Do Conditionalities Increase Support for Government Transfers?

Abstract Conditional Cash Transfers (CCTs) have spread through the developing world in the past two decades. It is often assumed that CCTs enjoy political support in the population precisely because they impose conditions on beneficiaries. This article employs survey experiments in Brazil and Turkey to determine whether, and in what contexts, making government transfers conditional on behaviour of beneficiaries increases political support for the programmes. Results show that conditional transfers are only marginally more popular than similar unconditional transfers in nationally representative samples, but that this difference is substantially larger among the better-off and among those primed to think of themselves as different from beneficiaries. These findings imply that conditionalities per se are not as strong a determinant of support for transfers as the literature suggests, but that they can still be helpful in building support for transfers among subsets of the population that are least likely to support them.


Introduction
Conditional cash transfers (henceforth CCTs) are a class of non-contributory social policies that attach a 'conditionality' to a social transfer. The conditionality is typically a requirement to use public services such as health and education ). 1 From their origins in the mid-1990s as a local-level policy innovation in a handful of Brazilian municipalities (Amaral & Ramos, 1999;World Bank, 2001), CCTs have expanded to large-scale and highly-visible national-level policies. Mexico's Progresa/Oportunidades/Prospera and Brazil's Bolsa Família Program (BFP) are the largest of these programmes, but CCTs are present in dozens of countries around the world.
The fast diffusion of CCTs, first throughout Latin America and then across the developing world, is often attributed to the acceptance of such transfers by voters. This acceptability of CCTs even among wealthier individuals could be driven by many different factors, but the literature suggests that a very likely culprit is the fact that these transfers are, as their name implies, conditional. This idea was clearly formulated in the influential 2009 World Bank report on CCTs , which suggested that transfers should be made conditional on the behaviour of beneficiaries whenever there exists an 'unfavourable political economy' that would make redistributive transfers politically hard to implement. In fact, the argument that conditionalities make transfers more acceptable to the better-off is so intuitive, that it is promoted without actually having been tested.
Granted, increasing political support for transfers is not necessarily the main reason to condition benefits. Conditionalities might, for instance, affect the welfare of beneficiaries directly, but separating effects of the transfers from those generated by the conditionalities attached to them is not straightforward and recent work on this issue yields mixed results. Even if conditionalities do not improve the welfare of beneficiaries, however, they might be justified by their positive political-economy effects. This is particularly relevant because differently than most of the factors known to affect support for redistribution, the conditional or unconditional nature of a transfer is almost entirely in the hands of policy-makers and policy implementers. In particular, even if conditionalities represent a burden, they might be politically necessary to guarantee the existence of distributive programmes.
In this article we examine whether the potential political-economy benefits from the conditional nature of CCTs exist. In order to make this assessment, we designed and fielded survey-experiments in Brazil and Turkey, two highly heterogenous middle-income countries that have implemented CCTs. We find that conditionalities increase average support for transfers only very marginally, and that these effects are concentrated almost exclusively among wealthier individuals. We then manipulate the perceptions of 'differences' between respondents and beneficiaries and find that the effects of conditionalities are larger among those primed to think of themselves as different from potential beneficiaries.
Our conclusion is that conditionalities produce a limited increase of support among nonbeneficiaries, but only among those that regard themselves as different from beneficiaries. These individuals are precisely the ones most likely to oppose any such transfers, which suggests that conditionality might be politically useful to overcome some resistance to redistribution. On the other hand, the existence of a significant administrative burden generated by the imposition of conditionalities requires that we first assess the potential magnitude of the political gain to determine whether it makes sense to condition benefits on behaviour.

