TY - JOUR
AU - Gambetta,, Diego
AB - Abstract Using a natural experiment, we find that in provinces where Turkey’s Islamic Justice and Development Party (AKP) just won the election in 2004, women, including those who are weakly religious or non-religious, now veil far more than in provinces in which AKP just lost, the more so the poorer they are. This effect, as we predict, does not occur for praying regularly which is more costly and harder to observe practice. We argue that veiling is higher in AKP provinces not only because of a generic aim to conform to the stricter mores fostered by the victorious party. We find that those who veil, particularly those in AKP provinces who are not pious, are more politically active than those who do not veil. This may be an indication that veiling could partly be a strategic response to policies, which favour those who are or appear pious. Our study suggests that observable religious practices may have their independent dynamics driven by the pursuit of instrumental goals. Our results also suggest that parties with a religious ideology have an advantage over their secular counterparts in solving the clientelistic information problem, for they can rely on religious symbols for screening and signalling. Most religions possess distinctive appurtenances donned by their members—skullcaps, veils, singular hairdos, coloured robes, symbolic jewellery, beards, and many more. Where the use of religious insignia is imposed by authorities—such as women’s veiling in Saudi Arabia—their universal diffusion testifies only of a given religion’s dominance, but it reveals nothing of the frequency of believers since unbelievers too are forced to adopt them. Conversely, where these appurtenances are banned—such as Muslim men’s beards in Tajikistan—their universal absence does not exclude the concealed presence of true believers. Imposition in these matters stunts information (Aksoy and Gambetta, 2016: p. 804). Where people are free to decide on their ‘plumage’, by contrast, the display of these appurtenances can afford reliable information on the presence of religious beliefs: information is perfect if all and only true believers adopt them (and by implication all and only unbelievers do not). This equilibrium, in which people’s outward appearances perfectly match their inner religious beliefs, is sustained by two states of affairs in neither of which there is any tangible benefit to be deemed either religious or non-religious: a universalistic one in which rewards are apportioned regardless of religiosity and a sectarian one in which religious and secular flocks receive comparable rewards by their respective communities. In neither case, one has any reason to appear as religiously other than one truly is. Seldom, however, are these states of affairs perfectly realized. In reality, we conjecture, social and political forces conspire to introduce a mismatch between outfit and beliefs. Particularly, we investigate whether a shift in tangible rewards in favour of religious people, brought about by political change, can foster such a mismatch and turns overt religious appurtenances into covert insignia of political allegiance/loyalty. We do so for one form of religious display, Muslim women’s veiling, and study what happens to veiling when the incentives to piety are increased. We use a Regression Discontinuity Design (RDD, Morgan and Winship, 2007; Gangl, 2010), exploiting marginal electoral wins of Turkey’s Islamic Justice and Development Party (AK Parti). We then validate the RDD analysis with multilevel regressions and a difference-in-difference analysis. Our study offers three contributions. First, contrary to a traditional sociological approach that considers religious practices a mere reflection of religiosity (see e.g. Voas, 2007; Fleischmann and Phalet, 2012; Brünig and Fleischmann, 2015), we add to the studies in the sociology of religion that argue that observable religious practices have their independent dynamics driven by the pursuit of mundane goals—e.g. signalling respectability and finding partners (Patel, 2012), gravitating to ‘respectable’ denominations with an aim to join the elite (Davidson, Pyle, and Reyes, 1995), or even increasing one’s children’s chances to be accepted into church-run schools (Francis and Hutchings, 2013). We uncover how macro political shifts affect religious micro behaviour among those whose religious beliefs are feeble or absent: if piety pays off, women can choose to adopt one of its outward manifestations, veiling. Next, we contribute to the literature on clientelism and political machines (Merton, 1968). The classic problem of clientelistic political parties concerns information: to whom to dispense patronage and to whom to exclude (Hicken, 2011; Stokes et al., 2013). To maximize electoral support, political machines must screen voters and target only those responsive to inducements and exclude others. Simultaneously, voters have an interest to signal to the party machine that they are worthy of patronage. Since some voters could mimic support, the genuine clients are compelled to search for credible signals to distinguish themselves from the mimics. Literature has mostly focused on the screening mechanisms (Hicken, 2011). We take into consideration both signalling and screening. We study the link between voters’ signalling of allegiance and the party’s payoffs from patronage; and argue that religious behaviours and appurtenances have particular properties that make them more efficient for signalling and screening political allegiance than comparable signals associated with secular parties. Third, we contribute to the understanding of political Islam’s success in expanding its power in Turkey. We document a strong positive effect of local Islamic governance on the prevalence of veiling. We explain this effect by the success of the AK Parti and its grassroots organizations in providing social assistance and welfare to create selective incentives for the ‘pious and the poor’ (Meyersson, 2014; Aksoy and Billari, 2018). In this sense, the case of political Islam epitomized by the rise of the AK Parti in Turkey is similar to successful political movements elsewhere, whether religious or ideological: a strong emphasis on mutual aid, charitable giving, and the provision of basic local services in recruiting and retaining followers (Berman, 2011; Blaydes, 2014). We find that veiled women, particularly those who are not pious, are more politically engaged than the unveiled ones—evidence that an upward shift in veiling can be politically rather than religiously driven. Theory and Hypotheses Veiling—it is commonly assumed—reflects a woman’s piety and modesty and hence should track religiosity. As various sociological theories predict (see Ruiter and Van Tubergen, 2009), modernization brings about a decline in religiosity, which should be accompanied by a decline in veiling. There is evidence supporting this prediction. However, it seems to apply only to women who are weakly religious or non-religious. Among highly religious women, the opposite seems to occur, and for them, modernizing forces—education, occupation and higher income, urban living, and contacts with non-Muslims—are associated with an increase in both the frequency and strictness of veiling (Aksoy and Gambetta, 2016). A plausible interpretation of this finding is that modernization raises women’s opportunities to abandon piety, or even just the suspicion that they may be doing so, and hence the need to reassure the community of their modesty: veiling then would represent a strategic response to the threat modernization poses to truly religious women’s reputation, a way for them to signal their resilient piety (Patel, 2012; Carvalho, 2013). The evidence, overall, shows that veiling responds to social circumstances and is not dictated merely by religious rulings or concerns; but it also shows that veiling remains consistent with religiosity or the lack of it. Here, we take a step further and investigate whether women’s veiling also has a strategic component independent of religiosity: are there conditions under which even weakly religious or non-religious women decide to veil? We conceive of two types of motives that could induce feebly religious women to veil voluntarily. One is to conform to prevailing mores and the other is to persuade specific observers of their piety or some other trait closely related to piety. The former is created by social pressures that produce a conformism effect. If veiling among the faithful spread, it becomes progressively harder not to conform even for women whose religiosity is weak or absent, and even in the absence of compulsory measures—for as simple a reason as not wanting to stick out and raise questions about their respectability or receive social approval from the community. Veiling in this case does not aim at a specific audience, but rather to avoid negative consequences from the community. Donning the veil despite being weakly religious due to community pressure is one instance of ‘preference falsification’, a phenomenon first studied by Kuran (1995: pp. 8–9). The second type of motives is signalling. Signalling motives are supported by the belief that piety matters crucially for those who can choose whether to bestow benefits on a woman, her family, or her wider ingroup and community. In this case, veiling is adopted strategically, as a signal of piety aimed at particular receivers, such as potential partners (Patel, 2012), probable employers, or religious and political organizations, which may offer welfare assistance and jobs to a women or members of her family or her wider ingroup and community (Aksoy, 2017). Unlike the conformism effect, the signalling effect is not frequency dependent—it does not necessarily become stronger as the frequency of veiled women increases. The two motives, community pressure and signalling, can coexist and reinforce each other. We believe that the signalling motives are rather strong in our context. To explain why this is so, we need to shift our attention to politics. ‘Under clientelism, parties distribute benefits to individuals and attempt to hold them accountable for their votes. (…) Parties need to know (…) who will vote for them come hell or high water, who will not vote for them come hell or high water, and who is on the fence’ (Stokes et al., 2013: p. 19). Past research has shown that parties rely on several mechanisms for screening. For example, repeated interactions with the client and social networks of the party and local brokers provide valuable information on the reliability of the client (Hicken, 2011). Religious parties may have an advantage over secular ones as they can exploit religious networks (churches, mosques, associations) and brokers (priests, imams) for screening. Indeed, Kalyvas (1996) argues that the Catholic Church’s extensive network of active clergy was a mechanism that contributed to the success of Christian Democracy in Europe. The clientelistic information problem also has a client side. Voters eager to benefit from patronage want to signal their loyalty to the victor ( Corstange, 2016: p. 5). Those with a history of supporting the winning party and connected with the party apparatus should have no difficulties to be identified as genuine. Yet, voters who want to benefit from patronage but whose allegiance is opaque need to invest resources to make their support manifest and credible. The incentive to do so is greater if the party’s patronage is expected to strongly favour its supporters, actual or potential, at the expense of the rest. At a minimum, these voters express their newly acquired political loyalties by publicly displaying the victors’ insignia, e.g. brown shirts for the Nazis, black shirts for the Fascists, green shirts for the Italian Northern League, red bonnets made fashionable by Hugo Chavez, hammer and sickle pins, or fascist Roman salutes. Not displaying the insignia of the victor party can prove costly: it sends the message that one is indifferent or hostile to the victor despite the benefits of jumping on the wagon—whether because the expected benefits are too low to bother or the conflicting preferences too strong. It is a credible signal of opposition. The reverse, however, is not as convincing a signal of support: there is no guarantee that an individual who displays the victor’s insignia voted for the victor or will do so in the future. Donning a red bonnet or making a Roman salute is a near-costless act, which opportunistic mimics too can afford. Before distributing resources, a secular party requires more reliable signals of voters’ allegiance. Parties with a religious connotation and their clients have a distinctive advantage in this regard. In addition to the methods available to secular parties, religious parties and their voters can rely on a set of well-established