overall populaton. Varaton n brth rates may also be lnked to those attrbutes of the populaton, whch we are partcularly 7

8 nterested n, namely ethncty and geography. Therefore, our methodology reles on a multpler method (extrapolatng from a sub-group to the overall populaton), wth a correcton procedure ntroduced n order to account for the varance n brth rates across socal groups, ethnc groups, and muncpaltes. 5 The model Data on the denttes of parents s reported n the brth statstcs of the Federaton of Bosna and Herzegovna, for each of the three consttuent groups (Serb, Croat, Bosnaks), and several other groups. We also have data on the overall sze of the populaton of the muncpaltes. Our model wll thus be based on the followng nformaton (and a set of further control varables, such as numbers of refugees/returnees and GDP estmates): - p T total populaton sze of muncpalty - b total number of newly born chldren n muncpalty - b e, parents (mothers and fathers) of newly born chldren n muncpalty, by ethnc dentty e, where e dstngushes Bosnaks (B), Croats (C), Serbs (S), and others (O) We wll use ths nformaton to derve an estmaton of the followng varable: p e, populaton of ethnc group e n muncpalty (p S, = ethnc Serbs n muncpalty, etc.) Nave Brth Rate Model A frst, drect applcaton of the multpler method allows us to obtan an estmate of the ethnc populaton structure. In ths case, we assume that the parents of newly born chldren are a nonselectve sample of the overall populaton, and treat them thus as a representatve sub-group. Hence, 5 The relance on brth-related baselne data from medcal nsttutons s not uncommon when t comes to the estmaton of populaton fgures (WHO, 2003, pp. 3-5). 8

9 we estmate a sngle brth rate for each muncpalty ( b p ). We derve the populaton structure by ethnc group by applyng the same sngle brth rate to each of the ethnc groups and each muncpalty, be, pe,. Ths drect applcaton would lead to vald results under the assumpton that there are no sgnfcant dfferences between the brth rates, age structures, or gender balances of dfferent ethnc groups,.e. that brth rates only co-vary wth local characterstcs, and that they are not connected to ethncty. We call ths the nave brth rate model, whch assumes that those condtons hold. If, however, the brth rate vares across ethnc groups, ether due to demographc or value-related dfferences between the ethnc groups, the estmates wll be based. Full Brth Rate Model The nave model underles the assumpton that brth-gvng parents are a representatve sample of the overall populaton. Ths s mplausble: war and mgraton have sgnfcantly altered the structure of the populaton, and not only partcularly affected the ethnc, but also the age structure and gender balance of the socety. For ths reason, a vald estmaton should consder possble varaton n brth rates, due to dfferences between the muncpaltes, whch mght affect the rate of ncluson nto our sub-sample (parents). In partcular, these are the dstrbuton of age groups and the gender balance across muncpaltes, and further correlates of brth rates, such as economc development or urbanty. Therefore, we allow the brth rate γ e, to vary between muncpaltes and between ethnc groups e. We defne the brth rate as the quotent of the number of newly-born chldren b e, from ethnc group e, n muncpalty and the number of nhabtants declarng themselves as belongng to ths ethnc group p e,. We do not have any nformaton of γ e, or of p e,. be, e, (1) p e, 9

10 We know, however, that the sum of the populaton of all ethnc groups p e, resdng n a muncpalty s equal to the total sze of the populaton, p T. p T p p p p (2) B C S O The two equatons (1) and (2) can be re-arranged as follows: p T bb, B, bc, C, bs, S, b O, O, We dvde the overall muncpal populaton p T by twce the overall number of local brths 2 b, and ths results n the nverse overall brth rate, 1 p b T. We add an arrow β x X, whch stands for further covarates of the local brth rate (demographc structure, economy, etc.). 1 p b T bb B b bc b C bs b S bo x X O b (3) The resultng equaton (3) s a lnear functon, where each of the fve unknown varables the brth rates γ B, γ C, γ S, γ O, and the parameter for the control varables, β x s part of one summand. We can estmate the functon wth a lnear regresson model, where 1 s our dependent varable. The equaton s not free of assumptons. We allow for brth rates to vary by ethnc groups,.e. we estmate brth rates for each ethnc group, γ B, γ C, γ S, γ O. Further control varables X allow brth rates to vary across muncpaltes. However, we assume that brth rates vary smlarly for all ethnc groups across muncpaltes. For nstance, f rural muncpaltes have hgher brth rates than urban muncpaltes, ths wll be captured by our control varable. Ther effect s fxed for all groups n a muncpalty. 10