Possible impacts of conditioning social transfers
We are interested in whether the imposition of conditionalities on beneficiaries of social transfers affects support for such transfers, particularly among non-beneficiaries. Social transfers can be targeted or universal and conditional or unconditional. 2 We focus only on variation in the second of these dimensions.
There is considerable debate over the economic rationale for conditioning government benefits on recipient behaviour. 3 On the one hand, each family knows best, one argument goes, how to allocate its own resources; restricting behaviour by imposing conditions cannot possibly improve a family's position. On the other, several purely economic arguments have been made as to how conditionalities help overcome imperfect credit markets, under-investment in education due to positive externalities (and hence increase human capital), or suboptimal outcomes driven by incomplete information, timeinconsistent preferences, or divergence of interest between children and their parents.
While there exists a copious literature in several fields examining different socio-economic impacts of CCTs (Barros, Carvalho, Franco, & Mendonça, 2010;Behrman, Piyali, & Petra, 2005;Januzzi & Pinto, 2013;Maluccio, 2010;Manacorda, Amarante, Miguel, & Vigorito, 2016;Soares, Soares, Medeiros, & Osório, 2006, inter alia), fewer authors have examined whether the conditionalities, in and of themselves, actually produce any independent effect on beneficiaries' welfare. Separating effects of the transfers from those of the conditionality is not an easy task. Some studies that have attempted to do so have found limited evidence of the effect of conditionalities. Unconditional programmes, for instance, succeed in changing the behaviours typically targeted by conditionalities (Case, Hosegood, & Lund, 2005;Edmonds & Schady, 2012), and one study even found that simply labelling an unconditional transfer as a 'schooling incentive' was as effective in increasing school attendance as a transfer that truly conditioned payments on attendance (Benhassine, Devoto, Duflo, Dupas, & Pouliquen, 2015).
Other studies that examined educational and health outcomes found that the effects of conditionalities, if they exist, are mediated by governments' enforcement efforts (Baird, Ferreira, Özler, & Woolcock, 2013;Paiva et al., 2016). Conditionalities need to be simple in order to be understood by beneficiaries and enforceable in large-scale programmes. But conditionalities that meet these feasibility requirements are too blunt to produce noticeable welfare effects. It seems, paradoxically, that for conditionalities to generate clear effects on their own, they would need to be tailored to beneficiaries' situation, making them hard to administer. Head-on comparisons of conditional and unconditional transfers, in contrast, finds superior welfare outcomes with conditionalities (Baird, McIntosh, & Özler, 2011;de Brauw & Hoddinott, 2011), and research suggests that conditionalities might be particularly adequate to deal with intergenerational conflicts and intra-household agency problems (Bursztyn & Coffman, 2012;Duflo, 2003).
While the case for justifying conditionalities due to their welfare effects is far from obvious, the possible effects of attaching conditionalities to government transfers are not limited to welfare improvements. Conditionalities also have implications for the administration of social programmes. On the positive side, conditionalities require the development of coordination mechanisms between different branches of government responsible for social services and between different levels of government. 4 Conditionalities might require that different bureaucracies operate in tandem and share information, and might also induce data collection and analysis that can shed light on the target population and help rationalise and plan social policy. As the case of Brazil's Bolsa Família shows, such coordination requirements can lead to the creation of a shared governance infrastructure that can generate further advances in social policy formulation and administration (Colin, Pereira, & Goneli, 2013). These advances, in turn, can also help spread the gospel of good governance (Grimes & Wängnerud, 2010).
On a more negative note, conditionalities might be too much of a burden for governments, though this would be a more pressing concern in lower-income countries than the ones we study Schubert & Slater, 2006). Even if the burden is not unsurmountable, conditionalities make the programmes more expensive and harder to implement (Caldés, Coady, & Maluccio, 2006;Freeland, 2007).
Conditionalities might also exclude the poorest of the poor in any country, especially when access to the services required by the conditionality is difficult. Simply put, the imposition of conditionalities changes the nature of the 'bureaucratic encounters' between potentially eligible individuals and the state apparatus (Kahn, Katz, & Gutek, 1976), often for the worse. Work on CCTs in the developing world has found that conditionalities constitute a significant de facto administrative burden that is associated with lower levels of receipt of the benefit and with significantly worse outcomes (Heinrich, 2015), and that the burden of complying with conditionalities is born disproportionately by women (Molyneux, 2007). In some contexts such requirements are often criticised as ideologically motivated attempts to 'influence the behaviour of the poor' (Calabrese, 2016;Ochs, 2015), or simply another instance of administrative burden imposed on individuals as means to prevent and limit access to social services (Lipsky, 1984). Hence, in these negative cases, one might ask whether beneficiaries would be better off with no programme at all than with a programme that burdens beneficiaries with conditionalities.
In practice, though, conditionalities might be more of a political requirement than a technical one. There is substantial evidence that CCTs are sufficiently popular with beneficiaries to significantly increase the likelihood of these voters supporting incumbent candidates (De La O, 2013;Diaz-Cayeros, Estevez, & Magaloni, 2009;Manacorda, Miguel, & Vigorito, 2011;Zucco, 2013). Less has been said about how non-beneficiaries view these programmes. 5 While CCTs have proven popular, and have contributed to a decline in inequality in many countries in recent years (Lopez-Calva & Lustig, 2010), in no country are CCT beneficiaries a majority of the population. The expansion and continuation of CCTs, therefore, depend on the supportor at least acquiescenceof non-beneficiaries.
One particularly strong argument for conditionalities is that they make CCTs more palatable to those footing the bill , increasing both public and legislative support for such programmes (De La O, 2015). In the next section we examine possible mechanisms behind these arguments, but if the argument is true, conditionalities might help ensure the survival of these safety net programmes in periods of economic hardship, such as the post-commodity supercycle period that has brought sluggish growth to most of the developing world and increased political polarisation.
Conditionalities support for transfers 529

