Attitudes to immigrants in South Africa: personality and vulnerability

ABSTRACT While many countries across the world face increasing numbers of immigrants, the literature on attitudes to foreigners and immigrants focuses on Western countries. This article tests broad insights from Western countries in a specific non-Western context – South Africa, a country marked by sporadic violence against some immigrant groups. This provides an important validity check. Data from the 2013 South African Social Attitudes Survey and the 2013 World Value Survey are used to model attitudes to immigrants. In line with research on Western countries, individual personality is associated with differences in attitudes; people in vulnerable positions and those who lack a personal support mechanism are more likely to oppose immigrants. When implemented to reflect the specific context, research on attitudes to immigrants appears to generalise to non-Western contexts.

bilateral agreements with neighbouring countries that precluded settlement and family reunification. The racial criterion was toned down in 1986 under international pressure, with a requirement to assimilate to any population group (Segatti 2011). Implementation was opaque, however, and the 1986 amendment did not change the practice of racial selection (Wa Kabwe -Segatti 2006). What is more, the South African state did not have the capacity to enforce the back gate to restrict immigration numbers in the long term.
After 1986 it became increasingly clear that the two-gate system neither met economic demand adequately, nor was there a means to deal with non-economic migrants like refugees from countries in the region, such as Mozambique. However, there was no change in policy beyond unifying existing policy into a single act in 1991. When the African National Congress (ANC) assumed power in 1994, immigration was low on an agenda focusing on economic transition (Segatti 2011). Since 1997 South Africa has a new constitution lauded for its progressive stances on inter-group relations, but despite increasingly obvious inadequacy to deal with contemporary migration challengesrefugees, labour shortages -, immigration policies were only revised in 2002 and 2004, and then only to a limited extent despite a progressive preamble to the Immigration Act claiming to encourage highly skilled workers. The 2007 and 2011 amendments failed to introduce policies that facilitate the recruitment of skilled workers (Nkomo 2014), and only increased enforcement mechanismstargeted at irregular immigrants but applicable to all immigrants, including the skilled workers the economy needs, and despite insufficient state capacity to enforce these policies. In this sense, South Africa has immigration policies that remain influenced by the apartheid-era and largely discourage immigration despite ostensible commitment to recruit skilled immigrants (Segatti 2011).
The result is that, like in most European countries, there are few legal channels to migrate to South Africa except for claiming asylum. South Africa maintains a relatively open asylum system that provides the right to work and move freely within the country after applying for asylum. Decisions often take long, and many rejected asylum seekers are thought to stay in the country, although officials frequently stop suspected rejected asylum seekers for identification and deport them (Amit 2015). Without identity documentsbecause their application is pendingand bank accounts, many immigrants are forced into the informal sector (Solomon and Kosaka 2013).
A common narrative is that negative attitudes in South Africa are linked to the postapartheid nation-building project (Crush 2001;Peberdy 2001), a context of low trust (Steenkamp 2009). The intuition is that negative identification against immigrants is a means to increase positive identification with the new in-group: a multi-racial South Africa. Violence against immigrants is thus regarded as a perverse upshot of the end of apartheid (Harris 2002). Given that violence against immigrants and negative attitudes are a recurring phenomenon, this explanation has face validity (Dodson 2010). By contrast, it fails to explain why anti-immigrant sentiments and violence against immigrants also occur in established Western countries without such a nation-building project. Leaving this counterfactual aside, the link between negative attitudes and violencebehaviouris one of the big unknowns in South Africa, as elsewhere: It is clearly not direct (see contributions in Nelson 2009;Carlsson and Eriksson 2016;Piper and Charman 2016).
Several contributors have noted that violence against immigrants in South Africa is directed against black Africans, not white immigrants (Sharp 2008;Matsinhe 2011). This may be a lingering influence of apartheid and its preference for white immigrants, but more plausible is probably a difference in opportunity: violence directed against the immigrant group that is 'there'. At the same time, opportunity alone is not a sufficient explanation, given that negative attitudes to different immigrant groups are dissimilarand that these negative attitudes are thought to be causally prior to violent behaviour . Piper and Charman (2016) use a survey of shopkeepers in areas affected by violence against immigrants, and find that the experience of violence is more about undercutting competitors than about nationality or being an immigrant. Nonetheless, the media and politicians systematically politicise against immigrants (Danso and McDonald 2001), unlike the more balanced coverage found in Western Europe (Van der Brug et al. 2015).