11 Model specfcaton and varables Ths paragraph dscusses the data on whch we rely n our estmaton and ssues whch we consder to be crucal for estmatng populaton structures n a post-war socety. The brth regstraton statstcs of Bosna and Herzegovna nclude nformaton about the ethnc dentty of both parents of newborns (Federaln zavod za statstku, ). As the data refers to the parents, not the chldren, and as there are two entres for each chld, the estmates would lead to a result, whch s twce hgher than the brth rate; hence we dvde the numbers by two. The dentty should be self-reported, and regstered by the state offcal ('matčar'). In practce, however, state offcals occasonally do regster parents who do not clam to belong to any of the offcal ethnc groups by selectng a group based on the parents names. 6 Further demographc and soco-economc varables were ncluded to control for possble bases. Frst, due to mgraton and war casualtes, the gender balance and the age structure mght vary across muncpaltes, leadng to varyng brth rates. The avalable data allows us to restrct the populaton to those aged 18 to 64, and to test for the robustness of the results, compared to the overall populaton. We cannot control drectly, however, for possble effects of the gender balance. Brth rates tend to vary n dfferent soco-economc contexts. We control for the economc development (local GDP estmate), the number of refugees and returnees, and ndrect measures of the degree of urbanty (populaton sze of the muncpalty and populaton densty, logarthmsed). In some muncpaltes, our estmates wll be less precse. In partcular, ths wll be the case for small muncpaltes, as we use a small sub-sample of the entre populaton (parents), and wth a smaller sze of the populaton, nference from a sub-sample becomes less accurate. Smlarly, as refugees were usually expelled from ther homes based on ther ethnc dentty, and ther return 6 Intervew wth a poltcal analyst, Sarajevo, on 21 May 2012 (kndly provded by Elena Stavrevska). The regstraton card can be found n the appendx of Federaln zavod za statstku (2009). 11

12 depends on the economc and socal context of the muncpaltes, refugees and returnees are no homogeneous group, and thus have not a homogeneous effect on muncpal brth rates. The estmaton wll also be based n muncpaltes, where large parts of the populaton do not belong to the age cohorts, who mght have chldren, or wth an unbalanced gender dstrbuton. Hence, these control varables have heteroscedastcty effects. Therefore, our models wll perform less well for muncpaltes wth hgh numbers of refugees and returnees, small muncpaltes, and those wth a specfc demographc structure. Because we expect such heteroscedastcty effects, we run lnear models wth separate varancepredctors. Our models use two sets of predctors. On the one hand, the model estmates the average effect on the mean of our dependent varable. On the other hand, a second set of explanatory varables s expected to estmate the accuracy of the model, and therefore s related to the unexplaned varance of the dependent varable, σ 2 (Davdan & Carroll, 1987; Braumoeller, 2006). In ths study, the man varables, ntroduced n equaton 3 brth numbers by ethnc dentty and a frst set of control varables are expected to affect the mean brth rate, and are therefore ncluded n the man part of the model. A second set of control varables s related to the accuracy of the predctons (effect on varance), consstng of the share of refugees and returnees among the populaton, the sze of the muncpalty, and the age structure (populaton <15 years). There s no nformaton avalable for the gender balance of the muncpaltes. Data on refugees and returnees s mssng for 9 out of 77 muncpaltes. We have mputed mssng nformaton for these muncpaltes. Estmaton results reman stable wth/wthout mputaton. Two muncpaltes have not regstered any brths at all (Ravno), or only very few brths (Neum). Whle Ravno s a rather small muncpalty, and thus susceptble to be an outler, Neum s surrounded by Croat terrtory, and s accordng to all sources overwhelmngly nhabted by ethnc Croats. Most parents wth resdence n Neum gve brth n hosptals n neghbourng Croata, where 12