Why conditionalities might increase support for transfers
Why would conditionalities imposed on beneficiaries affect attitudes of non-beneficiaries? The literature on redistribution based on the Meltzer and Richard (1981) model offers little insight into why features of transfers that do not directly affect their levels of redistribution should matter. We turn, instead, to non-economic differences between individuals, which are known to help explain individual preferences towards redistribution in the United States (Gilens, 2000), and can explain differences in levels of redistribution between the United States and Europe (Alesina & Glaeser, 2004).
Our main hypothesis is an extension of the idea that individuals tend to be more generous toward others who share common characteristics (Alesina & Giuliano, 2009;Fong & Luttmer, 2009;Luttmer, 2001). The ultimate psychological driver of this mechanism may lie in biased perceptions of the worthiness of beneficiaries. People are generally more willing to support 'industrious' poor through charity or state redistribution policies than by supporting the 'lazy' and 'unworthy' poor (Fong, Bowles, & Gintis, 2006;Gilens, 2000), and there is evidence that one tends to see members of one's own race as more deserving (Fong & Luttmer, 2009). In fact, and perhaps more generally as Milanovic observes, 'inequality may matter when people perceive each other as equals' (Milanovic, 2005, p. 155). Here, we extend this mechanism to other societal differences beyond just race and propose that the more similar those 'being helped' are perceived to be, the more likely one is to be willing to help them.
This also resonates with studies that examine how the structure of inequality affects levels of redistribution. Previous studies show that differences between groups, more than overall levels of inequality, shape the provision of public goods (Baldwin & Huber, 2010), and that when economic inequality overlaps with ethnic differences there is less support for redistribution (Morgan & Kelly, 2017). Lieberman's (2003) argument regarding the interplay of racial and regional politics is also particularly relevant. In Lieberman's study, racial disparities across regions in Brazil undermined federal redistribution efforts. Had redistribution been attempted mainly among more homogenous units, it would have led to more buy-inand, ultimately, increased compliance. Regionalism, in Lieberman's study, served as a proxy for racial, ethnic, and linguistic differences, and while these variables are often correlated, regionalism can also independently capture cultural (and other) non-racial attributes that define one's identity while informing the definition of the other. It follows, therefore, that individuals may also be less likely to support others in far-flung regions because they are perceived as being different.
Our hypothesis is that conditionalities make those who are being helped more worthy of the assistance in the eyes of non-beneficiaries. If the 'otherness' mechanism mentioned above is correct, we then expect that conditionalities will not always increase support for redistribution, or at least not always at the same rate. From the point of view of the non-beneficiaries, attaching conditionalities to transfers should generate larger increases in support for the transfer when nonbeneficiaries are perceived as different (and not very worthy of support), but little or no effect if beneficiaries are perceived as similar (and therefore worthy of support). Thus, the governing hypothesis in this article is that conditionalities have larger effects when individuals think of themselves as different from beneficiaries.
We define the 'conditionality premium' as the difference in support for conditional over identical but unconditional transfers, and test the following three implications of this general hypothesis: H 1 (Average effect): The conditionality premium is never negative. H 2 (Heterogeneous effects): The conditionality premium is larger among non-beneficiaries who are likely to think of themselves as different from beneficiaries, such as those respondents with higher socio-economic status (SES). H 3 (Otherness mechanism): The conditionality premium is larger when non-beneficiaries are primed to think of beneficiaries as being from distant places (and consequently different from themselves).