Theory and expectations
Over the years many reasons have been identified as to why some individuals are more likely to oppose immigrants and minority groups than others, amongst which (authoritarian) personality, perceptions of relative deprivation, and contact with immigrants find consistent empirical support (Pettigrew 2016). Historically, with the so-called authoritarian personality the focus was on individuals without regard to their social connections (Adorno et al. 1950). Although often frowned upon in social sciences, the influence of personality on attitudes points to genetic influencesin particular gene-environment interactions (Freese 2008). Recent contributions have both reaffirmed that personality plays a role for attitudes to immigrants (Gallego and Pardos-Prado 2014;Johnston, Newman, and Velez 2015), and placed them into a wider framework of basic personal values (Schwartz et al. 2014). Put simply, individuals who value tradition and conformity and reject universalism are more likely to reject immigrants. These tend to be individuals who value obedience to law and have conservative or right-wing preferences. Similarly, relevant personality may also be reflected in sociological liberalism (Crouch 1999), an association generally only implicit in recent research on the link between personality and attitudes to immigrants. Sociological liberalism is a broad outlook on inter-group relations more generally drawing on ideals of equality and social justice, supporting a peaceful cooperation and coexistence of different groups in society. Here, the intuition is that this general positive outlook on inter-group relations is associated with positive attitudes to immigrants: Immigrants are regarded largely the same way as other groups falling outside the cultural mainstream (Kalkan, Layman, and Uslaner 2009;Schwartz et al. 2014).
Personality expectations: People with more authoritarian personality have more negative attitudes to immigrants.
People with sociologically liberal views have more positive attitudes to immigrants.
When social interactions are taken into consideration, two broad explanations have been put forward. On the one hand individual and group interests are highlighted, on the other hand contact between groups is regarded in less competitive terms. In the former case, different groups in society are regarded as competitors over scarce resources (Blalock 1967). The larger an out-group is, the more competition can be expectedboth actual and perceived competition. Such competition can be at the individual level such as when an individual's wages are reduced, or at the societal level, such as when social benefits and the welfare system are put under strain. The psychological impact of this competition can lead to feelings of threat and prejudice, and it is expected that individuals more directly affectedby being in a vulnerable positionexpress more negative attitudes to their competitors and may feel bitter about their perceived relative deprivation (Alexseev 2015; Poutvaara and Steinhardt 2015;Gorodzeisky and Semyonov 2016; or Steenkamp 2009 on South Africa). Pettigrew (2016) highlights that relative deprivation is relevant, which is why economic outsiders tend to be more negative to immigrants (Jaime-Castillo, Marqués-Perales, and Álvarez-Gálvez 2015). In Western countries, the level of education is often used as a proxy for vulnerable positions in the labour marketusually under the assumption that most immigrants have low skills and thus compete with low educated people in the labour market, an assumption which does not necessarily hold outside Western countries (Facchini, Mayda, and Mendola 2013;Barceló 2016;Meseguer and Kemmerling 2016). While studies in Western countries usually find that individuals with higher levels of education are less opposed to immigration (Hewstone, Rubin, and Willis 2002;Dancygier and Laitin 2014), education is used to capture different mechanisms, from being more exposed to economic competition, to a supposed liberalising effect of education, or simple social desirability biases (Janus 2010;. Hello, Scheepers, and Gijsberts (2002) demonstrate that the 'effect' of education varies across countries; Weil (1985) argues the 'effect' of education may be limited to liberal democracies. Empirically, the different mechanisms are difficult to disentangle, although the common use as a proxy for labour-market competition has been challenged both conceptually and empirically (Pecoraro and Ruedin 2016).