13 they can receve health care as domestc ctzens. 7 We decded to exclude the two muncpaltes from the estmaton, as the estmates for the two cases are unrelable. However, the ncluson of the two muncpaltes does not alter the overall results of the estmaton substantally (results avalable from the authors). Results of the OLS estmaton of the brth rates are presented n table 1. Ethnc Bosnaks are our remote category, meanng that the coeffcents measure the dfferences between the brth rates of the Bosnaks and of other ethnc groups. The dependent varable (brth rates) s nverted, so that the postve coeffcent n the man model for populaton densty mples that brth rates are lower n muncpaltes wth a hgh populaton densty. Lkewse, they tend to be lower n places wth a hgh share of non-returned refugees. After controllng for these factors, we fnd that the overall brth rate lowers as the percentage of Serban parents ncreases. Among Croats, brth rates tend to be hgher than among Serbs, but lower than among Bosnaks. There are also sgnfcant effects n the varance part of the models: sze of the muncpalty, the age structure, and the share of refugees and returnees affect the accuracy of the estmaton. xxx nclude table 1 about here xxx In the second step, based on the models presented n table 1, we calculate predcted values for the brth rates n 77 muncpaltes for each ethnc group (γ B, γ C, γ S, γ O, for Bosnaks, Croats, Serbs, 'Others'), whlst consderng the soco-economc context. Equaton 1 allows us to predct the overall dstrbuton of ethnc denttes n these muncpaltes, whch we dsplay for 2008 n fgures 1 and 2. Certanly, the dstncton of muncpaltes n fgure 1 by the largest group after the war reproduces well-establshed knowledge. More mportantly however, our results also provde an estmaton of the actual sze of each group. The category Others s barely vsble; our hghest 7 (last accessed on 11 July 2013). 13

14 estmaton s for Bosansk Grahovo (0.5% of the 2008 populaton), but due to the very small number of brths by parents n ths group, the estmaton mght not be relable. xxx nclude fgure 1 about here xxx xxx nclude fgure 2 about here xxx Relablty and valdty In what follows we provde the results of a number of tests that assess the relablty and the valdty of our estmates reported above. These tests compare the results of our estmaton model to earler estmatons (by the Statstcal Offce and by the OSCE), and to the nave model. We am to show that our estmaton model outperforms the nave estmaton, and that they are also better suted than earler estmatons to measure ethnc denttes n the Federaton of Bosna and Herzegovna. Relablty test: Although the age structure and, especally n small muncpaltes the number of brths can change over tme, the dstrbuton of ethnc denttes should correlate strongly from one year to the next n the post-war perod. Hence, f relable, our estmaton results should almost be stable over the years. As a test of the relablty of our results, we therefore correlate our estmates for the avalable years ( ) (table 2). Valdty tests: We assess whether our populaton estmates correspond to the actual dstrbuton of ethnc denttes n muncpaltes n Bosna and Herzegovna. Due to the absence of offcal data, ths can only be done ndrectly. We perform two well-establshed tests of valdty. Frst, we test whether the new measures reproduce the results of other estmates (crteron valdty). We compare them to unoffcal and not publcly avalable prevous estmates by the 14

15 Federal Statstcal Offce (for all muncpaltes, n 2005) and the OSCE (coverng only 31 of the muncpaltes of the Federaton, n 2008, and matchng 30 of our cases). We correlated them wth our estmates for the year closest n tme to these, namely Ideally, the correlaton coeffcent (accordng to Pearson) should amount to 1. Our estmates of the full model are all above 0.96 for the ethnc Bosnaks and the ethnc Croats, and around for the estmates of ethnc Serbs (table 2). Our results estmate the share of Serbs to be hgher than the Federal Statstcal Offce; the dfference s partcularly pronounced for the muncpalty of Drvar (32% versus 78%). Gven these dfferences n the estmates of ethnc Serbs, t s crucal to test whch of these measures s more precse n explanng theoretcally expected patterns. xxx nclude table 2 about here xxx Second, we employ the new estmates to test for theoretcally expected relatonshps (construct valdty). In many ethncally dvded countres, ethnc denttes are good predctors of the vote for ethno-natonalst partes. Lkewse, electons n Bosna and Herzegovna are perceved as de-facto ethnc censuses. We use the vote share of ethno-natonalst partes n the 2006 parlamentary electons and n the 2008 local electons to assess our estmates. Ethno-natonalst partes obtaned 73%-75% of the votes n the Federaton n these electons. Two aspects are tested for. Frst, we expect that the muncpalty-level votng results can nform us about the dstrbuton of ethnc denttes. Due to the contextual effects of ethnc dversty on the propensty of voters to vote for natonalsts, the relatonshp s often curvlnear. 8 We analyse, thus, to what extent our measure explans the vote share of ethno-natonalst partes. Construct valdty should be hgh, f our measure s accurate, and f patterns of ethnc votng are strong. Fgure 3 8 For models explanng the curvlnearty of ths relatonshp, see Bochsler (2013) and Grofman and Handley (1995). 15