Empirical approach
There are many reasons why CCTs might be popular among the better off, not all of which relate to the fact that CCTs are conditional. CCTs, for instance, typically cater exclusively to those citizens whose basic needs are unmet, and experimental studies have shown that need can elicit sympathy from those who are better-off (Bowles & Gintis, 2000). Another factor is that in most of the developing world, CCTs are relatively new programmes, often implemented by a dedicated bureaucracy, in a non-clientelistic way, and frequently with funding and technical support of international organisations (De La O, 2015). As such, these programmes tend to be tightly monitored. They also tend to be perceived as more efficient because they provide small benefits to a relatively large number of beneficiaries. For this reason, simply comparing non-beneficiaries' perceptions about existing conditional and unconditional transfers will not address our specific hypotheses and answer our question of interest.
Determining whether a 'conditionality premium' exists, whereby conditionalities attached to transfers affects levels of support/acceptance among non-beneficiaries, requires a direct comparison of conditional and unconditional transfers that are otherwise identical (that is have the same eligibility criteria, and the same benefits). 6 Such a controlled comparison is best (and possibly only) achieved in an experimental setting, in which the conditional or unconditional nature of the transfer is manipulated by the researcher.
In order to test our hypotheses, we conducted survey-experiments in Brazil and Turkey. Although this is a somewhat unusual comparison in the academic literature, the two cases share many important characteristics. Both are large and heterogeneous upper-middle income countries that democratised in the 1980s. Both countries have comparable per capita income and levels of development. 7 Though income inequality (as measured by the Gini) is lower in Turkey than in Brazil, both countries are more unequal than most of their neighbours. 8 Crucially, for the present study, both countries show considerable regional and ethnic/racial heterogeneity. The Brazilian North and Northeast and the Turkish Eastern provinces are much poorer than the more developed South and West, respectively. Non-whites in Brazil and Kurds in Turkey are considerably overrepresented among poorer segments of the population that potentially benefit from direct government cash transfers. 9 In both countries, economic differences between regions and ethnic groups are pronounced, and strongly correlated with each other. Brazil and Turkey, therefore, are similar enough to serve as replicates of the study while also allowing us to probe the scope conditions of our hypotheses. Given the salience of an ethnic cleavage in Turkey and a racial cleavage in Brazil, studying both countries provides us with information on whether different types of social heterogeneity can be drivers of the otherness mechanism. This point is worth emphasising: in this paper, the comparison between Brazil and Turkey allows us to examine whether the results hold in countries that are comparable in many respects, but in which social heterogeneity' means different things.
Both countries have also had experience with CCTs. 10 In Brazil, CCTs began at the local level in the mid-to late-1990s, but by the end of Fernando Henrique Cardoso's second term, the Federal Government was already maintaining two relatively well-established programmes that covered approximately five million families. After 2003, during Luiz Inácio Lula da Silva's government, these programmes were merged with other subsidies into a single unified Bolsa Fam'lia Program, which gradually expanded to cover 14 million families. Today, the programme has a non-conditional component for the very poor, and a conditional component that imposes health and education conditionalities for those marginally better-off families with children.
CCTs were first introduced in Turkey in 2002, with the support of the World Bank, in order to mitigate the impact of the economic crisis on the poorest segment of the population. After a pilot, the Şartl Nakit Transferias the programme is generally referred toexpanded initially to randomly selected provinces and then to the entire country. Since 2007, the programme has been financed by the governments General Directorate of Social Assistance and Solidarity (GDSAS) Funds. Currently, the Turkish programme benefits more than one million families and includes transfers geared towards children's education and health as well as to pre-natal care.

Conditionalities support for transfers 531
Impact evaluation studies of the programmes in both countries have been largely positive, emphasising their role in reducing poverty and inequality11though some are circumspect with regard to CCTs effects on the empowerment of women or minority groups (Baykal, 2014). Other research has examined the political aspects of CCTs. Most work on Brazil has found positive pro-incumbent electoral effects at the national level (Hunter & Power, 2007;Zucco, 2013, inter alia), though it is not clear that similar electoral results hold at other levels, and some have highlighted that CCTs might generate backlash effects against incumbents in some groups (Correa, 2015). There has been less research on the political effects of CCTs in Turkey. Aytaç (2014) found that in a setting of multiparty competition, there are incentives for the incumbent party to channel disproportionately more CCT resources to districts with an ideologically close challenger, while Baykal (2014) concluded that Turkish CCTs have not suffered significantly from political manipulation except, perhaps, for an excessive coverage in Kurdish districts.
Our research approach consisted of polling non-beneficiaries of government transfers in order to evaluate a hypothetical government transfer targeted towards the poor. 12 We do this while manipulating the conditionalities attached to the transfer, the perceived racial/ethnic or regional differences between the respondent and potential beneficiaries. We acknowledge that participants in our study might have noticed the resemblance between the proposed policy and existing CCTs. We attempted to attenuate this connection by asking participants to evaluate a new programme. We also believe that any influence that the evaluation of existing programmes might have on respondents evaluation of the hypothetical programmes that were presented to them, should equally affect all treatment conditions in each study, and therefore have no impact on the treatment effects we estimate.
We designed and implemented three separate studies. 13 The first was embedded in nationallyrepresentative surveys in each country, and included one simple manipulation pertaining to the conditional nature of a hypothetical government transfer. This study focused primarily on assessing the existence and possible heterogeneity of the conditionality premium; as such, it provides a first assessment of Hypotheses 1 and 2.
The other two studies were fielded over the internet. In both of these studies, we not only manipulated the conditionality associated to the hypothetical transfer, we also introduced manipulations that sought to increase the perceived racial/ethnic differences and the geographical distance between respondents and potential beneficiaries. These two internet studies differ only in the way in which we manipulated distance and differences.

Study 1: the conditionality premium experiment
The first study consisted of a single experimental item embedded in nationally-representative surveys conducted in both countries. The goal was to determine whether a conditionality premium exists (Hypothesis 1), and whether it is larger in subgroups of 'privileged' respondents (Hypothesis 2).

Design.
This study employed a very simple experimental design, with a single response item and a single manipulation with two treatment conditions. Respondents were asked to evaluate their support for a hypothetical government transfer that was either unconditional or conditional depending on the treatment group to which the subject was randomly assigned. The question and answer options were as identically worded as possible in each language, and the value of the benefit was set to match lower-than-average values of CCTs currently in place in each country. The English translation of the two variants of the experimental questions are reported in Table 1.
In addition to the answer to this experimental item, which was recorded on a five-point scale, we have data on demographics and socio-economic status (SES) for each respondent. 14 4.1.2. Sample and data collection. We hired reputable local polling firms (Ibope in Brazil and TNS in Turkey) that conducted nationally-representative omnibus surveys, and which allowed for the inclusion of a single split-sample question. The surveys were conducted between December 2015 and January 2016 in both countries. The sample size was 2,002 in Brazil and 1,512 in Turkey.