Attitudes reflect perceptions of threat and deprivationnot necessarily objective competition -, which is why subjective measures are important: feeling vulnerable or perceived social position, although the subjective and objective measures are likely to be somewhat associated. Feelings of relative deprivation and vulnerability may be worse for individuals who lack a social support network, who feel relatively isolated in addition to relatively worse off. An individual's network may constitute a resource and coping mechanism commonly referred to as 'social capital' (Portes 1998;Stolle and Harell 2013;or Gordon and Maharaj 2015;and Steenkamp 2009 on South Africa). If individual-level competition dominates, we can expect a multiplicative effect (interaction) between being in a vulnerable positionexposed to competition with immigrantsand having a support network. In this case the support network can alleviate the impact of competition. If perceptions of society-level competition dominate (Hainmueller and Hopkins 2014), we can expect a cumulative effect where having a support network reduces perceptions of threat irrespective of individual exposure to competition.
Perceived relative deprivation expectation: People with insufficient income and lack of support from their neighbours have more negative attitudes to immigrants.
Commonly presented as a complement to threat theory, contact with immigrants has consistently been shown to reduce negative attitudes (Pettigrew and Tropp 2006). Relevant contact includes friendship, acquaintance, and having friends who are friends with immigrants (Turner et al. 2007;Hewstone and Swart 2011). While the findings are unambiguous, it is usually difficult to rule out self-selection and thus ascertain causal order (Enos 2016): Individuals with positive attitudes to immigrants are more likely to seek and accept friendship with immigrants in the first place, or experience such relationships positively and describe them as friendship. For this reason it may be more appropriate to capture contact with acquaintances (rather than contact with friends) where individual control over the relationship is less pronounced. What is more, increasing diversity may lead to social withdrawal (Laurence and Bentley 2016), actually leaving individuals more vulnerable to (perceived) competition with immigrants. With cross-sectional data, the present article will be unable to address this issue, and in common with the literature assumes that the main influence is from contact to attitudes and not the other way around (Kaufmann and Harris 2015).
Contact expectation: People who have acquaintances from other countries have more positive attitudes to immigrants.
Negative attitudes can also occur without contact, as highlighted by social identity theory where negative attitudes are regarded a symptom of in-group favouritism (Abrams and Hogg 2010). This is a common narrative in qualitative studies and comments on the situation in South Africa (Crush 2001;Peberdy 2001): negative attitudes to immigrants as an unintended consequence of the post-apartheid nation-building project. Creating a positive identity for the new in-group (South Africans) is facilitated by having negative attitudes to an out-group (immigrants). Such negative images of immigrants can be actively mobilised by political actors (Van der Brug et al. 2015; Gordon 2016 on South Africa). Adida (2011) highlights that the lack of obvious cultural difference can be a reason why community leaders politicise difference to other groups. In post-apartheid South Africa racial and ethnic identities remain strong, and individuals who regard their group as threatenedbecause community leaders politicise identitymay project negative feelings on immigrants (Gordon 2017).

Data and methods
While the general expectations outlined are formulated independent of context, we need to adapt the operationalisation to the specific context: Simply re-using the same variables could lead to an invalid test of the theories. Data from the South African Social Attitudes Survey (SASAS) 2013 are used to examine negative attitudes to immigrants (HSRC 2013). The data are nationally representative, collected using the 80,787 enumerator areas of the 2001 Census as primary sampling units (PSU), drawn with probability proportional to the number of non-vacant resident dwellings in the PSU. Dwellings and individuals in households were drawn randomly (HSRC 2013). As is common in such surveys, individuals in institutions (hospitals, prisons) were not sampled, and only South African citizens were targeted. There are 2885 respondents, aged 16 and older, all interviewed face-to-face. Data were weighted to account for variation in the population size of the selected PSU, and benchmarked to official statistics on gender and age. The response rate was 88%. To ascertain the robustness of the findings, I replicated the analysis with data from the World Value Survey (WVS) 2013 as far as possible (World Value Survey Association 2013). There are 3531 respondents aged 16 and older in the WVS, with the sampling strategy equivalent to that of the SASAS.