16 shows such a model exemplarly for the Croat partes, based on our full estmaton model. 9 The curvlnear lne s the best ft, and the graph demonstrates that our populaton estmates can explan the vote share of Croat partes reasonably well. Because we expect a curvlnear pattern, we use a measure of explanatory power (the adjusted R 2 ), rather than a correlaton coeffcent. Hgh values, that s, values close to 1, ndcate a hgh degree of construct valdty (fgure 4). However, some varaton unexplaned by ths model remans. Our second ndcator of construct valdty tests for the degree of underestmaton. As we are less concerned wth mnor devatons than wth larger ones, we use a measure of squared devatons (dev e ), only countng devatons, where the vote share of ethnc partes s hgher than ther group s share of the local populaton (fgure 5). 11 Small squared devatons (on a scale from 0 to 10000) ndcate a hgh degree of valdty. Ths s llustrated n fgure 3: ponts located below the dagonal lne mght ndcate cases where Croats mght not vote for Croat partes. Ponts located above the dagonal lne mark muncpaltes ndcate places where Croat partes reached a hgher vote than the estmated share of ethnc Croat resdents. Ths suggests that (assumng roughly equal turnout across all groups) our model underestmates the Croat populaton for these cases. xxx nclude fgure 3 about here xxx 9 Party classfcaton n appendx A. 11 The measure s calculated separately for each ethnc group, as follows: dev e N 1 max vote N p e, e,, 0 of the ethno-natonalst party of group e n muncpalty. 2, where N s the number of muncpaltes, vote e, s the vote share 16

17 Based on these two ndcators, our results clearly outperform two prevous estmates of ethnc dversty n Bosna and Herzegovna, the estmates of the Statstcal Offce of the Federaton of 2005, and the estmates provded by the OSCE. Our procedure appears to provde a much more accurate estmaton of the sze of the Serb and the Bosnak communtes than the estmates of the Statstcal Offce: valdaton models that are based on the fgures of the Statstcal Offce result n only moderate explanatory power for the success of Serb partes, wth an adjusted R 2 around 0.48 (2008 electons, fgure 4, left panel, and table B1 n appendx B). Our fgures allow a farly accurate explanaton (adj R 2 : 0.96). The Statstcal Offce appears to under-estmate the sze of the Serban communty (dev e : 23.6), and of the ethnc Bosnak communty (dev e : 15.5, fgure 5, left panel, and table B2 n appendx B) n certan muncpaltes, whch our full model does not. In contrast, we under-estmate the sze of the Croat communty n several muncpaltes. 12 Our measure (dev e =13.7) s slghtly less accurate than the Statstcal Offce s estmaton, but t devates less the Statstcal Offce s measure for Bosnaks and Serbs. In part, the naccuraces of the Statstcal Offce's fgures for the Serb communty mght stem from the consderable mgraton between 2005 (when the estmaton was done) and 2008 (when the electons were held). Therefore, we have repeated our valdty tests also for the 2006 parlamentary electons. Even n ths comparson, our measure consderably outperforms the populaton fgures of the Statstcal Offce for two ndcators (and s only moderately less accurate regardng a thrd ndcator) (see appendx B). To compare our model to the OSCE estmates, we restrct the valdaton procedure to the 30 muncpaltes for whch OSCE estmates are avalable (fgures 4 and 5, rght panels). The OSCE appears to under-estmate the number of ethnc Bosnaks n some muncpaltes, and our results 12 Ths s drven by Usora, we count 69% ethnc Croats and 31% ethnc Bosnaks, but 92% of the votes are cast for ethnc Croat partes. However, the number of regstered brths s very low, wth 26 brths on 7000 nhabtants n

18 outperform the OSCE on both ndcators for ths group. We also seem to have a slghtly better estmaton of the ethnc Serbs, judgng by the adjusted R 2. Both the OSCE and our fgures seem to under-estmate the number of ethnc Croats n certan muncpaltes, but here the OSCE slghtly outperforms our measures. Qute clearly, our full model mproves the estmaton, compared to the nave model, whch does not correct for dfferent brth rates across muncpaltes and ethnc groups. The dfference s most notable for the squared devaton measure. The nave model tends to under-estmate the sze of the ethnc Croat and the ethnc Serb communty systematcally for all muncpaltes. Ths also explans why the nave model and the full model have almost the same explanatory power (adjusted R 2 ). xxx nclude table 3 about here xxx 3. Conclusons In those countres, where data on the dstrbuton of ethnc denttes s most relevant for academc research, they are often also subject to poltcal struggles. As a result, questons about ethncty n the populaton census are hghly salent, and thus get poltcsed, or the census mght get delayed altogether. In the absence of (credble) census data, ethnc denttes are often estmated based on ndrect data, such as regster data (e.g. brth, marrage, death regsters), or proxes of ethnc denttes, such as membershp n relgous communtes or consumer data for goods and servces, assocated wth partcular groups. Ths paper proposes a new methodology to estmate the dstrbuton of ethncty at the subnatonal level (muncpaltes, dstrcts, etc.), relyng on ndrect data from a sub-sample of the entre populaton. Such ndrect approaches, also addressed as multpler method, assume that the subsample s representatve of the entre populaton. However, sub-samples, provdng for ndrect observatons of denttes, are not only ncomplete, but also selectve: the lkelhood that ndvdual ctzens are ncluded nto the sub- 18