Results.
We present the main results of Study 1 in abbreviated form, in Table 2. The table reports only the average treatment effects, which are measured as differences in average response of subjects in each of the two treatment groups. For simplicity, we treated the five-point answer scale as a linear variable in the estimation. Positive values indicate greater support for the conditional transfer. We standardised the outcome variable prior to analysis in order to facilitate the comparison with subsequent results.
The first row reports results for the whole sample. These are the average treatment effects in the survey-experiment, expressed in standard deviations of the outcome variable. For each country, we report estimates with and without controls for individual characteristics. 15 We employed difference-in -means tests to estimate the former, and linear regression with robust standard errors for the latter (Samii & Aronow, 2012).
Estimates of average effects are positive, indicating that conditional transfers do enjoy greater support than unconditional ones, in both countries. Results are similar with or without controls, as would be expected due to randomisation. Estimates are small, ranging from 0.04 standard deviations of the outcome variable in Brazil to 0.12 in Turkey, and only statistically significant in Turkey.
Results change once we consider the heterogeneous effects of the conditionality manipulation. Effects are two to three times stronger than the average among those with higher SES. 16 Although standardised effects among high-SES respondents are not particularly large, they now range between 0.11 and 0.21 and are statistically significant in both cases.
For the poorest respondents, the conditionality has very small effects that are not statistically significant. In the case of Brazil, these are actually negative. One explanation for this observation could be that most beneficiaries are included in this (lowest) SES stratum, and that for them, conditionalities are a burden. Even though we cannot directly exclude beneficiaries from this study, the heterogeneous effects are most likely not driven solely by a possible negative reaction by CCT beneficiaries. Treatment effects for those in the middle of the SES scale are very similar to what we observed in the lowest SES strata, and considerably different from results for high-SES respondents.
As a whole, results suggest that a conditionality premium exists (hypothesis 1), but that it is small and much more pronounced among the better-off in each country. If we consider that the high-SES respondents are probably 'more different' from the potential beneficiaries of the transfer than others, these results are compatible with the idea that the conditionality premium is a function of perceived differences between beneficiaries and non-beneficiaries (hypothesis 2).
This study is not without its shortcomings, and these are mostly related to the practical limitations of including experimental items in commercial surveys. We were limited to using a single manipulation, so we neither directly measured nor manipulated perceived differences between respondents and potential beneficiaries of the hypothetical transfer. As a result, we cannot say whether it is the perception of differences or any characteristic that correlates with SES that is driving the heterogenous effects we found. To overcome these limitations, we designed and executed two follow-up studies.

Study 2: otherness with sequential manipulations
In Study 2, we directly manipulate the level of perceived differences between respondents and potential beneficiaries. To accomplish this, we used a convenience sample in each country, recruited over the internet. This sample was not statistically representative of the better-off population in each country, but was considerably more diverse than what one would typically obtain using undergraduate students as subjects. As such, it is a sample suitable for examining the effects of our proposed manipulations.

Study design.
We employed a 3 Â 2 experimental design, in which we manipulated the perception of otherness (the perceived similarity between respondents and potential beneficiaries) and the conditionality of a transfer. There were three otherness conditions: a control group, one in which we highlighted the regional aspect of poverty, and one in which we highlighted its ethnic/ racial aspect. Our manipulation sought to induce respondents in the treatment conditions to think about the poor as being different from them in either a regional or ethnic/racial dimension. 17 The conditionality manipulation amounted to presenting respondents with a hypothetical cash transfer that was either non-conditional or conditional. The descriptions of each transfer are presented in Table 3, and were nearly identical to those in Study 1.
The two manipulations were implemented sequentially. The experimental items were embedded in a short survey that began with background demographic questions, information related to the presence of certain items in the household, and a standard question on attitudes toward redistribution. We also included a warning, stating that there would be an attention check among the subsequent questions. After these initial background items, respondents were asked two placebo questions on their knowledge of basic facts about poverty and inequality in the country. Those in the otherness control group received no further questions on the topic but others received a variant of either the racial or regional manipulation of otherness. 18 The conditionality manipulations were straightforward. Respondents were presented with either a conditional or non-conditional hypothetical cash transfer programme and asked about their levels of approval on a seven-point scale. The response to this question is the outcome of interest in the study. Immediately after the conditionality question, respondents were subjected to the attention screener. The survey concluded with a few additional filler questions and with the collection of information for participation in a lottery.