The SASAS includes several questions on attitudes to immigrants, and I construct an additive scale of anti-immigrant attitudes as the outcome variable where all variables carry the same weight (Cronbach α = 0.78, see Supplement S1 for details on the scale and other variables). This approach is common in existing studies on attitudes to foreigners in Europe and North America. The items used in the scale ask 'How much do you agree or disagree with the following statements?', each measured on a 5-point scale that is treated as continuous in the analysis: 'immigrants increase crime rates' [agree], 'immigrants are generally good for South Africa's economy' [disagree], 'immigrants take jobs away from people who were born in South Africa' [agree], 'immigrants make South Africa more open to new ideas and cultures' [disagree], 'immigrants bring disease to South Africa' [agree], and 'immigrants bring skills that are needed in South Africa' [disagree]. Values on the scale range from −5 to +2.6, with most responses between −2 and +2 and a mean of 0.3 (standard deviation 0.87). Higher values denote negative attitudes.
As a test of robustness, I replicate the findings with a single question rather than the scale: 'Immigrants take jobs away from people who were born in South Africa' (34% of respondents strongly agree with this statement). Like the scale, this single question measures negative attitudes. These results are included in the supplement alongside the results using the scale (Supplement S3 and S6).
For the replication with the World Value Survey, two questions are used as outcome variablesand analysed separately. The first is a statement with which respondents can agree, disagree, or neither agree or disagree ('When jobs are scarce, employers should give priority to South African people over immigrants', inverted). Fifty-one per cent of respondents agree with this statement. The responses are treated as a continuous variable. The second is part of the question on undesirable neighbours ('On this list are various groups of people. Could you please mention any that you would not like to have as neighbours?', respondents mentioning 'Immigrants/foreign workers'). The variable is binary, and 43% mentioned immigrants/foreign workers as undesirable neighbours. See Supplement S1 for details.
For the predictor variables, the following question was used to capture whether an individual is in a vulnerable position and ostensibly more exposed to competition with immigrants, reflecting qualitative work in South Africa (Adjai and Lazaridis 2013): having an insufficient income (self-declared, 'Sometimes people find that their income does not quite cover their living costs. In the last 12 months, has this happened to you?', binary variable). Fourty-one per cent of respondents report insufficient income. Supplement S1 also describes alternative variables used in bivariate analysis only. One question asks whether respondents consider themselves wealthy, using six response categories; one asks about income relative to other households in the neighbourhood, using five response categories; and I use an additive count of objective ownership of household goods and amenities as they are frequently asked in middle-and low-income countries. These variables allow me to differentiate absolute from relative deprivation.
Two questions capture access to support networks and community help and are combined ('How comfortable would you be asking a neighbour to lend you a cup of sugar if you needed it?', 'If you were short of money, how comfortable would you be asking a neighbour if you could borrow R20?'). For both underlying variables, 28% of respondents say that they are 'very comfortable'. Contact with immigrants is captured by 'How many acquaintances do you know who have come to live in South Africa from another country?'. Thirty eight per cent of respondents report 'none', and 9% report 'very many' such contacts. See Supplement S1 for details.
Personality is approached through individual beliefs and ideology in two variables, which were chosen on the basis that they resonate contemporary political debates in the country and because of their closeness to the theoretical constructs suggested in the literature: supporting the death penalty for murder (authoritarianism: 'People convicted of murder should be subject to the death penalty'; 21% of respondents agree or strongly agree); and whether they believe that it is (morally) wrong that gay men and lesbians live their own life as they wish (lack of sociological liberalism; 48% of respondents agree or strongly agree). Furthermore, self-placement on a political liberalconservative scale is available (conservatism: 'In political matters, people talk of "the left" and "the right" or "liberal" and "conservative". Where would you place your views on this scale?', 10 response categories treated as continuous). This is a standard measure of conservatism, with its wording adapted to the way politics are commonly described in South Africa.