19 sample wll often correlate wth ethnc denttes and wth further soco-economc varables. For nstance, brth and marrage rates, lfe expectancy, but also the share of the populaton who are members of relgous communtes vares across groups, and due to other contextual factors across muncpaltes. The method proposed n ths paper corrects for these effects. It allows thus to mprove the estmaton of ethnc denttes from ndrect data, but could also transfer to other estmatons of populatons, where some characterstcs are ndrectly observed on a sub-sample of the entre populaton. Ths paper apples the procedure the to 77 muncpaltes n Bosna and Herzegovna n the post-war perod. We rely on data from brth regsters on the ethnc dentty of parents of newly born chldren. To our knowledge, ths s the frst attempt to provde estmates of ethnc denttes for the post-war perod n Bosna, based on a transparent and systematc procedure that covers a large number of muncpaltes. Multple tests of robustness and valdty ndcate that our procedure leads to a good approxmaton of the dstrbuton of the htherto offcally recognsed ethnc denttes n many muncpaltes of Bosna and Herzegovna. Ths, despte the fact that the estmaton s based on subset of ctzens, whch relates to a specfc age group, and despte potental problems as the gender balance of the populaton mght vary across muncpaltes. Most mportantly, our numbers also seem to outperform prevous measures by the OSCE and the Statstcal Offce of the Federaton. In muncpaltes wth very low brth fgures, the measurements are less accurate, as they seem to under-estmate the sze of the Croat populaton. Whle the fndng that muncpaltes of the Federaton of Bosna and Herzegovna are ether Bosnak- or Croat-domnated mght be lttle surprsng to area specalsts, the estmaton also dentfes the places wth a genunely mult-ethnc face. Examples are the muncpaltes of Glamoć or Bosansk Petrovac, where none of the ethnc groups consttutes a clear majorty of the populaton, and over a dozen of further places, whch are ethncally dverse. Apart from the manyfold applcatons for research, these results can serve as a bass of comparson once data on ethnc 19

20 denttes from the census of Bosna and Herzegovna wll be released. Due to the ncluson of more nuanced questons on ethncty - ncludng proxes such as the use of language and relgon - the 2013 census mght show a more mult-faceted and dversfed pcture of ethnc denttfcaton (Perry, 2013, p. 12; Beber, 2013). 13 On the other hand, the heated poltcal debate about the census mght also leave ts footprnt on the emprcal results (Fredman, 2002; Perry, 2013, p. 4). Overall, the results of the census mght therefore reveal a more dverse and cvc face of the country, where non-recognsed, cvc, or mult-faceted denttes are more vsble. Our data mght be used n studes that test hypotheses about dentty shfts, return processes, regstraton poltcs, and other ethnctyrelated processes n the years between the war, and the frst post-war populaton census. Our method can travel to other cases, where ethnc denttes are estmated ndrectly, based on observatons from a sub-sample of the populaton. References Abrahamse, A. F., Morrson, P. A., & Bolton, N. M. (1994) Surname analyss for estmatng local concentraton of Hspancs and Asans, Populaton Research and Polcy Revew, 13, pp Atken, R. (2007) Cementng dvsons? An assessment of the mpact of nternatonal nterventons and peace-buldng polces on ethnc denttes and dvsons, Polcy Studes, 28 (3), pp Ball, P., & Asher, J. (2002) Statstcs and Slobodan: Usng Data Analyss and Statstcs n the War Crmes Tral of Former Presdent Mlosevc, Chance, 15 (4), pp Bellon, R. (2007) State Buldng and Internatonal Interventon n Bosna (Abngdon, Routledge). Beber, F. (2013). When countng counts. The Bosnan Census (Vol. 2013). Bshop, Y. M. M., Fenberg, S. E., & Holland, P. W. (1975) Dscrete Multvarate Analyss: Theory and Practce (Cambrdge (MA), MIT Press). Bochsler, D. (2013) Radcalzng Electoral System Effects on Support for Natonalst Hardlners, Natons and Natonalsm, 19 (1), pp Braumoeller, B. F. (2006) Explanng Varance; Or, Stuck n a Moment We Can t Get Out Of, Poltcal Analyss, 14, pp Buhaug, H., & Rød, J. K. (2006) Local determnants of Afrcan cvl wars, , Poltcal Geography, 25 (3), pp Burg, S. L., & Shoup, P. S. (2000) The war n Bosna-Herzegovna: ethnc conflct and nternatonal nterventon (Armonk, M.E.Sharpe). Caspersen, N. (2004) Good Fences Make Good Neghbours? A Comparson of Conflct-Regulaton Strateges n Postwar Bosna, Journal of Peace Research, 41 (5), pp See also Lee (2009) and Chandra (2006) on mult-layered and mult-faceted denttes. 20