Sample and data collection.
In each country, we placed a series of ads on Facebook offering the chance to win an iPad in exchange for completing an academic survey, as in previous research by Samuels and Zucco (2014). Facebook users who clicked on the ads were sent to a page on the survey platform Qualtrics, which contained the informed consent form. Those who accepted the terms of the study were subsequently referred to the survey itself.
In order to test the otherness hypothesis (hypothesis 3), we were primarily interested in the reactions of individuals who could be considered 'advantaged' in both the racial/ethnic and regional dimensions. Hence, we attempted to recruit only respondents residing in areas with higher socioeconomic status and who belonged to the majority racial/ethnic groups in each country. This was done by employing two different strategies simultaneously. First, we employed custom targeting of Facebook ads, focusing only on adult users with at least a high-school diploma and who lived in the advantaged regions of each country. Second, we employed regional and ethnic self-identification questions early in the survey. If a respondent reported being younger than 18 years of age or living abroad, the study was interrupted. Respondents who were flagged as being from a disadvantaged ethnic or racial group or from a disadvantaged region were excluded from the otherness manipulations and dropped from the analysis. 19 Eligible respondents were then assigned to one of the six treatment conditions using Qualtrics' random number generator.
We further restricted the sample by employing two ex-post exclusions. First, we eliminated from the analysis all individuals who reported having a CCT beneficiary in the family. 20 Second, we restricted the sample being analysed to respondents who also passed the attention screener. As our manipulations provide respondents with different information, they will only have an effect if the information was minimally processed. While we do not examine directly whether respondents in fact processed the information, we argue that paying attention to a subsequent question (the attention check) is a proxy for having paid attention to previous items (the manipulations). 21 Data were collected in May and June of 2015 in Turkey, and in September and October of the same year in Brazil. In Turkey, we had a total of 693 respondents, 230 of which qualified for the experiment sample. In Brazil, we had a total of 687 responses, 266 of which were included in the experiment sample.

Results.
We assess support for the otherness hypothesis by examining the interaction effects of the two manipulations in the experiment. For ease of presentation, we do this graphically in Figure 1. We present one figure for each country, and each figure contains three pairs of columns. Each pair represents one otherness condition, control in the centre, racial/ethnic to the left, and regional to the right. Within each pair, the two columns depict support for the non-conditional and the conditional transfer. Confidence intervals of the estimates are shown for the conditionality treatments; the difference between the lighter-coloured bars to the darker bars adjacent to them represents the conditionality premium.
The otherness hypothesis leads us to expect that the conditionality premium will be larger in the regional and ethnic/racial manipulations than in the control group. In other words, our hypothesis is that the difference between the lighter and darker bars in the central pair should be smaller than in the other pairs.
The hypothesis is corroborated for the regional manipulation in both countries. The conditionality premiums increase considerably relative to what is observed in the otherness control group. The ethnic manipulation in Turkey also generated the expected effects, though in Brazil the racial manipulation did not.
We also observe a positive conditionality premium in both countries in all otherness conditions, bolstering the results reported in the previous section. The premium for those in the otherness control condition is larger than what we observed in Study 1, similar to what we found among high-SES respondents, though not statistically significant (likely due to the smaller sample sizes). The existence of larger effects in the internet sample than in the nationally representative sample of Study 1 was expected, due to the income and wealth differences between the two samples.
Still, while our strategy of providing information highlighting some type of difference between the respondent and the poor is reasonable, it might be too weak for our sample sizes, especially given that we lost many respondents due to lack of attention. To address the possibility that our otherness manipulation was not achieving the desired effect, we designed a third study, with a more direct manipulation of otherness.

Study 3: otherness with simultaneous manipulations
In Study 3, we implemented a more direct variant of the otherness manipulation. Instead of priming respondents to first think of themselves as different from the poor, we informed respondents of some aspect of the identity of potential beneficiaries of the hypothetical transfer. This study was also fielded over the internet with a sample of respondents recruited using Facebook ads.  4.3.1. Study design. As in Study 2, we employed a 3 Â 2 design, in which we manipulated the perception of similarity between beneficiaries and non-beneficiaries, as well as the conditionality associated with a hypothetical cash transfer. There were three otherness conditions: one baseline control group, one in which we highlighted that the transfer would disproportionately benefit individuals of a different race/ethnic group, and another asserting that beneficiaries lived, disproportionately, in a different region of the country. As in the previous studies, the conditionality manipulations included the non-conditional and conditional hypothetical cash transfer. The experiment consisted of six conditions produced by the combination of the two experimental manipulations. Unlike Study 2, both manipulations were implemented simultaneously. Each respondent, therefore, saw one of the six variations of the experimental item, as described in Table 4.
The experimental items were embedded in an online survey that was otherwise identical to the one used in Study 2. We assessed the otherness hypothesis (Hypothesis 3) in a core sample restricted to respondents from advantaged regions and advantaged racial/ethnic groups who passed the attention screener.
Data were collected in November and December of 2015 in both countries. In Brazil, there was a total of 949 respondents, 437 of which qualified for the experiment sample. In Turkey, there was a total of 903 responses, 273 of which were included in the core experiment sample. 22 Figure 2 reports the average support for the hypothetical transfers in all six treatment conditions in the study, by country. As in all the previous studies, the conditionality premium in the baseline otherness condition is always positive, but not particularly large. The standardised point estimates of this effect are very similar to those found in Study 2. Due to the larger sample sizes, these estimates are associated with considerably smaller p-values and the result is now statistically significant in Brazil.