religious signals. This set, which is greater than the set at the disposal of secular parties and their voters, has two further advantages. First, truly religious people incur no extra cost to signal their preferences, as these transpire from what they naturally do. If voters’ religiosity is a strong indication that they support the party that best takes care of their interests and beliefs, then these behaviours also obliquely signal political support. Many religious practices, which have intrinsically high costs—think of long-term fasting, celibacy, self-flagellation, bodily modifications, generously giving to charity, attending regularly religious functions, and more—can work in this way, for only true believers are prepared to afford them. If having religious beliefs is associated to the support of a religious political party, then performing any of the above acts would credibly signal not just one’s true religious beliefs but indirectly one’s political preferences too. However, political parties with a religious connotation cannot pick their supporters on these very costly religious practices. Parties seek to obtain popular majority and hence cannot afford to appeal only to the extreme fringes of religiosity. Second, although voters who are feebly or non-religious have to pay a cost to pretend religiosity, the set of signals at the religious parties’ disposal is likely to contain also some signal that is costly enough to be persuasive but not exorbitantly so. Veiling, we argue, is just such a signal. According to the signalling theory, if a receiver fears to be deceived by a signaller, that is by signallers who display evidence of having a certain quality when this is untrue, honest signallers need to undertake an action the cost of which is such as to separate them from dishonest signallers, the mimics, for it is unaffordable by them. To be convincing, this action must be affordable only or nearly only by those who are signalling honestly but not by mimics (Spence, 1974; Zahavi and Zahavi, 1998; Gambetta, 2009). Veiling does not have high intrinsic costs, it can be bought inexpensively and wearing it does not impose forbidding costs for weak or unbelievers either: it is rather a nuisance for the non-religious, a constraint on vanity and freedom of bodily movement.1 This is not to dismiss that veiling could be highly costly depending on the context. In Europe, for example veiling attracts discrimination (Aksoy and Gambetta, 2016). In secular Turkey too, before the AK Parti reign began, veiled women were under constant state pressure not to veil (Somer, 2017). In such high cost environments, veiling clearly signals religiosity and could be used as a reliable signal of support by Islamic political parties. But why even in relatively low-cost environments, such as in the late AK Parti reign, veiling could be an effective signal of political allegiance? It is not difficult to grasp why clients who want to signal allegiance to the religious party would prefer veiling to more costly (or less visible) practices, such as engaging daily in regular prayers; but why should veiling persuade politicians of women’s loyalty to the party even external pressures for the veiled are not very high? Even those lacking the corresponding beliefs may find it expedient to afford to veil and mimic political preferences they do not have. There is however a difference between secular political insignia and veiling, which we may call viscosity: political insignia are typically cheap to acquire—parties themselves often donate them to supporters—and can be put on or taken off as the occasion demands; veiling, by contrast, even if inexpensive to produce and buy, can be expensive to display as to be credible qua religious (and hence political) signal it requires consistency. In this regard, veiling is less costly than many religious practices, but it is more costly than simple political insignia for taking it off would reveal the falsification—once a woman is seen veiling, she cannot, except in anonymous environments, unveil publicly without raising questions about her true piety. Veiling carries a public commitment, and this affords veiling some persuasive weight. Hence, while by itself may be a weak signal of genuine religiosity—it may fail to persuade an imam or a pernickety pious would-be spouse—it is just reliable enough as a signal of loyalty to a political movement, more so than secular insignia could be. To summarize, we argue that veiling is a religious practice that can work well as a screening device in a political context. On the one hand, due to its viscosity, veiling is costly enough to deter those who would only pretend to be prepared to vote for a religious political party from using it. On the other hand, by not being too costly, veiling can become widely adopted. In addition, unlike other practices, such as praying or charitable donations, veiling is easier to observe. Thus, screening clients on veiling is efficient and gives political parties the chance to still pursue a popular majority, which would be denied if they screened on more expensive signals. Our theoretical framework has the potential to be applied to other behaviours that can work as political signals—for men, one can think of beards and regular mosque attendance, particularly to the Friday prayer. Data on these other religious practices however are, regrettably, scarce. To check the solidity of our reasoning, we develop a formal model demonstrating how moderately costly and highly visible practices work best as signals of loyalty to religious parties (Supplementary data A). Our theoretical framework has some ancestors that deserve to be mentioned. In Iannaccone’s (1992) religious club, good model costly religious acts work as signals of commitment to the religious organization and the organization uses those signals to screen out less committed members and free riders. The main difference of this model with ours is that religious organizations (e.g. sects and communities) can require more stringent signals from their members, for a certain number of committed members would suffice for the community to thrive. Political parties, by contrast, need to solve the screening problem of identifying loyal clients while at the same time maximizing popular votes. This makes the free-rider problem experienced by political parties different from the one local religious community experience. Chen (2010) builds on Iannaccone’s model and shows that economic shocks increase the intensity of religious involvement. Chen argues that religious involvement works as social insurance against hardship: the pious who are in need can turn to their religious community for help. Similar to ours, Chen’s work too demonstrates the important role social insurance plays in religious involvement. In our case, however, we have a more tangible insurance provider than the ‘community’, namely the local authorities and the AK Parti. Carvalho (2013) proposes a formal model of veiling, which is tested empirically by Aksoy and Gambetta (2016). In Carvalho (2013), veiling is modelled as a commitment device rather than signalling. Women may veil to bind themselves so that they are unable to misbehave even if they were inclined to. In this sense, commitment is an inward-looking self-control motive, although it can still inform the community of the piety of the women. Finally, the social psychological literature distinguishes between extrinsic and intrinsic religious orientations (Allport and Ross, 1967). Those with the former use religion for attaining instrumental goals (status, security, power, self-justification, etc.), while those with the latter ‘live’ the religion. Signalling motives in our case can be seen as a subset of extrinsic motives. In contrast with a social psychological approach, here, we test whether a shift in external macro conditions increases the incentives for extrinsic religiosity. Hypotheses By applying these insights on political signalling using religious behaviour to Turkey’s Islamist AK Parti, we derive a set of hypotheses. We will describe in detail the Turkish context and the AK Parti politics after formulating our hypotheses. We predict that: H1: The likelihood of veiling is higher in areas where the local authority is under the AK Parti’s control. This does not imply that religiosity itself increases under the AK Parti. In fact, we predict from a signalling perspective that that less visible or more costly religious practices such as performing the regular daily prayers (namaz) are unaffected by the AK Parti governance. H2: The effect of the AK Parti predicted in H1 does not extend to more costly or less observable religious practices, such as regular praying. Note that the payoffs for the client, such as access to jobs, welfare, and social assistance, should be more beneficial for the poor (Calvo and Murillo, 2004). Hence, we expect that: H3: The effect of the AK Parti on veiling predicted in H1 are stronger the poorer the women are These hypotheses are about the client’s signalling behaviour. We further predict that the AK Parti selects on religious signals in distributing the patronage premium. Local-level data on the party’s patronage decisions are scarce, but our datasets nevertheless include one measure that can reveal AK Parti’s patronage: access to free health care. We will describe Turkey’s healthcare and local welfare provision system in the next section. In short, we expect to find the AK Parti to be more efficient than its secular counterparts in providing local welfare. This is because the central and local AK Parti governments work in tandem, the central government distributes resources strategically to its local municipalities, and the local welfare system the AK Parti has developed relies on religious charity motives, which give the AK Parti an edge over secular parties. We thus expect that: H4: The likelihood of having the general health insurance versus no insurance is higher in areas where the local authority is under the AK Parti’s control. Our model also predicts that a religious signal such as veiling increases the chances of access to patronage in AK Parti municipalities. H5: Veiling is positively associated with access to free healthcare in AK Parti municipalities. We drive the hypotheses above from the signalling perspective. After presenting the tests of these hypotheses, we will provide additional analyses using which we will try to disentangle the signalling motives from community pressure and conformism. Let us now describe in detail the Turkish context and the evolution of AK Parti’s clientelistic local welfare system. Evolution of Clientelism in Turkey under AK Parti Since Atatürk founded it in 1923, modern Turkey had been secular. Atatürk set a modernizing agenda that emphasized economic development, Westernization, and secularization in all domains of social life. This ‘Kemalist’ vision, aimed at relieving the individual from the pressures of tradition and religion (Mardin, 2010), was embraced by urban denizens while most of the population remained marginalized, poor, and indeed pious (Yavuz, 2000). Practicing Muslims found themselves discriminated against and rejected by the ruling elite (Somer, 2017). The rise of the AK Parti (Justice and Development Party) in 2002 marked an end to state-controlled secularization. The AK Parti represents continuity in political Islam, but with the scale of its power it signifies a paradigm shift (Aksoy, 2016). Under Recep Tayyip Erdoğan’s leadership, the AK Parti gained a parliamentary majority in 2002 and captured most Turkey’s municipalities in its first local elections in 2004. Since then, the AK Parti has consolidated its political power. Change however did not happen immediately. Until around 2010, many believed the AK Parti to be a modernizing force that while giving a voice to the marginalized poor and pious, it also promoted democratization and prosperity. Indeed, the early terms in power of the AK Parti were marked by rapid economic development (Rodrik, 2012). Gradually, but more forcefully after the recession of 2009, the AK Parti changed course boosting the presence of Islam in public life and curtailing Turkey secular tradition (Kuran, 2018): it reversed the ban on veiling at public universities and offices, introduced restrictions on alcohol, erected new Mosques, supported religious vocational schools (İmam-Hatip), and increased 4-fold the budget of Turkish Presidency of Religious Affairs (TPRA). The causal force we expect to affect veiling came into full realization 6 years after the elections of 2004. The effect of the recession on the AK Parti became apparent in the 2009 elections in which the party lost more of its mayoralties than it won new ones, and its vote fell by three points to 39 per cent. After these warning signs, the party increased its Islamic tone and policies and