21 Chandra, K. (2006) What s Ethnc Identty and Does It Matter?, Annual Revew of Poltcal Scence, 9, pp Davdan, M., & Carroll, R. J. (1987) Varance Functon Estmaton, Journal of the Amercan Statstcal Assocaton, 82 (400), pp DellaVgna, S., Enkolopov, R., Mronova, V., Petrova, M., & Zhuravskaya, E. (2014) Cross-border meda and natonalsm, Amercan Economc Journal: Appled Economcs, 6 (3), pp Ellott, M. N., Fremont, A., Morrson, P. A., Pantoja, P., & Lure, N. (2008) A New Method for Estmatng Race/Ethncty and Assocated Dspartes Where Admnstratve Records Lack Self-Reported Race/Ethncty, Health Research and Educatonal Trust, 43 (5), pp Fearon, J. D., & Latn, D. D. (1996) Explanng Interethnc Cooperaton, Amercan Poltcal Scence Revew, 90 (4), pp Federaln zavod za statstku. ( ). Žvorođen prema naconalnoj - etnčkoj prpadnost majke oca. Sarajevo: Federaln zavod za statstku. Federaln zavod za statstku. (2005). Procjene ukupnog broja prsutnh stanovnka, po naconalnoj struktur, stanje Sarajevo: FZS. Federaln zavod za statstku. (2009). Demografska statstka Sarajevo: FzS. Franck, R., & Raner, I. (2012) Does the Leader's Ethncty Matter? Ethnc Favortsm, Educaton, and Health n Sub-Saharan Afrca, Amercan Poltcal Scence Revew, 106 (2), pp Fredman, V. (2002) Observng the Observers: Language, Ethncty, and Power n the 1994 Macedonan Census and Beyond, New Balkan Poltcs, 2002 (3/4). Grofman, B., & Handley, L. (1995) Racal context, the 1968 Wallace vote, and southern presdental dealgnment: evdence from North Carolna and elsewhere. In M. Eagles (Ed), Spatal and Contextual Models n Poltcal Research., pp ((London, Taylor and Francs). Harrs, J. A. (2015) What's n a Name? A Method for Extractng Informaton about Ethncty from Names, Poltcal Analyss, 23 (2), pp Horowtz, D. L. (1985) Ethnc Groups n Conflct (Berkeley, Los Angeles, London, Unversty of Calforna Press). Internatonal Crss Group. (1999). Preventng Mnorty Return n Bosna and Herzegovna (No. 73). Sarajevo: ICG. Internatonal Crss Group. (2003). Buldng Brdges n Mostar. Brussels: Internatonal Crss Group. Kertzer, D. I., & Arel, D. (Eds). (2001) Census and Identty. The Poltcs of Race, Ethncty and Language n Natonal Censuses. Cambrdge, Cambrdge Unversty Press. Lee, T. (2009) Between Socal Theory and Socal Scence Practce. Toward a New Approach to the Survey Measurement of 'Race'. In R. Abdelal, Y. M. Herrera, A. I. Johnston & R. McDermott (Eds), Measurng Identty, pp ((Cambrdge, Cambrdge Unversty Press). Mateos, P., Longley, P. A., & Sullvan, D. O. (2011) Ethncty and Populaton Structure n Personal Namng Networks, PLoS one, 6 (9). McCrudden, C., & O'Leary, B. (2013) Courts and Consocatons. Human Rghts versus Power- Sharng (Oxford, Oxford Unversty Press). McCulloch, A. (2014) Consocatonal settlements n deeply dvded socetes: the lberal-corporate dstncton, Democratzaton, 21 (3), pp Melander, E. (2007). Ethnc Cleansng n Bosna-Herzegovna, , Conference "Dsaggregatng the Study of Cvl War and Transnatonal Volence". Colchester. Mtchell, P., Evans, G., & O Leary, B. (2009) Extremst Outbddng In Ethnc Party Systems Is Not Inevtable: Trbune Partes n Northern Ireland, Poltcal Studes, 57, pp