Main results.
The conditionality premium is roughly twice as large for those who received the regional manipulation and statistically significant in both countries. The racial manipulation generates similar Table 4. Wording of the experimental items in Study 3

Non-conditional × Control
The federal government is studying a new social programme that will pay benefits of approximately [R$100/TL 50] to every poor person in the country.

Conditional × Control
The federal government is studying a new social programme that will pay benefits of approximately [R$100/TL 50] to every poor person in the country, provided the recipients meet a number of conditions, such as ensuring children attend school and make regular visits to the doctor.

Non-conditional × Region
The federal government is studying a new social programme that will pay benefits of approximately [R$100/TL 50] to every poor person in the country. This transfer will disproportionately benefit [Brazilians in the Northern and Northeastern regions/Eastern Anatolia].

Conditional × Region
The federal government is studying a new social programme that will pay benefits of approximately [R$100/TL 50] to every poor person in the country, provided the recipients meet a number of conditions, such as ensuring children attend school and make regular visits to the doctor. This transfer will disproportionately benefit [the Northern and Northeastern regions/Eastern Anatolia]. Non-conditional × Ethnic/Racial The federal government is studying a new social programme that will pay benefits of approximately [R$100/TL 50] to every poor person in the country. This transfer will disproportionately benefit [non-whites/the Kurdish population].

Conditional × Ethnic/Racial
The federal government is studying a new social programme that will pay benefits of approximately [R$100/TL 50] to every poor person in the country, provided the recipients meet a number of conditions, such as ensuring children attend school and make regular visits to the doctor. This transfer will disproportionately benefit [non-whites/the Kurdish population]. effects in Brazil, but yield less pronounced increases in the conditionality treatment in Turkey. The regional manipulation, once more, produced more consistent increases in the conditionality premium than the racial/ethnic manipulation, which is something we discuss in the final section.

Discussion of the results
Results for the existence of the conditionality and heterogeneity of the conditionality premium were similar in both countries in Study 1. These findings provide some evidence that a conditionality premium exists and that this premium tends to be concentrated among the better-off members of each polity.
The otherness experiments (Studies 2 and 3) corroborated this proposition that the average conditionality premium is small. While some individual results are not statistically significant on their own, together they consistently showed that the conditionality premium increased considerably when subjects were primed to think of beneficiaries as being different from them. In other words, results suggest that the conditionality premium is indeed heterogenous, and linked to perceptions of similarity between respondents and potential beneficiaries.
To facilitate the interpretation of these results, we pooled the data from these two studies, and estimated the conditionality premium for the different otherness conditions (Figure 3).
Pooling these two studies makes for a cogent summary of the results: the regional manipulation more than doubled the conditionality premium in both countries. The ethnic/racial manipulation doubled the the conditionality premium in Turkey and produced a smaller, though still substantial, increase in Brazil. If respondents are primed to think of themselves as different from beneficiaries, the conditionality premium is always unambiguously different than zero. In the case of the regional manipulation of otherness, the increase in the conditionality premium is also statistically significant relative to the conditionality premium in the control group.
All results based on the otherness experiments seem to indicate that priming regional differences has a more consistent effect on the conditionality premium than priming racial or ethnic differences. Our rationalisation of this result is that, at least in the two countries studied, regionalism can be  a stand in for any of many politically relevant differences between groups, of which racial and ethnic differences are but one of the possible driving forces. More importantly, however, given the strength of the prior expectations that conditionalities should be consequential for support, we consistently find that the conditionality premium is very small. While we acknowledge that it is not obvious how one should make this assessment, all three studies indicate that simply adding a child-related conditionality to a transfer hardly changes the levels of support we observe. In this sense, the political-economy argument about why one should condition benefits does not carry much empirical weight, however intuitive it may seem, and should not be the driving force behind the adoption of conditionalities. Caution is particularly advisable considering that conditionalities represent an added burden to potential beneficiaries, who are already among the most vulnerable segments of the population.
One limitation of our studies is that we did not stress the administrative costs of enforcing conditionalities. On the other hand, there might exist other positive effects from conditionalities, be them positive spillover effects or even greater demand or concern for the quality of services that could generate incentives for better governance. Stressing these positive and negative effects of conditionalities could affect estimates of the conditionality premium from Study 1, and are a promising path for future research.
It is also true that political polarisation was on the rise in both countries when data was collected. In Turkey, in particular, the refugee crisis might have increased the perceptions of otherness, though our survey instrument focused explicitly on Turks. In Brazil, increased political polarisation also led to greater divergence in evaluations of Bolsa Família, with supporters of the Worker's Party holding Figure 3. Estimates of the the effects of the conditionality manipulation (that is the conditionality premium) and 95 per cent confidence intervals for the different otherness conditions, after pooling the data from Studies 2 and 3, by country. Notes: Effects are reported in standard deviations of the outcome variable. much more favourable views of the programme than their antagonists. While these facts might have affected levels of support for the programmes, they should not directly affect the estimates we obtained through the random manipulation of otherness.