succeeded in overturning the effects of the recession: in 2010 it won a remarkable 58 per cent of all votes in a constitutional referendum and, in the 2011 general elections, it obtained a comfortable majority in parliament for the third time after receiving a whopping 49.83 per cent, almost twice the vote share of the runner up (CHP with 25.98 per cent). We thus expect to find evidence for H1 after 2009, when the ‘Islamization’ has taken pace. The Islamization of the AK Parti is well-captured by two trends. First trend is the growth of state-funded religious vocational schools (İmam-Hatip): until 2010, the number of Imam-Hatips was stable at around 450; from 2010 to 2012, they increased to 537; and then shot up to 1,452 by 2017 (Figure 1). The students in these schools increased by about 11,700 annually until 2009, but between 2009 and 2015, the yearly increase climbed up to 67,000 students, a nearly 6-fold increase. The Education and Science Workers’ Union (Eğitim-Sen) explains this dramatic increase as the result of the generous funding of the Ministry of Education and the conversion of secular schools into religious schools by the government.2 Figure 1. Open in new tabDownload slide Upper panel: number of mosques and religious personnel by year. Lower panel: annual number of state-funded religious high schools (İmam-Hatip) and students in these schools Data : Turkish Presidency of Religious Affairs (upper panel) and Education and Science Worker’s Union (Eğitim-Sen) (lower panel). Figure 1. Open in new tabDownload slide Upper panel: number of mosques and religious personnel by year. Lower panel: annual number of state-funded religious high schools (İmam-Hatip) and students in these schools Data : Turkish Presidency of Religious Affairs (upper panel) and Education and Science Worker’s Union (Eğitim-Sen) (lower panel). Second trend is the growth after 2005 (Figure 1) of Mosques and religious personnel—administrative and service staff at mosques, Imams, müezzins, and Quran course instructors: until 2010, the number of Mosques grew by about 700 per year. Between 2011 and 2012 however, as many as 2,000 new Mosques opened (from 82,693 in 2011 to 84,684 in 2012). According to TPRA, the size of religious personnel, stable until 2010, increased sharply between 2010 and 2013 (84,000–121,000). After year 2000, the number of female preachers employed by TPRA increased by a staggering 713 per cent and that of female Quran course instructors increased 4-fold.3 Although the gender breakdown of the change in religious personnel numbers in Figure 1 is unavailable, it is plausible that women benefitted more than men from this increase in religious employment opportunities. We argue that the ‘Islamization’ of the country increased the payoffs for the pious, either directly as access to newly available religious personnel jobs and local welfare provision or indirectly through social approval by the religious community. Indeed, since rising to power, the AK Parti has launched a capillary charity-based welfare system (Buğra and Keyder, 2006). In this system, the poor are assisted with cash, food and fuel, or social security cover, using the funds raised by municipalities and local organizations. Religious motivations are at the centre. NGOs in welfare (such as the infamous Deniz Feneri4) use Islamic values as to foster charitable giving (Buğra and Keyder, 2006). Province- and district-level governments play a vital role in this system (Metin, 2011). The ‘Social Assistance and Solidarity Foundations’ organized at the provincial and district levels are responsible for the distribution of social assistance. Province and district mayors often sit on the board of these foundations. This system gives mayors substantial discretion, which offers ample opportunities for clientelism (Metin, 2011). For instance, Umit (2018) analysed the distribution of Istanbul’s budget across its provinces and found that the investment plan for 2018 does indeed privilege the AK Parti districts of Istanbul, which receive around 250 million Turkish liras more than non-AK Parti districts ceteris paribus. The social welfare system is organized locally, but the central AK Parti government too invests strategically. Analysing government’s spending across the provinces of Turkey from 2003 to 2014, Cammett, Luca, and Sergenti (2018) find that the central government systematically used the allocation of social and economic budget expenditures to channel resources (e.g. conditional cash transfers) to hotly contested districts to attract new voters to the AK Parti. Given the increasing power of the AK Parti and the expansion of local welfare provision, the incentives to become a client of the AK Parti or pro-AK Parti organizations also increased. As the pro-government Confederation of Public Servant Trade Unions (Memur-Sen) proudly announced, its membership shot up by almost 18-fold from 47,871 in 2002 to 836,505 in 2015.5 It is important to describe Turkey’s General Social Security (GSS), also known as the ‘green card’ in detail. GSS is only one form of social assistance provided by local authorities. AK Parti’s social assistance programme comprises many elements such as conditional cash transfers, prenatal training, housing, and food aid. Detailed data on these forms of local welfare are unavailable (Buğra and Keyder, 2006); we therefore have to rely on GSS not only as a measure of free health care but as an indirect indicator of many other forms of welfare that can be dispensed clientelistically. The green card programme was started in 1992 to provide free health care to poor citizens. In 2003, AK Parti expended it so that citizens lacking other forms of social security and earning less than a certain threshold became eligible for free hospital services and prescription medicines. Importantly, green card applications need to be approved by a local administrative council, which comprises the local governor and other officials appointed by the central government. The council is required to gather information on applicants’ income. The council also aims to collect information on the applicant’s informal sources of income. How far the council pushes this scrutiny is rather flexible. Yoltar’s (2009) ethnographic study shows that in deciding eligibility green card officers can resort to as vague criteria as being a ‘fine citizen deserving of the state’s compassion’. The findings of Kemahlıoğlu and Bayer (2020), who have analysed the diffusion of green cards across Turkey, lend clear support to our H4. Using data from the Ministry of Health, which differ from our data, they conclude ‘that municipalities whose mayors are affiliated with the national incumbent party, the AKP, receive more green cards because local government resources are used to build stronger local party organizations, which then help reach citizens that the incumbent party wants to target’ (p. 2) and that ‘local party organizations, which are stronger in AKP-controlled municipalities, helped the central government bureaucracy identify such voters, aided them in applying for benefits, and, hence, increased the number of green cards in areas controlled by AKP co-partisans’ (p. 4). In short, local organizations and authorities have large discretion on whom to deny the green card and to whom to dispense it, either by encouraging individuals to apply or by taking advantage of flexible eligibility criteria. This enables vast opportunities for clientelism. To test our hypotheses, we exploit a natural experiment made possible by the first local (mayoral) elections of the AK Parti in 2004. In 2004, 20 parties and independent candidates competed for 81 mayoralties at the province centres (İl), about 1,000 mayoralties at the lower-level district centres (İlçe), and about 2,000 lowest-level offices. Mayors are elected by a first-past-the-post system. This enables us to compare provinces, which were narrowly won by the AK Parti with those narrowly won by a non-Islamist party. In 2004, the AK Parti obtained 42 percent of all votes, followed by the centre left secular CHP (Republican People’s Party) and the ultra-nationalist right-wing MHP (Nationalist Movement Party), which secured, respectively, 18 and 10 percent of all votes. There were two other Islamist parties, Saadet (Felicity Party) and BBP (Great Union Party), but Saadet and BBP received only about 4 and 0.4 percent of the votes respectively and neither could win any province-level seats. Consequently, we will use AK Parti rule and Islamic governance interchangeably. While we focus on the 2004 mayoral elections, as they were AK Parti’s first, we will also analyse how the results of the 2009 mayoral elections affected our findings. Data, Variables, and Descriptive Statistics We match three surveys—Turkey’s 2003, 2008, and 2013 Demographic and Health Surveys (henceforth TDHS03, TDHS08, and TDHS13)—with the election results. All three surveys provide a representative sample of Turkish households, but only women at reproductive ages (15–49 years) were interviewed. In 2003 and 2008, only women who were ever married whereas in 2013 all women were interviewed. The sample sizes are 8,075, 7,405, and 9,746 in 2003, 2008, and 2013, respectively. The response rates were consistently above 90 per cent.6 The surveys report the province of the respondent but not lower-level geographical units. Consequently, we could match electoral results with respondents at the provincial level. Veiling is a binary variable: whether the respondent covers her head regularly outside the home. We also use a measure of whether the respondent performs the five-times-a-day prayer rituals (Namaz), a less observable and more costly religious behaviour. Veiling is measured in all waves, prayer in 2008 and 2013 only. The key independent variable is the AK Parti’s win/lose margin in a province in 2004, the difference between the vote share of the AK Parti and that of the largest remaining party in the province main centre or the metropolitan municipality. In some analyses, we will use an ‘Islamic District Ratio’, which is the proportion of lower-level district municipalities in a province won by the AK Parti in 2004. We further use two electoral variables (from TurkStat) for robustness checks: the provincial level vote share of the Virtue Party (Fazilet), which is the predecessor of the AK Parti in 1999 to measure the local support for political Islam before 2004 and the 2009 mayoral election results. As controls, we use age and education (in years), occupational status (employed or not), marital status (married, divorced, single), ethnicity per respondent’s mother tongue (Turkish, Kurdish, Arabic, and other), urban or rural residence, the ideal number of children a woman wants, a household wealth calculated by the survey designers, and a composite indicator that captures gender norms and values (see Supplementary data D). We also use the type of healthcare coverage (or no coverage) of the respondent. Table 1 presents descriptive statistics for the variables. In 2003, 75 per cent and in 2008 76 per cent of ever married women cover their head when outside the home, whereas in 2013, 73 per cent of ever married women do so. However, in 2013, ever married women are on average more educated and more likely to be employed than in 2008/2003. When all those factors are accounted, the difference between 2008 and 2013 (difference = 0.01, z = 1.02, P = 0.311) and between 2003 and 2013 (difference = 0.02, z = 1.43, P = 0.158) disappears. Never married women sampled in 2013 are much less likely to veil (41 per cent) than ever married women. Table 1. Descriptive statistics for key variables (NA = Not Available) . 2003 Ever married (N = 8,075) . 2008 Ever married (N = 7,405) . 2013 Ever married (N = 7,219) . 2013 Never married (N = 2,526) . Mean . SD . Mean . SD . Mean . SD . Mean . SD . Veil 0.75 0.43 0.76 0.43 0.73 0.45 0.41 0.49 Prayer NA NA 0.50 0.50 0.54 0.50 0.35 0.48 Age (years) 33.70 8.41 34.08 8.39 34.81 8.06 20.84 6.28 Education (years) 5.39 4.03 5.62 3.94 6.46 4.18 9.85 3.34 Married 0.95 0.21 0.95 0.22 0.95 0.22 NA NA Divorced 0.02 0.13 0.03 0.17 0.04 0.19 NA NA Widowed 0.02 0.15 0.02 0.14 0.02 0.13 NA NA Turk 0.77 0.42 0.77 0.42 0.77 0.42 0.76 0.43 Kurd 0.19 0.39 0.20 0.40 0.19 0.39 0.21 0.41 Arab 0.02 0.15 0.02 0.15 0.03 0.16 0.02 0.14 Other ethnicity 0.00 0.00 0.01 0.09 0.01 0.11 0.01 0.09 Ideal number of children 2.61 1.18 2.62 1.25 2.90 1.31 2.40 1.08 Urban 0.72 0.45 0.73 0.44 0.73 0.44 0.74 0.44 Wealth (z-scr) 0.00 1.00 0.00 1.00 −0.01 1.01 0.02 0.96 Work 0.26 0.44 0.56 0.50 0.58 0.49 0.39 0.49 Values (z-scr) 0.00 1.00 0.00 1.00 0.05 1.00 −0.14 0.96 GSS coverage 0.10 0.30 0.18 0.39 0.13 0.34 0.15 0.36 No insurance 0.32 0.47 0.16 0.37 0.12 0.32 0.10 0.30 . 2003 Ever married (N = 8,075) . 2008 Ever married (N = 7,405) . 2013 Ever married (N = 7,219) . 