22 Nagle, J., & Clancy, M.-A. C. (2012) Constructng a shared publc dentty n ethno natonally dvded socetes: comparng consocatonal and transformatonst perspectves, Natons and Natonalsm, 18 (1), pp Nalepa, M. (2012) Reconclaton, Refugee Returns, and the Impact of Internatonal Crmnal Justce: The Case of Bosna and Herzegovna. In M. Wllams, R. Nagy & J. Elster (Eds), NOMOS, Proceedngs of the Amercan Socety for Poltcal and Legal Phlosophy, pp ((New York, New York Unversty Press). Perry, V. (2013). The 2013 Census n Bosna and Herzegovna - A Basc Revew. Sarajevo: Democratzaton Polcy Councl. Pugh, M., & Cobble, M. (2001) Non-Natonalst Votng n Bosnan Muncpal Electons: Implcatons for Democracy and Peacebuldng, Journal of Peace Research, 38 (1), pp Ralegh, C., & Hegre, H. (2009) Populaton sze, concentraton, and cvl war. A geographcally dsaggregated analyss, Poltcal Geography, 28 (4), pp Susewnd, R. (forthcomng) What's n a Name? Probablstc Inference of Relgous Communty from South Asan Names, Feld Methods. Szreter, S., Sholkamy, H., & Dharmalngam, A. (Eds). (2004) Categores and Contexts: Anthropologcal and Hstorcal Studes n Crtcal Demography. Oxford, Oxford Unversty Press. Toal, G., & Dahlman, C. T. (2011) Bosna Remade. Ethnc Cleansng and ts Reversal (Oxford, Oxford Unversty Press). UNAIDS/WHO Workng Group on Global HIVAIDS and STI Survellance. (2003). Estmatng the Sze of Populatons at Rsk for HIV: Issues and Methods: Famly Health Internatonal (FHI). Varshney, A. (2001) Ethnc Conflct and Cvl Socety. Inda and Beyond, World Poltcs, 53, pp Vsoka, G., & Gjevor, E. (2013) Census poltcs and ethncty n the Western Balkans, East European Poltcs, 29 (4), pp WHO. (2003). HIV/AIDS Epdemologcal Survellance Update for the WHO Afrcan Regon Harare: WHO. Wllams, M., & Husk, K. (2013) Can we, should we, measure ethncty?, Internatonal Journal of Socal Research Methodology, 16 (4), pp

23 An ndrect approach to map ethnc denttes n post-conflct socetes Fgures and tables Table 1: Explanatory model for nverted brth rates (1/γ ) n the Federaton BH, Varance model (Lnear regresson wth multplcatve heteroscedastcty) a Varable mputed for 9 muncpaltes. year=2008 year=2009 year=2010 Robust Robust Robust Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Man model (mean) b C / b ** * ** 4.42 b S / b ** ** * b O / b ** ** % employed women populaton densty (log) *.85 GDP per capta (log) populaton (log) returnees (log) a refugees non-returned (log) a ** ** ** 3.12 constant * * Varance model returnees (log) a 1.96 * **.91 refugees non-returned (log) a ** ** ** 1.22 populaton (log) -.89 ** ** **.26 % chldren ** ** ** 6.23 constant ** ** ** 2.78 Number of obs Model χ 2 (12) Prob > χ Pseudo R VWLS R

24 Table 2: Results of relablty and crteron valdty tests Type of test Relablty Crteron valdty Indcator mean correlaton across years correlaton wth OSCE estmates (N=30) correlaton wth FZS estmates (N=77) Bosnaks Croats Serbs nave brthrate full model nave full nave model brth rate model brth rate model model full model

25 Fgure 1: Populaton estmates by ethnc group, 2008, muncpaltes of the Federaton of Bosna and Herzegovna. Full model estmaton. 25

26 Fgure 2: Ethnc map of the Federaton of Bosna and Herzegovna, Full model estmaton. Source map: GD GISDATA d.o.o. Sarajevo (Gauss-Kruger - Zone 6). Data: own estmaton. 26

27 Fgure 3: Estmated share of Croat populaton and vote share for Croat natonalst partes, 2008 muncpal electons. (Full estmaton model) vote for Croat partes vote share populaton share 27