Conclusions and implications
The debate over the long-term consequences of CCTs is ongoing, but there is widespread agreement that CCTs can be an important instrument among a wider set of social policies. In some instances, at least, CCTs have contributed to positive outcomes in fighting malnutrition, in keeping children in school, and in improving the lives of the neediest. But even successful programmes are subject to changing political winds. This risk is exacerbated by a global climate that has seriously impacted lower-income countries, most notably with the sharp fall of commodity prices after one decade of sustained increases. 23 Sound policies are often subject to the imperatives of governments' budget priorities and electoral incentives. In this context, a sound policy that is also politically viable is more likely to prevail over the long-term. While popular among beneficiaries, the political viability of CCTs depends on its acceptance by non-beneficiaries.
This article assessed whether and why the imposition of conditions on beneficiaries of transfers elicits support from non-beneficiaries. We examined whether a conditionality premium exists, and how it varies with the perceptions of differences between beneficiaries and non-beneficiaries. Results showed very consistent evidence that the existence of a conditionality premium is not as straightforward as the literature implies. Once we focus squarely on the addition of conditionalities while keeping other transfer characteristics constant, the conditionality premium is fairly elusive, at least in highly heterogenous societies such as Brazil and Turkey.
There are four important take-aways from these results. First of all, our results mean that the political-economy justification for conditioning benefits caries very limited weight. This result adds to other work that has found that the welfare implications of conditioning are also limited. This does not mean, however, that there are no worthwhile reasons to condition benefits. If conditions can spur the development of government capacity, increase beneficiaries self-perceived legitimacy, or even solve intra-household agency problems, they might very well be justified.
Second, the evidence that the conditionality premium is stronger among the better-off is both compelling and consistent. This suggests that some attribute that correlates with income serves as a moderator of the conditionality premium. Thus, even if one is reluctant to accept our hypothesis that perceptions of similarity between beneficiaries and non-beneficiaries are the driving force behind this result, the fact that the conditionalities have heterogenous effects on approvals of transfers should give us pause when considering whether to condition a transfer. It suggests, on the one hand, that conditionalities might shore up support among those more averse to redistribution, which is a politically useful bit of information. On the other hand, it suggests that this effect might, in some polities, exist only among a very small subset of the population and might be quite small. If the political gains are uncertain, one should, perhaps, reconsider the potential downsides, such as the added administrative burden imposed on potential beneficiaries.
Third, most factors that are known to affect the amount of redistribution a society will accept are immutable, or change only very slowly, and are therefore out of the hands of policy-makers. This article focused, in contrast, on whether malleable characteristics of government transfers affect support for them. Knowing that certain redistributive policies might elicit more support (and less resistance) from the better-off, and understanding why this is the case, are potential game changers when it comes to designing poverty-reducing policies. These characteristics can be manipulated by policy-makers in the short-term to overcome some types of political resistance to programmes with potential positive welfare effects.
Finally, our studies show that referencing differences between non-beneficiaries and beneficiaries in potentially politically relevant dimensions leads to an increase in the conditionality premium. While this does not prove that all the heterogeneity discussed above is due to these perceptions, it does suggest that it is a relevant mechanism in the variation of the conditionality premium. Moreover, in both countries, regionalism appears to be a stronger and more consistent driver of the variation in the conditionality premium than race/ethnicity. We believe this makes sense because regionalismat least in the countries studiedis correlated not only with race, but with many other political factors. As such, regionalism might be a stand-in for other relevant political cleavages.
The fact that perceived differences between beneficiaries and non-beneficiaries have an impact on preferences on redistribution also has wider implications. There is a growing body of literature that shows that political identities are socially constructed. When this is considered in the context of our results, we conclude that support for redistributive programmes can be affected by the design of transfers and that divisive political discourse can be particularly detrimental to redistribution efforts.

Acknowledgments
Early versions of this paper were presented at seminars at Rutgers and PUC Chile, and at the 2014 and 2016 Meetings of the APSA, 2014 meeting of the ECPR, as well as the 2017 REPAL Conference, and LASA Congress. The authors would like to thank participants at these venues, as well as Andy Baker, Clèment Gèvaudan, Jere Behrman, Kerem Yildirim, and Thad Dunning for comments on earlier drafts. The studies were originally approved by the IRBs of Rutgers University (E12-621) and Koç University (

Disclosure statement
No potential conflict of interest was reported by the authors.