2013 Never married (N = 2,526) . Mean . SD . Mean . SD . Mean . SD . Mean . SD . Veil 0.75 0.43 0.76 0.43 0.73 0.45 0.41 0.49 Prayer NA NA 0.50 0.50 0.54 0.50 0.35 0.48 Age (years) 33.70 8.41 34.08 8.39 34.81 8.06 20.84 6.28 Education (years) 5.39 4.03 5.62 3.94 6.46 4.18 9.85 3.34 Married 0.95 0.21 0.95 0.22 0.95 0.22 NA NA Divorced 0.02 0.13 0.03 0.17 0.04 0.19 NA NA Widowed 0.02 0.15 0.02 0.14 0.02 0.13 NA NA Turk 0.77 0.42 0.77 0.42 0.77 0.42 0.76 0.43 Kurd 0.19 0.39 0.20 0.40 0.19 0.39 0.21 0.41 Arab 0.02 0.15 0.02 0.15 0.03 0.16 0.02 0.14 Other ethnicity 0.00 0.00 0.01 0.09 0.01 0.11 0.01 0.09 Ideal number of children 2.61 1.18 2.62 1.25 2.90 1.31 2.40 1.08 Urban 0.72 0.45 0.73 0.44 0.73 0.44 0.74 0.44 Wealth (z-scr) 0.00 1.00 0.00 1.00 −0.01 1.01 0.02 0.96 Work 0.26 0.44 0.56 0.50 0.58 0.49 0.39 0.49 Values (z-scr) 0.00 1.00 0.00 1.00 0.05 1.00 −0.14 0.96 GSS coverage 0.10 0.30 0.18 0.39 0.13 0.34 0.15 0.36 No insurance 0.32 0.47 0.16 0.37 0.12 0.32 0.10 0.30 Open in new tab Table 1. Descriptive statistics for key variables (NA = Not Available) . 2003 Ever married (N = 8,075) . 2008 Ever married (N = 7,405) . 2013 Ever married (N = 7,219) . 2013 Never married (N = 2,526) . Mean . SD . Mean . SD . Mean . SD . Mean . SD . Veil 0.75 0.43 0.76 0.43 0.73 0.45 0.41 0.49 Prayer NA NA 0.50 0.50 0.54 0.50 0.35 0.48 Age (years) 33.70 8.41 34.08 8.39 34.81 8.06 20.84 6.28 Education (years) 5.39 4.03 5.62 3.94 6.46 4.18 9.85 3.34 Married 0.95 0.21 0.95 0.22 0.95 0.22 NA NA Divorced 0.02 0.13 0.03 0.17 0.04 0.19 NA NA Widowed 0.02 0.15 0.02 0.14 0.02 0.13 NA NA Turk 0.77 0.42 0.77 0.42 0.77 0.42 0.76 0.43 Kurd 0.19 0.39 0.20 0.40 0.19 0.39 0.21 0.41 Arab 0.02 0.15 0.02 0.15 0.03 0.16 0.02 0.14 Other ethnicity 0.00 0.00 0.01 0.09 0.01 0.11 0.01 0.09 Ideal number of children 2.61 1.18 2.62 1.25 2.90 1.31 2.40 1.08 Urban 0.72 0.45 0.73 0.44 0.73 0.44 0.74 0.44 Wealth (z-scr) 0.00 1.00 0.00 1.00 −0.01 1.01 0.02 0.96 Work 0.26 0.44 0.56 0.50 0.58 0.49 0.39 0.49 Values (z-scr) 0.00 1.00 0.00 1.00 0.05 1.00 −0.14 0.96 GSS coverage 0.10 0.30 0.18 0.39 0.13 0.34 0.15 0.36 No insurance 0.32 0.47 0.16 0.37 0.12 0.32 0.10 0.30 . 2003 Ever married (N = 8,075) . 2008 Ever married (N = 7,405) . 2013 Ever married (N = 7,219) . 2013 Never married (N = 2,526) . Mean . SD . Mean . SD . Mean . SD . Mean . SD . Veil 0.75 0.43 0.76 0.43 0.73 0.45 0.41 0.49 Prayer NA NA 0.50 0.50 0.54 0.50 0.35 0.48 Age (years) 33.70 8.41 34.08 8.39 34.81 8.06 20.84 6.28 Education (years) 5.39 4.03 5.62 3.94 6.46 4.18 9.85 3.34 Married 0.95 0.21 0.95 0.22 0.95 0.22 NA NA Divorced 0.02 0.13 0.03 0.17 0.04 0.19 NA NA Widowed 0.02 0.15 0.02 0.14 0.02 0.13 NA NA Turk 0.77 0.42 0.77 0.42 0.77 0.42 0.76 0.43 Kurd 0.19 0.39 0.20 0.40 0.19 0.39 0.21 0.41 Arab 0.02 0.15 0.02 0.15 0.03 0.16 0.02 0.14 Other ethnicity 0.00 0.00 0.01 0.09 0.01 0.11 0.01 0.09 Ideal number of children 2.61 1.18 2.62 1.25 2.90 1.31 2.40 1.08 Urban 0.72 0.45 0.73 0.44 0.73 0.44 0.74 0.44 Wealth (z-scr) 0.00 1.00 0.00 1.00 −0.01 1.01 0.02 0.96 Work 0.26 0.44 0.56 0.50 0.58 0.49 0.39 0.49 Values (z-scr) 0.00 1.00 0.00 1.00 0.05 1.00 −0.14 0.96 GSS coverage 0.10 0.30 0.18 0.39 0.13 0.34 0.15 0.36 No insurance 0.32 0.47 0.16 0.37 0.12 0.32 0.10 0.30 Open in new tab Figure 2 shows the distribution of the 2004 AK Parti win/lose margin and veiling across the 81 provinces. A total of 58/81 provinces are won by the AK Parti in 2004. Also, in 59 provinces, the AK Parti either won or lost by a wide absolute margin. The RDD is based on the provinces (N = 22) where the win/lose margin is low. The multilevel regression and difference-in-difference analysis are based on all provinces. Figure 2. Open in new tabDownload slide Distribution of the AK Parti win/lose margin in 2004 and veiling across the 81 provinces of Turkey Figure 2. Open in new tabDownload slide Distribution of the AK Parti win/lose margin in 2004 and veiling across the 81 provinces of Turkey Analysis Strategy The majoritarian nature of the local elections in Turkey helps us use a strong design, known as RDD (Thistlethwaite and Campbell, 1960; de la Cuesta and Imai, 2016). In this design, a treatment t is assigned to cases based on a forcing or running variable x such that, when x is larger than a cut-off value c, the observation receives the treatment (t = 1) otherwise control (t = 0). In our application, x is the AK Parti’s win/lose margin, c = 0, and t is whether the province centre is ruled by the AK Parti after the 2004 elections. Moreover, only cases within h bandwidth around c are included in the analysis. Our RDD can be written formally: Ey=α0+α1x+α2xt+ βt with t= 1 if x≥00 if x<0 for -h<x<h.(1) In equation (1), β captures the effect of treatment, namely AK Parti governance. The α1 and α2 represent the control function of the running variable x (win/lose margin) with a local linear specification, recommended by de la Cuesta and Imai (2016), which can vary below and above the cut-off. The equation is fitted using cases within the h bandwidth around the zero cut-off.7 We use the method proposed by Imbens and Kalyanaraman (2012) to calculate an optimum bandwidth value ĥ and use it in our main analyses. Below we present robustness checks with alternative bandwidths (Figure 6). RDD is often characterized as a natural experiment (Gangl, 2010; Titiunik, 2020). Titiunik (2020) formally defines natural experiments as designs that have the following three properties: ‘the treatment assignment mechanism is neither designed nor implemented by the researcher, unknown to the researcher, and probabilistic by virtue of depending on an external factor’ (p. 2). Titiunik (2020) argues that RDD is a natural experiment as it satisfies all three conditions. In our case, for example the assignment of an AK Parti governance to provinces is out of our control (condition 1). The second condition is satisfied in RDD too, for, while we know the assignment mechanism as x > c implies that t = 1 in equation (1), this is only known conditional on x. The probability distribution of x is essentially unknown to us, for there is always an element of chance in the exact vote shares of political parties. Finally, in the small window h around the cut-off c, it is plausible to think that treatment assignment is probabilistic due to electoral noise, unpredictability of x, and the sharp RD rule that when x > 0, t = 1. In large-scale popular elections, there is much unsystematic ‘noise’—such as accident, illness, weather conditions, and any other random factor that affects the turnout among a few voters. These factors are thought to be sufficient to shift the result of razor-sharp election in either direction (Eggers et al., 2015). Hence, the assumption is that potential outcomes of all confounders are continuous at the cut-off c, except t. If this assumption is correct, a comparison of the provinces around the threshold will reveal the ‘causal’ effect of the AK Parti rule on the outcome variable y. It is possible to test the above assumption, first by checking whether the pre-treatment variables jump abruptly at the threshold [i.e. by replacing y in equation (1) by pre-treatment variables]. Significant differences would invalidate the RDD. Next, McCrary (2008) proposed a density test that does not require any pre-treatment variable. Sorting around the cut-off, that is some cases being more likely to be just above the threshold, would also invalidate an RDD. But if there is sorting, one would expect to see a discontinuity in the density of x. McCrary’s test checks density around the cut-off. We carried out both tests (see Supplementary data B). Importantly, as a placebo test, we find that the 2004 elections had no ‘retro-effect’ as it was on veiling prevalence in 2003. While one can test the underlying assumptions of RDD and disprove them, it is not possible to prove definitively that those assumptions are met. Hence, we validate the main RDD results in two additional ways. First, we use multilevel regressions with random effects for provinces and households and regress veiling on ‘Islamic district ratio’, proportion of districts won by the AK Parti within a province. Second, we use a difference-in-difference analysis testing how veiling changes in a province from 2003 to 2013 and whether this change is different between AK Parti versus non-AK Parti provinces. In both multilevel and difference-in-difference models, we control for various key covariates. We are not alone in applying an RDD to Turkish elections. The pathbreaking study was by Meyersson (2014), who used the 1994 elections and found that Islamist mayors facilitate female educational attainment. Meyersson explains this finding by arguing that Islamist mayors lower the educational entry barriers women experience by, for instance, stretching the veil bans in schools and providing educational facilities to the conservative and poor citizens. Aksoy and Billari (2018) too apply an RDD and show that a local AK Parti rule increases marriage rates and fertility. They explain this by AK Parti’s efficiency in providing social welfare and assistance at the local level. This local social welfare system of the AK Parti is key in our theoretical framework. Finally, Çörekçioğlu (2018) shows using an RDD that the AK Parti municipalities hired more female employees after the repeal of the veil ban. Results Using a combination of the RDD detailed above and more conventional regression models, we test our hypotheses. We use linear probability models, unless stated otherwise. Results with binary logistic regressions for binary outcomes are qualitatively the same.8 Starting from H1, we find that in 2008 an Islamic governance in 2004 has a small and statistically insignificant positive effect on the likelihood of veiling (β [marginal effect] = 0.03, province-level cluster robust S.E. = 0.12, P = 0.787 for the optimal bandwidth of 0.17), but in 2013 it does have a very strong positive effect in 2013 (β [marginal effect] = 0.25, province-level cluster robust S.E. = 0.07, P = 0.001 for the optimal bandwidth of 0.087). See Figure 3 for a graphical representation of those effects. This is consistent with a slowly growing effect picking up speed after 2010 when religion came at the forefront in AK Parti’s political agenda. As a placebo check we find no significant effect of Islamic governance in 2004 on veiling in 2003 (Figure 3). Figure 3. Open in new tabDownload slide Islamic win/lose margin in the province centre in 2004 versus veiling prevalence in the province in 2003, 2008, and 2013 with loess smoother Figure 3. Open in new tabDownload slide Islamic win/lose margin in the province centre in 2004 versus veiling prevalence in the province in 2003, 2008, and 2013 with loess smoother We further check what happens in provinces gained by the AK Parti in 2004 but lost to another party in the 2009 elections. We expect that, in provinces in which the AK Parti lost in 2009, the effect predicted in H1 should be lower. Once the cause is off, the effect should wane. We find support for this expectation. RD effect of the AK Parti in 2004 on veiling probability in 2013 is about 0.34 in provinces that did not swing away from the AK Parti, while in provinces lost by the AK Parti in 2009 the effect on veiling does not disappear but is significantly smaller (interaction coefficient = −0.16, P = 0.022). See Supplementary data C for further details. One could suspect that the effect of AK Parti governance on veiling might be due to selection: more religious and hence veiled women might have immigrated to AK Parti provinces after 2008. There is no evidence for this explanation. Aksoy and Billari (2018) show that an AK Parti governance has no effect on sub-national net immigration of women. H2 finds support: Islamic governance affects veiling in 2013 substantially, but it does not affect praying significantly neither in 2013 (RDD marginal effect = 0.09, P = 0.304) nor in 2008 (RDD marginal effect = 0.02, P = 0.898) (Figure 4). Figure 4. Open in new tabDownload slide Islamic win/lose margin in the province centre in 2004 and subsequent praying prevalence with loess smoother Figure 4. Open in new tabDownload slide Islamic win/lose margin in the province centre in 2004 and subsequent praying prevalence with loess smoother We test H3 with an interaction between the treatment indicator and the household wealth (z-score) in the RDD. In addition, we control for nesting within households and provinces with random intercepts. We find that for every standard deviation increase in household wealth, the effect of Islamic governance in 2004 on veiling probability in 2013 decreases by about 6 percentage points (interaction coefficient = –0.06, P = 0.012).9 H4 and H5 concern whether the AK Parti distributes resources disproportionately to its voters and to those who signal religiosity. Figure 5 shows how access to free health care (GSS coverage) varies by the AK Parti win/lose margin in the 2004 elections. Supporting