28 Fgure 4: Results of construct valdty tests, explanatory power (adjusted R 2 ) Values close to 1 ndcate hgh construct valdty Fgure 5: Results of construct valdty tests, squared devatons (dev e ) Values close to 0 ndcate hgh construct valdty 28

29 Appendx A: Categorsaton of poltcal partes and electoral lsts Bosnak Croats Bosanskohercegovačka Patrotska Stranka- Sefer Hallovć Bošnjačka Seljačka Stranka BOSS - Bosanska Stranka-Mrnes Ajanovć Demokratska Narodna Zajednca BH Lsta za Čapljnu SBH-BPNS Narodna Bošnjačka Stranka Savez SDA-S BH Socjaldemokratska Unja Bosne Hercegovne Stranka Demokratske Akcje Stranka Demokratske Aktvnost za Evropsku BH A-SDA Stranka za Bosnu Hercegovnu HDZ 1990 HSS-NHI HSP Hrvatska Demokratska Zajednca 1990 Hrvatska Demokratska Zajednca BH Hrvatska Koalcja HSS - NHI HDZ BH HDZ 1990 HNZ Hrvatska Koalcja (HDZ BH HDZ 1990 HKDU BH) Hrvatska Koalcja HDZ 1990 HSS NHI Hrvatska Koalcja HDZ 1990 HSS-NHI Za Žepče Hrvatska Koalcja HDZ 1990-HKDU Hrvatska Koalcja Kantona Sarajevo Hrvatska Koalcja za Drvar HDZ HDZ BH-HSP Ðapć-Juršć Hrvatska Koalcja za Pougarje HSP-Ð-J HDZ1990 HSS-NHI Hrvatska Koalcja za Ravno HDZ1990 HDZBH Hrvatska Koalcja za Travnk Hrvatska Koalcja za Zencu HDZ BH HDZ 1990 HSS-NHI HNZ Hrvatska Koalcja-HDZ BH-HSS-NHI- HDZ 1990 Hrvatska Kršćanska Demokratska Unja BH Hrvatska Narodna Zajednca Hrvatska Stranka Prava Bosne I Serban Mxedethnc Hercegovne - Ðapć Dr. Juršć Hrvatsko Zajednštvo Herceg-Bosne Hsp BH Ðapć-Dr.Juršć-HNZ BH Hss NHI Koalcja Hrvatskh Stranaka HDZ BH HSS-NHI HDZ 1990 HKDU BH Koalcja: HNZ-HDU Koalcjska Lsta HDZ BH HDZ 1990 Demokratska Stranka Naroda Srpske Demokratsk Narodn Savez Demokratsk Pokret Srpske-Depos Koalcja Za Bolj Žvot-DNS DS Narodna Demokratska Stranka Partja Demokratskog Progresa Savez Nezavsnh Socjaldemokrata - SNSD - Mlorad Dodk Savez Srpskh Stranaka Srebrence: SDS- PDP-SP SRS RS SDS-DNS-PDP-SRS RS SNSD-PDP Socjalstčka Partja Srpska Demokratska Stranka Srpska Demokratska Stranka Izvorna - Pokret za Srpsku Srpska Napredna Stranka Republke Srpske Srpska Narodna Radkalna Stranka - Banja Luka Srpska Radkalna Stranka Dr Vojslav Šešelj Srpska Radkalna Stranka Republke Srpske Srpska Stranka Republke Srpske Srpsk Pokret Obnove Republke Srpske Graðanska Demokratska Stranka BH Narodna Stranka Radom za Boljtak Naša Stranka Socjaldemokratska Partja Bosne Hercegovne 29

30 Appendx B: Results of construct valdty tests Construct valdty tests, comparng the populaton estmates wth the 2006 electons, are provded, n order to allow for a better tme match wth the 2005 estmates of the Statstcal Offce of the Federaton (FZS). Therefore, the comparson to the 2006 electons s only ncluded for the models for all 77 muncpaltes, and not for the subset of 30 muncpaltes. Table B1: Results of construct valdty tests, explanatory power (adjusted R 2 ) Values close to 1 ndcate hgh construct valdty. group compared to electons of... FZS estmates Nave Full (2005) model model All 77 muncpaltes Bosnak Croat Serb OSCE estmates Nave model Full model Only 30 muncpaltes Bosnak Croat Serb Table B2: Results of construct valdty tests, squared devaton (dev e ) Values close to 0 ndcate hgh construct valdty. group compared to electons of... FZS estmates Nave Full (2005) model model All 77 muncpaltes Bosnak Croat Serb OSCE estmates Nave model Full model Only 30 muncpaltes Bosnak Croat Serb