H4, there is a very strong effect of AK Parti on GSS coverage in 2013 (β [marginal effect] = 0.18, province-level cluster robust S.E. = 0.08, P = 0.032 for the optimal bandwidth of 0.07). There is a positive RD effect of AK Parti rule on GSS coverage in 2008 which is statistically insignificant (β [marginal effect] = 0.11, province-level cluster robust S.E. = 0.10, P = 0.239 for the optimal bandwidth of 0.19). The effect in 2003 (a year before the 2004 elections) is virtually zero (β [marginal effect] = –0.03, province-level cluster robust S.E. = 0.03, P = 0.270). Figure 5. Open in new tabDownload slide Prevalence of access to health care (GSS/green card) relative to all other forms of insurance by AK Parti win/lose margin in 2004 and by year with loess smoother Figure 5. Open in new tabDownload slide Prevalence of access to health care (GSS/green card) relative to all other forms of insurance by AK Parti win/lose margin in 2004 and by year with loess smoother We also find in line with H5 that, conditional on various controls (viz. marital status, education, age, household wealth, employment status, ideal and actual number of children, gender norms and values, urbanicity, ethnicity, survey year fixed effects) being veiled is positively associated with the likelihood of having GSS coverage versus no coverage in AK Parti provinces (marginal effect = 0.06, S.E. = 0.02, P = 0.022). This result is obtained with a logistic regression model that predicts the probability of having GSS versus having no insurance (see Table 2). Note that this result is not obtained using RDD, so we make no strong claim of causality. Table 2. Logistic regressions predicting the probability of having GSS (green card) versus no health insurance in AK Parti and non-AK Parti provinces . AK Parti Provinces . Non-AK Parti Provinces . Participant veils (binary) 0.262* (0.116) 0.234 (0.225) AKP win/lose margin –1.244+ (0.684) –1.133* (0.556) Year (reference: 2003)  2008 1.314*** (0.156) 1.508*** (0.232)  2013 1.343*** (0.140) 1.614*** (0.397) Education (years) –0.069*** (0.013) –0.037* (0.019) Age –0.028*** (0.006) –0.027** (0.010) Marital status (ref: single)  Married –0.510** (0.189) –0.094 (0.272)  Divorced –0.496* (0.206) 0.110 (0.448)  Widowed 0.137 (0.372) 0.019 (0.599) Ethnicity (ref: Turk)  Kurd 0.516*** (0.107) 0.701*** (0.166)  Arab 0.217 (0.170) 0.000 (0.330)  Other –0.453 (0.445) 0.472 (0.482) Ideal number of children –0.000 (0.021) –0.016 (0.043) Actual number of children 0.122*** (0.023) 0.063 (0.047) Urban versus rural –0.214 (0.134) –0.068 (0.175) Wealth –0.042 (0.058) –0.138+ (0.078) Traditional values –0.064* (0.030) –0.027 (0.079) Employed –0.248*** (0.066) –0.237+ (0.129) _constant 0.157 (0.270) –0.727+ (0.478) N (individual) 5,103 1,532 Log likelihood –3,113.993 –923.39 . AK Parti Provinces . Non-AK Parti Provinces . Participant veils (binary) 0.262* (0.116) 0.234 (0.225) AKP win/lose margin –1.244+ (0.684) –1.133* (0.556) Year (reference: 2003)  2008 1.314*** (0.156) 1.508*** (0.232)  2013 1.343*** (0.140) 1.614*** (0.397) Education (years) –0.069*** (0.013) –0.037* (0.019) Age –0.028*** (0.006) –0.027** (0.010) Marital status (ref: single)  Married –0.510** (0.189) –0.094 (0.272)  Divorced –0.496* (0.206) 0.110 (0.448)  Widowed 0.137 (0.372) 0.019 (0.599) Ethnicity (ref: Turk)  Kurd 0.516*** (0.107) 0.701*** (0.166)  Arab 0.217 (0.170) 0.000 (0.330)  Other –0.453 (0.445) 0.472 (0.482) Ideal number of children –0.000 (0.021) –0.016 (0.043) Actual number of children 0.122*** (0.023) 0.063 (0.047) Urban versus rural –0.214 (0.134) –0.068 (0.175) Wealth –0.042 (0.058) –0.138+ (0.078) Traditional values –0.064* (0.030) –0.027 (0.079) Employed –0.248*** (0.066) –0.237+ (0.129) _constant 0.157 (0.270) –0.727+ (0.478) N (individual) 5,103 1,532 Log likelihood –3,113.993 –923.39 Notes: Cluster (province) robust standard errors are in parentheses; + P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001. Data: Turkey’s 2003, 2008, and 2013 Demographic and Health Survey. Open in new tab Table 2. Logistic regressions predicting the probability of having GSS (green card) versus no health insurance in AK Parti and non-AK Parti provinces . AK Parti Provinces . Non-AK Parti Provinces . Participant veils (binary) 0.262* (0.116) 0.234 (0.225) AKP win/lose margin –1.244+ (0.684) –1.133* (0.556) Year (reference: 2003)  2008 1.314*** (0.156) 1.508*** (0.232)  2013 1.343*** (0.140) 1.614*** (0.397) Education (years) –0.069*** (0.013) –0.037* (0.019) Age –0.028*** (0.006) –0.027** (0.010) Marital status (ref: single)  Married –0.510** (0.189) –0.094 (0.272)  Divorced –0.496* (0.206) 0.110 (0.448)  Widowed 0.137 (0.372) 0.019 (0.599) Ethnicity (ref: Turk)  Kurd 0.516*** (0.107) 0.701*** (0.166)  Arab 0.217 (0.170) 0.000 (0.330)  Other –0.453 (0.445) 0.472 (0.482) Ideal number of children –0.000 (0.021) –0.016 (0.043) Actual number of children 0.122*** (0.023) 0.063 (0.047) Urban versus rural –0.214 (0.134) –0.068 (0.175) Wealth –0.042 (0.058) –0.138+ (0.078) Traditional values –0.064* (0.030) –0.027 (0.079) Employed –0.248*** (0.066) –0.237+ (0.129) _constant 0.157 (0.270) –0.727+ (0.478) N (individual) 5,103 1,532 Log likelihood –3,113.993 –923.39 . AK Parti Provinces . Non-AK Parti Provinces . Participant veils (binary) 0.262* (0.116) 0.234 (0.225) AKP win/lose margin –1.244+ (0.684) –1.133* (0.556) Year (reference: 2003)  2008 1.314*** (0.156) 1.508*** (0.232)  2013 1.343*** (0.140) 1.614*** (0.397) Education (years) –0.069*** (0.013) –0.037* (0.019) Age –0.028*** (0.006) –0.027** (0.010) Marital status (ref: single)  Married –0.510** (0.189) –0.094 (0.272)  Divorced –0.496* (0.206) 0.110 (0.448)  Widowed 0.137 (0.372) 0.019 (0.599) Ethnicity (ref: Turk)  Kurd 0.516*** (0.107) 0.701*** (0.166)  Arab 0.217 (0.170) 0.000 (0.330)  Other –0.453 (0.445) 0.472 (0.482) Ideal number of children –0.000 (0.021) –0.016 (0.043) Actual number of children 0.122*** (0.023) 0.063 (0.047) Urban versus rural –0.214 (0.134) –0.068 (0.175) Wealth –0.042 (0.058) –0.138+ (0.078) Traditional values –0.064* (0.030) –0.027 (0.079) Employed –0.248*** (0.066) –0.237+ (0.129) _constant 0.157 (0.270) –0.727+ (0.478) N (individual) 5,103 1,532 Log likelihood –3,113.993 –923.39 Notes: Cluster (province) robust standard errors are in parentheses; + P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001. Data: Turkey’s 2003, 2008, and 2013 Demographic and Health Survey. Open in new tab A marginal effect of 0.06 corresponds to a 16 per cent increase in the probability of having GSS versus no insurance, for the baseline probability of having no insurance is 19 per cent. Note that this result is an outcome of an individual-level analysis rather than the province-level analysis. That is, within AK Parti provinces, veiled women have better chances of receiving GSS than unveiled women even after adjusting for household wealth, education, employment status, and other covariates. Indeed, veiled women could be poorer on average, which could, in turn, affect the chance of having GSS, so it is important to control for household wealth, education, and employment status in the model. There is also a positive, albeit insignificant association between being veiled and GSS coverage in non-AK Parti provinces (marginal effect = 0.05, S.E. = 0.05, P = 0.298). Praying is not associated significantly with the probability of having GSS versus no insurance in AK Parti (marginal effect = 0.02, P = 0.179) and non-AK Parti provinces (marginal effect = 0.03, P = 0.457). These findings on praying are obtained using the same set of control variables as in Table 2 but excluding veiling. These findings establish that the AK Parti mayors invest more strongly in social welfare than the non-AK Parti mayors and that veiling could increase the chances of access to this welfare, particularly in AK Parti provinces. In the rest of the paper, we will subject both our theoretical interpretation and our empirical results to scrutiny. First, we will impose a large number of checks on our results. Second, we will test the implications of alternative explanations vis-a-vis the implication of our model. Robustness Checks Validating RDD: Multilevel Regression and Difference in Difference We report here the results of a multilevel regression model, which may suffer from internal validity issues but has arguably greater externally validity than RDDs because it uses all observations. In this analysis, we also measure AK Parti’s power in a province in a more elaborate way. Rather than measuring if AK Parti won or lost the province centre, we calculate an ‘Islamic district ratio’, defined as the proportion of districts in a province ruled by the AK Parti after 2004. The results of this conventional analysis are in line with our RDD results (see Table 3). Islamic district ratio in 2004 is associated with veiling positively (in 2013 marginal effect = 0.19, S.E. = 0.073, P = 0.01, in 2008 marginal effect = 0.14, S.E. = 0.069, P = 0.042). Table 3. Table 3. Regression models predicting the probability of veiling. . M1 . M2 . M3 . 2013 . 2008 . Islamic district ratio in 2004 0.189** (0.073) 0.140* (0.069) AKP win/lose margin –0.020 (0.077) 0.065 (0.071) Year (reference: 2003)  2008 –0.022 (0.105)  2013 –0.211 (0.129) AKP × year interaction  AKP × 2008 0.204+ (0.124)  AKP × 2013 0.425* (0.167) Age 0.001 (0.001) –0.002*** (0.001) –0.014** (0.004) Education (years) –0.026*** (0.001) –0.037*** (0.001) –0.190*** (0.009) Marital status (Ref: single) (Ref: widowed)  Married 0.204*** (0.013) –0.044 (0.030) 0.882*** (0.102)  Divorced 0.024 (0.026) –0.240*** (0.038) –0.122 (0.171)  Widowed 0.247*** (0.038) 1.349*** (0.173) Ethnicity (reference: Turk)  Kurd 0.011 (0.015) –0.008 (0.015) 0.152 (0.143)  Arab –0.061+ (0.032) –0.162*** (0.035) –0.672 (0.558)  Other –0.118** (0.039) 0.042 (0.050) –0.464* (0.234) Ideal number of children 0.029*** (0.003) 0.019*** (0.004) 0.268*** (0.023) Actual number of children 0.007+ (0.003) 0.017*** (0.003) 0.304*** (0.028) Urban versus rural –0.011 (0.012) –0.058*** (0.011) –0.368*** (0.104) Wealth –0.072*** (0.006) –0.044*** (0.005) –0.347*** (0.034) Traditional values 0.050*** (0.004) 0.048*** (0.005) 0.405*** (0.030) Employed –0.052*** (0.009) –0.046*** (0.009) –0.263*** (0.053) _constant 0.521*** (0.047) 1.008*** (0.054) 0.267 (0.169) Province fixed effects Yes (suppressed for brevity) Random part  SD(province) 0.117*** (0.011) 0.108*** (0.011)  SD(household) 0.178*** (0.008) 0.193*** (0.016)  SD(individual) 0.341*** (0.004) 0.294*** (0.011) N (individual) 9,665 7,241 20,932 Log likelihood –4,500.089 –2,760.598 –8,367.278 . M1 . M2 . M3 . 2013 . 2008 . Islamic district ratio in 2004 0.189** (0.073) 0.140* (0.069) AKP win/lose margin –0.020 (0.077) 0.065 (0.071) Year (reference: 2003)  2008 –0.022 (0.105)  2013 –0.211 (0.129) AKP × year interaction  AKP × 2008 0.204+ (0.124)  AKP × 2013 0.425* (0.167) Age 0.001 (0.001) –0.002*** (0.001) –0.014** (0.004) Education (years) –0.026*** (0.001) –0.037*** (0.001) –0.190*** (0.009) Marital status (Ref: single) (Ref: widowed)  Married 0.204*** (0.013) –0.044 (0.030) 0.882*** (0.102)  Divorced 0.024 (0.026) –0.240*** (0.038) –0.122 (0.171)  Widowed 0.247*** (0.038) 1.349*** (0.173) Ethnicity (reference: Turk)  Kurd 0.011 (0.015) –0.008 (0.015) 0.152 (0.143)  Arab –0.061+ (0.032) –0.162*** (0.035) –0.672 (0.558)  Other –0.118** (0.039) 0.042 (0.050) –0.464* (0.234) Ideal number of children 0.029*** (0.003) 0.019*** (0.004) 0.268*** (0.023) Actual number of children 0.007+ (0.003) 0.017*** (0.003) 0.304*** (0.028) Urban versus rural –0.011 (0.012) –0.058*** (0.011) –0.368*** (0.104) Wealth –0.072*** (0.006) –0.044*** (0.005) –0.347*** (0.034) Traditional values 0.050*** (0.004) 0.048*** (0.005) 0.405*** (0.030) Employed –0.052*** (0.009) –0.046*** (0.009) –0.263*** (0.053) _constant 0.521*** (0.047) 1.008*** (0.054) 0.267 (0.169) Province fixed effects Yes (suppressed for brevity) Random part  SD(province) 0.117*** (0.011) 0.108*** (0.011)  SD(household) 0.178*** (0.008) 0.193*** (0.016)  SD(individual) 0.341*** (0.004) 0.294*** (0.011) N (individual) 9,665 7,241 20,932 Log likelihood –4,500.089 –2,760.598 –8,367.278 Notes: M1 and M2: three-level (individuals in households in provinces) random effects linear probability regression predicting Pr(veiling) in 2013 and in 2008. M3: difference-in-difference logit model predicting probability veiling in 2003, 2008, and 2013; fixed effects for provinces included; standard errors are robust to clustering in province. Standard errors are in parentheses; + P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001. Open in new tab Table 3. Table 3. Regression models predicting the probability of veiling. . M1 . M2 . M3 . 2013 . 2008 . Islamic district ratio in 2004 0.189** (0.073) 0.140* (0.069) AKP win/lose margin –0.020 (0.077) 0.065 (0.071) Year (reference: 2003)  2008 –0.022 (0.105)  2013 –0.211 (0.129) AKP × year interaction  AKP × 2008 0.204+ (0.124)  AKP × 2013 0.425* (0.167) Age 0.001 (0.001) –0.002*** (0.001) –0.014** (0.004) Education (years) –0.026*** (0.001) –0.037*** (0.001) –0.190*** (0.009) Marital status (Ref: single) (Ref: widowed)  Married 0.204*** (0.013) –0.044 (0.030) 0.882*** (0.102)  Divorced 0.024 (0.026) –0.240*** (0.038) –0.122 (0.171)  Widowed 0.247*** (0.038) 1.349*** (0.173) Ethnicity (reference: Turk)  Kurd 0.011 (0.015) –0.008 (0.015) 0.152 (0.143)  Arab –0.061+ (0.032) –0.162*** (0.035) –0.672 (0.558)  Other –0.118** (0.039) 0.042 (0.050) –0.464* (0.234) Ideal number of children 0.029*** (0.003) 0.019*** (0.004) 0.268*** (0.023) Actual number of children 0.007+ (0.003) 0.017*** (0.003) 0.304*** (0.028) Urban versus rural –0.011 (0.012) –0.058*** (0.011) –0.368*** (0.104) Wealth –0.072*** (0.006) –0.044*** (0.005) –0.347*** (0.034) Traditional values 0.050*** (0.004) 0.048*** (0.005) 0.405*** (0.030) Employed –0.052*** (0.009) –0.046*** (0.009) –0.263*** (0.053) _constant 0.521*** (0.047) 1.008*** (0.054) 0.267 (0.169) Province fixed effects Yes (suppressed for brevity) Random part  SD(province) 0.117*** (0.011) 0.108*** (0.011)  SD(household) 0.178*** (0.008) 0.193*** (0.016)  SD(individual) 0.341*** (0.004) 0.294*** (0.011) N (individual) 9,665 7,241 20,932 Log likelihood –4,500.089 –2,760.598 –8,367.278 . M1 . M2 . M3 . 2013 . 2008 . Islamic district ratio in 2004 0.189** (0.073) 0.140* (0.069) AKP win/lose margin –0.020 (0.077) 0.065 (0.071) Year (reference: 2003)  2008 –0.022 (0.105)  2013 –0.211 (0.129) AKP × year interaction  AKP × 2008 0.204+ (0.124)  AKP × 2013 0.425* (0.167) Age 0.001 (0.001) –0.002*** (0.001) –0.014** (0.004) Education (years) –0.026*** (0.001) –0.037*** (0.001) –0.190*** (0.009) Marital status (Ref: single) (Ref: widowed)  Married 0.204*** (0.013) –0.044 (0.030) 0.882*** (0.102)  Divorced 0.024 (0.026) –0.240*** (0.038) –0.122 (0.171)  Widowed 0.247*** (0.038) 1.349*** (0.173) Ethnicity (reference: Turk)  Kurd 0.011 (0.015) –0.008 (0.015) 0.152 (0.143)  Arab –0.061+ (0.032) –0.162*** (0.035) –0.672 (0.558)  Other –0.118** (0.039) 0.042 (0.050) –0.464* (0.234) Ideal number of children 0.029*** (0.003) 0.019*** (0.004) 0.268*** (0.023) Actual number of children 0.007+ (0.003) 0.017*** (0.003) 0.304*** (0.028) Urban versus rural –0.011 (0.012) –0.058*** (0.011) –0.368*** (0.104) Wealth –0.072*** (0.006) –0.044*** (0.005) –0.347*** (0.034) Traditional values 0.050*** (0.004) 0.048*** (0.005) 0.405*** (0.030) Employed –0.052*** (0.009) –0.046*** (0.009) –0.263*** (0.053) _constant 0.521*** (0.047) 1.008*** (0.054) 0.267 (0.169) Province fixed effects Yes (suppressed for brevity) Random part  SD(province) 0.117*** (0.011) 0.108*** (0.011)  SD(household) 0.178*** (0.008) 0.193*** (0.016)  SD(individual) 0.341*** (0.004) 0.294*** (0.011) N (individual) 9,665 7,241 20,932 Log likelihood –4,500.089 –2,760.598 –8,367.278 Notes: M1 and M2: three-level (individuals in households in provinces) random effects linear probability regression predicting Pr(veiling) in 2013 and in 2008. M3: difference-in-difference logit model predicting probability veiling in 2003, 2008, and 2013; fixed effects for provinces included; standard errors are robust to clustering in province. Standard errors are in parentheses; + P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001. Open in new tab We further report a difference-in-difference analysis. We probe how veiling frequency changes in a province between 2003, 2008, and 2013, and whether the change is different between provinces won by the AK Parti versus provinces lost by the AK Parti in 2004. In a logistic regression, we include dummies for 2008, 2013, and their interactions with a dummy for AK Parti governance. We also control for province fixed effects and other variables (see Table 3, M3). Results show that in non-AK Parti provinces, veiling did not change significantly in 2008 (logit coefficient = –0.022, P = 0.836) and in 2013 (coefficient = –0.21, P = 0.100) compared with 2003. However, the difference between 2003 and 2013 was significantly different in AK Parti provinces than the same difference in non-AK Parti provinces (interaction coefficient = 0.42, S.E. = 0.17, P = 0.011). In fact, in AK Parti provinces, veiling is estimated to be higher (0.43–0.20) in 2013 than in 2003 (sum of the coefficients = 0.21, P = 0.050). This analysis shows that from 2003 to 2013 (before and after the treatment), controlling for the covariates listed above, veiling increased in AK Parti provinces whereas it decreased in non-AK Parti provinces, corroborating our RDD results. Robustness to RDD Specification We carried out several robustness checks some of which we summarize here (full details in the Supplementary data). The RD effect on veiling in 2013 remains virtually unchanged after controlling for several key covariates. Support for political Islam before the 2004 elections does not affect the effect of the 2004 elections. We also carried out 20 placebo cut-off tests, which use a non-zero cut-off value c for the AK Parti win/lose margin (Supplementary Table B2). We estimated the main RD effects using Cattaneo, Idrobo, and Titiunik’s (2019)bias-corrected and robust method which resulted in very similar estimates as those reported above (with the Cattaneo et al.’s method the RD effect on veiling in 2013 is estimated as 0.30, S.E. = 0.08, P < 0.001 and on green card coverage in 2013 as 0.18, S.E. = 0.08, P = 0.028). Finally, Figure 6 shows the robustness of the estimated RDD effects for various outcome variables for a set of different bandwidths including the optimal bandwidth. The significant RDD effect obtained for the outcome variable veiling in 2013 remains positive, large, and statistically significant for most alternative bandwidths. Insignificant effects remain insignificant in all cases, except for praying (namaz) in 2013 for a very small bandwidth. Figure 6. Open in new tabDownload slide Robustness of RDD effect estimate with respect to bandwidth selection, a vertical line per panel shows the optimal bandwidth, 95 per cent confidence intervals are given Figure 6. Open in new tabDownload slide Robustness of RDD effect estimate with respect to bandwidth selection, a vertical line per panel shows the optimal bandwidth, 95 per cent confidence intervals are given Alternative Explanations We find support for all five hypotheses. We interpret these effects as outcomes of a signalling mechanism that drive women to veiling in AK Parti provinces. There may be other mechanisms that could explain these findings. One can argue that people may experience an increase in genuine piety, driven, for example by a sense of enhanced group identity in regions in which the Islamist party improved pastoral care. If that was the case, we would expect a positive effect of AK Parti governance on praying too—there is evidence that real piety is associated positively with an array of religious practices including praying (Aksoy and Gambetta, 2016). As we noted above, however, we do not find a significant effect of AK Parti on praying. Another alternative mechanism we discussed above is conformism with the mores of one’s community: community or normative pressures can spread through local networks and increase the number of veiled women to the point of inducing even women who are less religious to veil. To test whether conformism matters, we carry out the following three analyses. First, we test whether the AK Parti effect on veiling in 2013 in our RDD is stronger in provinces that had a high veiling prevalence 10 years earlier, in 2003. If the effect was driven by adaptation to community customs and in response to the normative pressure in AK Parti provinces, one would expect this effect to be stronger the higher was the veiling prevalence before the 2004 elections. We find no significant interaction between veiling prevalence in 2003 and AK Parti governance (interaction coefficient: –0.11, P = 0.704). Also, controlling for veiling prevalence in the province in 2003 does not affect the estimated coefficient for the AK Parti governance. Second, to probe more directly the political signalling motive that may draw women to veiling in AK Parti provinces, we analyse how political participation is linked with veiling. We use the likelihood of voting as an indicator of political participation. THDS13 includes a question on whether the participant votes regularly (TDHS08 and TDHS03 do not include the voting question). Figure 7 presents the marginal effects of veiling on the probability of voting in 2013. These marginal effects are obtained after adjusting for key covariates. The effects are qualitatively the same without covariate adjustment. The left-hand panel in Figure 7 provides the marginal effects in AK Parti and non-AK Parti provinces. The right-hand panel in Figure 7 further breaks down the marginal effects between the more and the less religious (based on whether the respondent prays or not). While the differences between the marginal effects in the likelihood of voting between AK Parti and non-AK Parti provinces do not seem statistically significant, we find three important and statistically significant results: veiling in 2013 is Figure 7. Open in new tabDownload slide Marginal effects of veiling on the probability of voting in 2013 and 90 per cent confidence interval; adjusted for marital status, education, age, wealth, employment status, ideal number of children, gender norms and values, ethnicity, urbanicity, praying, random effects for provinces, and households. Left panel: average of high and low religious. Right panel: by religiosity. Figure 7. Open in new tabDownload slide Marginal effects of veiling on the probability of voting in 2013 and 90 per cent confidence interval; adjusted for marital status, education, age, wealth, employment status, ideal number of children, gender norms and values, ethnicity, urbanicity, praying, random effects for provinces, and households. Left panel: average of high and low religious. Right panel: by religiosity. associated with women’s voting in AK Parti provinces; not associated with women’s voting in non-AK Parti provinces. Finally, the veiling–voting association in AK Parti provinces seems driven by the less religious—for the highly religious veiling is not significantly associated with voting. Third, to study further the possible reasons behind veiling, we use a question asked by KONDA Research and Consultancy to a random sample of individuals in Turkey in 2007 and 2010.10 This question probes directly the link between wearing the conservative veil türban (as opposed to the less conservative traditional headscarf) and politics: ‘In your opinion, by wearing türban rather than traditional headscarves, what do you or others demonstrate’.11 While not perfect, this question tells us something about how subjects perceive the link between veiling and politics. One of the answer categories is ‘they show their political tendency/standing’. The answer categories also include conformism and social identity as possible reasons. Respondents are instructed to select only one answer. Table 4 shows the answers broken down by year of the survey and by whether the respondent (or their wife) veils regularly. Unsurprisingly, the most popular answer is ‘they abide by Islam’s dictate’. Strikingly, however, the second most popular answer is showing one’s political tendency, among both veiled and unveiled respondents. Note that since respondents are asked to choose only one answer category, this finding is even more revealing, for it means that, for those who reply ‘showing one’s political tendency’, this must be the first answer that comes their mind. If respondents had been given the chance to answer with more than one category, it would be reasonable to expect that some of those who chose other categories, especially the dutiful group who replied ‘they abide by Islam’s dictate’, might also have ticked ‘showing one’s political tendency’ as their second answer. Table 4. Answers to ‘In your opinion, by wearing türban rather than traditional headscarves, what do you or others demonstrate?’ by year and veiling . 2007 . 2010 . Unveiled . Veiled . Unveiled . Veiled . They show that they conform to their social surrounding 115 (7.7) 238 (7.0) 39 (5.6) 63 (3.7) They show that they are virtuous 57 (3.8) 95 (2.8) 12 (1.72) 55 (3.2) They abide by Islam’s dictate 639 (42.5) 2,622 (76.9) 222 (31.9) 1,188 (69.2) They show their political tendency/standing 450 (29.9) 260 (7.6) 299 (43.0) 251 (14.6) They show their identity 143 (9.5) 79 (2.3) 69 (9.91) 51 (3.0) They protect themselves (2007) / Other (2010) 36 (2.4) 61 (1.8) 46 (6.61) 92 (5.4) No answer 63 (4.2) 54 (1.6) 9 (1.29) 18 (1.1) Total 1,503 (100) 3,409 (100) 696 (100) 1,718 (100) . 2007 . 2010 . Unveiled . Veiled . Unveiled . Veiled . They show that they conform to their social surrounding 115 (7.7) 238 (7.0) 39 (5.6) 63 (3.7) They show that they are virtuous 57 (3.8) 95 (2.8) 12 (1.72) 55 (3.2) They abide by Islam’s dictate 639 (42.5) 2,622 (76.9) 222 (31.9) 1,188 (69.2) They show their political tendency/standing 450 (29.9) 260 (7.6) 299 (43.0) 251 (14.6) They show their identity 143 (9.5) 79 (2.3) 69 (9.91) 51 (3.0) They protect themselves (2007) / Other (2010) 36 (2.4) 61 (1.8) 46 (6.61) 92 (5.4) No answer 63 (4.2) 54 (1.6) 9 (1.29) 18 (1.1) Total 1,503 (100) 3,409 (100) 696 (100) 1,718 (100) Notes: Column percentages are in parentheses, the answer category on political signalling is indicated in bold. Data: KONDA 2007 and 2010 surveys. Open in new tab Table 4. Answers to ‘In your opinion, by wearing türban rather than traditional headscarves, what do you or others demonstrate?’ by year and veiling . 2007 . 2010 . Unveiled . Veiled . Unveiled . Veiled . They show that they conform to their social surrounding 115 (7.7) 238 (7.0) 39 (5.6) 63 (3.7) They show that they are virtuous 57 (3.8) 95 (2.8) 12 (1.72) 55 (3.2) They abide by Islam’s dictate 639 (42.5) 2,622 (76.9) 222 (31.9) 1,188 (69.2) They show their political tendency/standing 450 (29.9) 260 (7.6) 299 (43.0) 251 (14.6) They show their identity 143 (9.5) 79 (2.3) 69 (9.91) 51 (3.0) They protect themselves (2007) / Other (2010) 36 (2.4) 61 (1.8) 46 (6.61) 92 (5.4) No answer 63 (4.2) 54 (1.6) 9 (1.29) 18 (1.1) Total 1,503 (100) 3,409 (100) 696 (100) 1,718 (100) . 2007 . 2010 . Unveiled . Veiled . Unveiled . Veiled . They show that they conform to their social surrounding 115 (7.7) 238 (7.0) 39 (5.6) 63 (3.7) They show that they are virtuous 57 (3.8) 95 (2.8) 12 (1.72) 55 (3.2) They abide by Islam’s dictate 639 (42.5) 2,622 (76.9) 222 (31.9) 1,188 (69.2) They show their political tendency/standing 450 (29.9) 260 (7.6) 299 (43.0) 251 (14.6) They show their identity 143 (9.5) 79 (2.3) 69 (9.91) 51 (3.0) They protect themselves (2007) / Other (2010) 36 (2.4) 61 (1.8) 46 (6.61) 92 (5.4) No answer 63 (4.2) 54 (1.6) 9 (1.29) 18 (1.1) Total 1,503 (100) 3,409 (100) 696 (100) 1,718 (100) Notes: Column percentages are in parentheses, the answer category on political signalling is indicated in bold. Data: KONDA 2007 and 2010 surveys. Open in new tab Moreover, the proportion of respondents who think türban is worn to show ones’ political tendency increased significantly from 2007 to 2010. For veiled respondents, the proportion is almost doubled (7.6–14.6 per cent) and for the unveiled it increased by almost 50 per cent (30 per cent versus 43 per cent). The difference in the answers between 2007 and 2010 is statistically significant both for the unveiled (χ2 (6) = 81.65, P < 0.001) and the veiled respondents (χ2 (6) = 140.12, P < 0.001). Interestingly, in 2010, almost no veiled women thought that türban shows conformism to the surrounding (3.7 per cent) or identity (3 per cent). This shows that conformism and identity are not the main drivers of türban, and turban is seen as a signal of political tendency by a sizable number of people. Based on these additional analyses, we cannot rule out that veiling is driven also by conformism but we can exclude that it is only driven by that: the association between veiling and politics is an indication that women’s strategic considerations could be an important factor driving our results. Lift of Veil Ban in 2013 There was a major legislation change in Turkey that concerned veiling and which could challenge our explanation: until October 2013, veiling was forbidden in public offices in Turkey, and then, on the 4th of that month, parliament lifted the ban. One can argue that this change might have had a stronger positive effect on veiling in AK Parti provinces than in non-AK Parti provinces. Women who always wanted to veil but did not do so because of the ban, now felt legitimized to veil especially in AK Parti provinces. AK Parti provinces may have also been quicker in implementing the repel of the ban than non-AK Parti provinces, resulting in an early difference in veiling frequency. These factors could have brought about the difference we observe in Figure 3between the AK Parti and non-AK Parti provinces. To test this conjecture, we exploit the variation in the interview date in the 2013 wave of TDHS. In 2013, TDHS was fielded conveniently between 1 September and end of December. Figure 8 below shows how prevalence of veiling varies by the date on which a respondent was interviewed (centred on the day of lift of veil ban) broken down by whether the province is ruled by the AK Parti. As the figure shows, we do not find any change in the effect of the lift of the ban across the AK Parti and non-AK Patri provinces. Neither the change immediately before and after the day of the lift of the ban nor the change over time varies significantly between AK Parti and non-AK Parti provinces. Figure 8. Open in new tabDownload slide Binned scatterplot of prevalence of veiling by interview date and whether the province is ruled by the AK Parti Data: Turkey’s 2013 Demographic and Health Survey. Figure 8. Open in new tabDownload slide Binned scatterplot of prevalence of veiling by interview date and whether the province is ruled by the AK Parti Data: Turkey’s 2013 Demographic and Health Survey. Discussion and Conclusions In this study, we provide evidence that religious practices respond to incentives. If, for instance, the semblance of piety pays off, even those whose religious beliefs are feeble or absent may choose to mimic religiosity by adopting its outward manifestations. Comparing the provinces with narrow AK Parti wins with those with narrow AK Parti losses, we find that an increase in benefits for the faithful increases veiling without increasing praying. This finding that we derive from an RDD survives alternative specifications and statistical models, such as difference-in-difference analysis and traditional multilevel regressions. This finding is compatible with our theoretical model, which links voters’ signalling of allegiance through perceivable religious practice with the ruling party’s payoffs from patronage. Our study thus contributes to the sociology of religion literature showing that observable religious practices have their independent dynamics driven by the pursuit of mundane goals. We also contribute to the literature on clientelism by developing and testing a theory based on signalling and screening. We think that this theory helps us understand better the interplay between politics and religion. We argue how religious behaviours and appurtenances can turn into efficient signals of allegiance for the voters and screening devices for the ruling party. We also argue that political parties with a religious ideology and their voters may have an advantage over their secular competitors in solving a classical problem of clientelism: whom to distribute patronage and whom to exclude. Our primary aim in our empirical analyses is to test the predictions of this theory. Still, we also scrutinize other mechanisms that are consistent with the finding that women are more likely to veil in AK Parti provinces than in non-AK Parti provinces. A possible alternative is conformism resulting from community pressure and related social norms. An AK Parti win may signal to a non-religious woman that the community is conservative. This may make her veil not to stick out and risk to raise questions about her respectability. We cannot fully rule out that veiling increase that we register in provinces governed by the AK Parti owes also to this conformism effect. In fact, a definitive test of exact mechanisms requires more fine-grained data in the form of qualitative interviews, ethnographic studies, and experimental work. Nevertheless, we disentangle signalling from conformism empirically in four different ways. First, we show that in AK Parti provinces veiling among the non-religious is strongly associated with voting. Next, we find that veiling prevalence before the elections makes no difference to the effect of AK Parti wins on subsequent veiling—if conformism was the dominant factor, one would expect otherwise. Third, we show that at the province level an AK Parti majority is associated with more widespread social security coverage, and at the individual level, being veiled is associated with a higher chance of having access to social security. And finally, we find that the proportion of veiled and unveiled people who think that women who wear the turban do it to show their political standing has increased significantly from 2007 to 2010 and that very few veiled people think that turban is worn to show conformism or social identity. Taken together, these findings suggest that women’s strategic considerations are the driver of our results. Finally, our study enhances our understanding of political Islam’s success in expanding its power in Turkey, a once staunchly secular country. We show that AK Parti and its grassroots organizations may be providing social assistance and welfare very effectively to create selective incentives for the pious and the poor. The rise of the AK Parti in Turkey, thus, resembles the successful ideological and religious movements elsewhere: a strong emphasis on mutual aid, charitable giving, and the provision of basic local services in recruiting and retaining followers. Here, we show how a wafer-thin AK Parti win causes 25 percentage points difference in the likelihood of veiling. It is, however, yet to be seen if this effect is permanent. We find that only outward displays of religiosity rather than more demanding or unobservable religious practices are affected by the AK Parti rule. We also find that in provinces in which the AK Parti subsequently lost, the positive effect of the party on veiling wanes. This suggests that the shift in veiling could be a ‘bubble’ bound to burst: should the payoff bonanza set by the AK Parti cease, the difference between AK Parti and non-AK Parti provinces may diminish. Yet, there is a second possibility. One may choose to flaunt by outward practices an otherwise feeble religiosity for mundane reasons at first. But still, if Blaise Pascal’s famous assumption was correct, after some unspecified length of time true religiosity may creep up and eventually align itself to the outward displays of it.12 As the great 13th century Muslim poet Rumi put it, ‘Appear as you are. Or be as you appear.’ The theory of cognitive dissonance reduction would predict such an outcome (Festinger, 1957; Kuran, 1995). It predicts paradoxically that, if the threat of punishment for having a certain view or the rewards for changing it publicly are large, then the person is less likely to change her private views because modifying the public view is fully justified by the large size of the reward or punishment. By contrast, where the actions initially misaligned with beliefs are more freely chosen, move forward by short steps, and are induced, rather than by dramatic shifts, by small payoffs or punishments, then private beliefs are more likely to change to match the public view. The second condition seems to fit our case: women do not receive a dramatic punishment or reward if they do or do not veil, at least not immediately, and are free not to do it. So the cognitive dissonance reduction prediction would be that to justify what they do, freely and for no great benefit, veiled women are more likely to align their private beliefs to their public ones, and the increase in any initially opportunistic religiosity may eventually lead to a genuine increase in religiosity. Notes Ozan Aksoy is an associate professor of social science at University College London. His research interests include cooperation, trust, and religious behaviour. He uses game theory, statistical and computational methods, and laboratory and natural experiments as research tools. He is the recipient of the 2019 Raymond Boudon Award for Early Career Achievement of the European Academy of Sociology. His work has been published in European Sociological Review, American Journal of Sociology, Sociological Science, Electoral Studies, Social Psychology Quarterly, Social Science Research, Journal of Mathematical Sociology, Journal of Experimental Social Psychology, and Games. Diego Gambetta is a Carlo Alberto Chair at Collegio Carlo Alberto, Turin, and Emeritus Fellow at Nuffield College, Oxford. His main interests are trust, signalling theory and its applications, organized crime, and violent extremists. Since 2000, he is a fellow of the British Academy. His latest book, with Steffen Hertog, is Engineers of Jihad: The Curious Connection between Violent Extremism and Education (Princeton UP 2016). Footnotes 1 A potentially serious cost of veiling is vitamin D deficiency due to lower exposure to sunrays. This effect is yet to be properly established (Baroncelli et al., 2008). 2 For the full report (in Turkish), see http://egitimsen.org.tr/wp-content/uploads/2017/09/E%C4%9Fitimin-Durumu-Raporu-13-Eyl%C3%BCl-2017.pdf [accessed 14 August 2020]. 3 See (in Turkish) http://www2.diyanet.gov.tr/DinHizmetleriGenelMudurlugu/KadnveAileyeYonelikCalsmalar/KadinVeA%C4%B1leyeYonel%C4%B1kCalismalar.pdf [accessed 14 August 2020]. 4 See (in English) http://www.dw.com/en/german-court-hands-down-jail-terms-in-islamic-charity-scandal/a-3652266 [accessed 14 August 2020]. 5 See (in Turkish) http://www.memursen.org.tr/show_file.php?attachid=293 [accessed 14 August 2020]. 6 All surveys employ a weighted, stratified, multi-stage cluster sampling approach. For further details, see Hacettepe University Institute of Population Studies (2008–2014). 7 For a logistic specification with a binary outcome variable, E(y) is replaced by log[p/(1 − p)]. 8 Full replication material (data and code) is submitted with the manuscript for peer review. 9 We do not find any significant interaction between local Islamic governance and household wealth for veiling in 2008 (interaction coefficient = 0.037, P = 0.206) and when random effects for provinces and households were omitted (interaction coefficient = −0.05, P = 0.186). 10 Further details of the datasets can be found in KONDA Research and Consultancy (2007, 2010). 11 Aksoy and Gambetta (2016) compare türban and the traditional headscarf and show that the former is used for strategic signalling purposes 12 See Pascal’s Pensée 250: ‘The external must be joined to the internal to obtain anything from God, that is to say, we must kneel, pray with the lips, etc., in order that proud man, who would not submit himself to God, may be now subject to the creature. To expect help from these externals is superstition; to refuse to join them to the internal is pride’ (www.samizdat.qc.ca/arts/lit/Pascal/Pensees_English.pdf [accessed 14 August 2020]). Supplementary Data Supplementary data are available at ESR online. Acknowledgements We thank Timur Kuran, Werner Raub, Aron Szekely, David Laitin, David Voas, the editor, and the three anonymous reviewers of ESR for their comments on earlier drafts; the participants of the Fifth International Meeting on Experimental and Behavioral Social Sciences particularly Johanna Greke, Macartan Humphreys, Krzysztof Krakowski, Cristine Legare, and Wojtek Przepiorka; and the participants of the UCL Department of Quantitative Social Science Seminar Series for insightful suggestions. References Aksoy O. ( 2017 ). Motherhood, sex of the offspring, and religious signalling . Sociological Science , 4 , 511 – 527 . Google Scholar Crossref Search ADS WorldCat Aksoy O. ( 2016 ). Regression discontinuity diagnostics reveal statistical anomalies in Turkish elections . Electoral Studies , 44 , 284 – 292 . Google Scholar Crossref Search ADS WorldCat Aksoy O. , Billari F. C. ( 2018 ). Political Islam, marriage, and fertility: evidence from a natural experiment . American Journal of Sociology , 123 , 1 – 45 . Google Scholar Crossref Search ADS WorldCat Aksoy O. , Gambetta D. ( 2016 ). Behind the veil: the strategic use of religious garb . European Sociological Review , 32 , 792 – 806 . Google Scholar Crossref Search ADS WorldCat Allport G. W. , Ross J. M. ( 1967 ). Personal religious orientation and prejudice . Journal of Personality and Social Psychology , 5 , 432 . Google Scholar Crossref Search ADS PubMed WorldCat Baroncelli G. I. et al. ( 2008 ). Rickets in the Middle East: role of environment and genetic predisposition . Journal of Clinical Endocrinology & Metabolism , 93 , 1743 – 1750 . Google Scholar Crossref Search ADS WorldCat Berman E. ( 2011 ) Radical, Religious, and Violent: The New Economics of Terrorism. Massachusetts: MIT. Blaydes L. ( 2014 ). How does Islamist local governance affect the lives of women? Governance , 27 , 489 – 509 . Google Scholar Crossref Search ADS WorldCat Brünig B. , Fleischmann F. ( 2015 ). Understanding the veiling of Muslim women in the Netherlands . Journal for the Scientific Study of Religion , 54 , 20 – 37 . Google Scholar Crossref Search ADS WorldCat Buğra A. , Keyder Ç. ( 2006 ). The Turkish welfare regime in transformation . Journal of European Social Policy , 16 , 211 – 228 . Google Scholar Crossref Search ADS WorldCat Calvo E. , Murillo M. V. ( 2004 ). Who delivers? Partisan clients in the Argentine electoral market . American Journal of Political Science , 48 , 742 – 757 . Google Scholar Crossref Search ADS WorldCat Cammett M. , Luca D. , Sergenti E. ( 2018 ). Unfair Play: The Politics of Turkey's Central Government Spending under AKP Rule. Working Paper, available from: https://bit.ly/2GG2VI8 [accessed 05 May 2018]. Carvalho J.-P. ( 2013 ). Veiling . Quarterly Journal of Economics , 128 , 337 – 370 . Google Scholar Crossref Search ADS WorldCat Cattaneo M. D. , Idrobo N. , Titiunik R. ( 2019 ). A Practical Introduction to Regression Discontinuity Designs: Foundations . Cambridge: Cambridge University Press . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Chen D. L. ( 2010 ). Club goods and group identity: evidence from Islamic resurgence during the Indonesian financial crisis . Journal of Political Economy , 118 , 300 – 354 . Google Scholar Crossref Search ADS WorldCat Çörekçioğlu G. ( 2018 ). Headscarves and Female Employment. European University Institute Working Paper: Florence, available from: https://www.dropbox.com/s/pgp1y7vtnfspdwv/jmp_GC.pdf?dl=0 [accessed 07 June 2018]. Corstange D. ( 2016 ). The price of a vote in the middle east: Clientelism and communal politics in Lebanon and Yemen . New York : Cambridge University Press . Davidson J. D. , Pyle R. E. , Reyes D. V. ( 1995 ). Persistence and change in the protestant establishment, 1930–1992 . Social Forces , 74 , 157 – 175 . Google Scholar Crossref Search ADS WorldCat de la Cuesta B. , Imai K. ( 2016 ). Misunderstandings about the regression discontinuity design in the study of close elections . Annual Review of Political Science , 19 , 375 – 396 . Google Scholar Crossref Search ADS WorldCat Eggers A. C. et al. ( 2015 ). On the validity of the regression discontinuity design for estimating electoral effects: new evidence from over 40,000 close races . American Journal of Political Science , 59 , 259 – 274 . Google Scholar Crossref Search ADS WorldCat Festinger L. ( 1957 ) A Theory of Cognitive Dissonance . Stanford, CA : Stanford University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Fleischmann F. , Phalet K. ( 2012 ). Integration and religiosity among the Turkish second generation in Europe: a comparative analysis across four capital cities . Ethnic and Racial Studies , 35 , 320 – 341 . Google Scholar OpenURL Placeholder Text WorldCat Francis B. , Hutchings M. ( 2013 ). Parent Power? Using Money and Information to Boost Children’s Chances of Educational Success. Sutton Trust Report, available from: http://dera.ioe.ac.uk/30277/1/1parentpower-final.pdf [accessed 07 June 2018]. Gambetta D. ( 2009 ). Signaling. In Hedström P. , Bearman P. (Eds.), The Oxford Handbook of Analytical Sociology . Oxford : Oxford University Press , pp. 168 – 194 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Gangl M. ( 2010 ). Causal inference in sociological research . Annual Review of Sociology , 36, 21-47 . Google Scholar OpenURL Placeholder Text WorldCat Hacettepe University Institute of Population Studies ( 2008 –2014). Turkey Demographic and Health Survey. Ankara, Turkey: Hacettepe University Institute of Population Studies, T.R. Ministry of Development and TÜBİTAK. Hicken A. ( 2011 ). Clientelism . Annual Review of Political Science , 14 , 289 – 310 . Google Scholar Crossref Search ADS WorldCat Iannaccone L. R. ( 1992 ). Sacrifice and stigma: reducing free-riding in cults, communes, and other collectives . Journal of Political Economy , 100 , 271 – 291 . Google Scholar Crossref Search ADS WorldCat Imbens G. , Kalyanaraman K. ( 2012 ). Optimal bandwidth choice for the regression discontinuity estimator . The Review of Economic Studies , 79 , 933 – 959 . Google Scholar Crossref Search ADS WorldCat Kalyvas S. N. ( 1996 ). The Rise of Christian Democracy in Europe . New York : Cornell University Press . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Kemahlıoğlu Ö. , Bayer R. ( 2020 ). Favoring co-partisan controlled areas in central government distributive programs: the role of local party organizations. Public Choice, 1 – 19 . KONDA Research and Consultancy ( 2007 ). Religion, Secularism and the Veil in Daily Life, available from: https://konda.com.tr/wp-content/uploads/2017/03/2007_09_KONDA_Religion_Secularism_and_the_Veil_in_Daily_Life_Survey.pdf [accessed 17 August 2020].. KONDA Research and Consultancy ( 2010 ). Başını Örtme ve Türban Araştırması [Research on headscarf and türban], available (in Turkish) from: https://konda.com.tr/wp-content/uploads/2017/03/KONDA_1011_BASINI_ORTME_VE_TURBAN.pdf [accessed 14 August 2020]. Kuran T. ( 1995 ). Private Truths, Public Lies: The Social Consequences of Preference Falsification . Cambridge, MA : Harvard University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kuran T. ( 2018 ). Islam and economic performance: historical and contemporary links . Journal of Economic Literature , 56, 1292-1359 . Google Scholar OpenURL Placeholder Text WorldCat Mardin Ş. ( 2010 ). Türkiye’de Din ve Siyaset [Religion and Politics in Turkey]. Istanbul: İletişim. McCrary J. ( 2008 ). Manipulation of the running variable in the regression discontinuity design: A density test . Journal of Econometrics , 142 , 698 – 714 Google Scholar Crossref Search ADS WorldCat Merton R. K. ( 1968 ) Social Theory and Social Structure . New York : Free Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Metin O. ( 2011 ). Sosyal Politika Açısından AKP Dönemi: Sosyal Yardım Alanında Yaşananlar [Social policies of the AKP reign: the case of social aids] . Çalışma ve Toplum Dergisi [Work and Society] , 1 , 179 – 200 . Google Scholar OpenURL Placeholder Text WorldCat Meyersson E. ( 2014 ). Islamic rule and the empowerment of the poor and pious . Econometrica , 82 , 229 – 269 . Google Scholar Crossref Search ADS WorldCat Morgan S. L. , Winship C. ( 2007 ). Counterfactuals and Causal Inference. Methods and Principles for Social Research . Cambridge : Cambridge University Press . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Patel D. S. ( 2012 ). Concealing to reveal: the informational role of Islamic dress . Rationality & Society , 24 , 295 – 323 . Google Scholar Crossref Search ADS WorldCat Rodrik D. ( 2012 ). Turkish economy after the global financial crisis . Ekonomi-tek , 1 , 41 – 61 . Google Scholar OpenURL Placeholder Text WorldCat Ruiter S. , Van Tubergen F. ( 2009 ). Religious attendance in cross-national perspective: a multilevel analysis of 60 countries . American Journal of Sociology , 115 , 863 – 895 . Google Scholar Crossref Search ADS WorldCat Somer M. ( 2017 ). Conquering versus democratizing the state: political Islamists and fourth wave democratization in Turkey and Tunisia . Democratization , 24 , 1025 – 1043 . Google Scholar Crossref Search ADS WorldCat Spence M. ( 1974 ). Market Signaling . Cambridge, MA : Harvard University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Stokes S. C. et al. ( 2013 ). Brokers, Voters and Clientelism. The Puzzle of Distributive Politics . Cambridge : Cambridge University Press . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Thistlethwaite D. L. , Campbell D. T. ( 1960 ). Regression-discontinuity analysis: an alternative to the ex post facto experiment . Journal of Educational Psychology , 51 , 309 . Google Scholar Crossref Search ADS WorldCat Titiunik R. ( 2020 ). Natural experiments. arXiv preprint arXiv: 2002.00202. Umit R. ( 2018 ) Istanbul puts the money where AKP voters are. Blog post at Medium, available from: https://medium.com/@resulumit/istanbul-puts-the-money-where-the-akp-voters-are-4f60727c88e1 [accessed 18 December 2018]. Voas D. ( 2007 ). Surveys of behaviour, beliefs and affiliation. In Beckford J. , Demerath N. J. (Eds.), Handbook of the Sociology of Religion . London : Sage , pp. 128 – 50 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Yavuz M. H. ( 2000 ). Cleansing Islam from the public sphere . Journal of International Affairs , 54 , 21 . Google Scholar OpenURL Placeholder Text WorldCat Yoltar Ç. ( 2009 ). When the poor need health care: ethnography of state and citizenship in Turkey . Middle Eastern Studies , 45 , 769 – 782 . Google Scholar Crossref Search ADS WorldCat Zahavi A. , Zahavi A. ( 1998 ). The Handicap Principle . Oxford : Oxford University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC © The Author(s) 2020. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - The Politics behind the Veil
JF - European Sociological Review
DO - 10.1093/esr/jcaa035
DA - 2021-01-30
UR - https://www.deepdyve.com/lp/oxford-university-press/the-politics-behind-the-veil-cb9sBA07x0
SP - 67
EP - 88
VL - 37
IS - 1
DP - DeepDyve
ER -