With regard to control variables, one question is whether individuals contacted a politician or local government official in the past 12 months, assuming that this reflects individuals who are politically mobilised so that latent negative attitudes are triggered and crystallised (Claassen 2014). This mechanism has been suggested in the South African context in particular (Claassen 2014;Gordon 2016). Social identity is approached with two questions on nationalism that are combined ('I would rather be a citizen of South Africa than of any other country in the world' and 'Generally speaking, South Africa is a better country than most other countries', combined into a single continuous variable), reflecting social identity theory. There are control variables for education, age (continuous), gender, and race. Education is a central control variable in the literature on attitudes to immigrants, capturing anything from competition in the labour market to the liberalising effect of education or the likelihood to give a socially desirable answer (Facchini, Mayda, and Mendola 2013;. The highest level of education was converted into years of education using typical completion times (Supplement S2). Age, gender, and race are standard sociodemographic control variables.
In the World Value Survey, insufficient income was measured with satisfaction with the financial situation ('How satisfied are you with the financial situation of your household?'). Six per cent of respondents are completely dissatisfied with their financial situation, with most responses around the middle. While the question in the SASAS focuses on feeling comfortable asking a neighbour a favour when needed, the best equivalent in the WVS focuses on feeling part of the community ('I see myself as part of my local community'). Forty five per cent of respondents strongly feel part of the community, with under 2% strongly disagreeing with the statement. There is no variable on contact with immigrants in the WVS, but respondents with immigrant parents can be identifiedwho will have contact with immigrants ('Are your mother and father immigrants to this country or not?'). Four per cent of respondents report having immigrant parents. See Supplement S1 for further details and a description of variables in the extended models.
Although missingness is not a major problem in the present dataset, in line with current best practice (Allison 2001;Lall 2016), multiple imputations were carried out to maximise the use of the information in the data (5 imputations). The same substantive results can be obtained with list-wise deletion (Supplement S3). In the regression analysis, (Gaussian) normal regression models (R Core Team 2008) are used when predicting negative attitudes using the scale in the SASAS data, as well as the preference of South African workers in the WVS dataequivalent to ordinary least squares. The question on undesirable neighbours in the WVS is binary and probit models are used. In all the models, (negative) attitudes to foreigners are the outcome variable, and having insufficient income, help from the community, and contact with foreigners the predictor variables. In each case, I start with perceived relative deprivation and add other predictor and control variables in more complex models to show that the reported findings are robust to alternative explanations.

Findings
Competitive threat theory stipulates that individuals in vulnerable positions exposed to economic competition with immigrants are more opposed to immigrants. Objective measures of absolute income and wealth are not associated with negative attitudes to immigrants in South Africa (income r = −0.01 [95% CI −0.05, 0.04], sum of 'does your household have … ' r = 0.01 [95% CI −0.02, 0.05], cell phone ownership (no substantive differences, Supplement S4). Self-declared social class as a subjective measure of absolute income is not clearly associated with negative attitudes, but suggests more positive attitudes for the 'upper middle class' and 'upper class' (Supplement S4). Measures that more clearly capture (perceptions of) relative positioning are associated with negative attitudes: household income relative to neighbourhood (r = −0.05 [95% CI −0.08, −0.01], Supplement S4) (compare Dambrun et al. 2006), and individuals who perceive their Table 1. Negative attitudes to immigrants, coefficients of regression models.   (Semyonov et al. 2004;Stephan et al. 2005;Sides and Citrin 2007;Weber 2015): Negative attitudes in South Africa are not associated with absolute deprivation, but a sentiment that what one has is not enoughfeeling vulnerable. Consistent across the models in Table 1 is the finding that individuals who think that they can count on the community in times of need have more positive attitudes. These are the consistently negative coefficients on the second line in the table. The statistical interaction between this variable of community help and having insufficient income, however, is not clearly substantive in model M4. Compared to the coefficient, the standard error is large. The sign of the interaction indicates an association between insufficient income and negative attitudes to immigrants only among those with little or no community support. This points to individual-level concerns over competition with immigrants. Both variables in this interaction point to weak or vulnerable positions in society, but not necessarily direct competition with immigrants. This finding is reminiscent of Billiet, Meuleman, and Witte (2014) who found more negative attitudes in economically insecure communities; here the same phenomenon may be picked up at the individual level.
In line with the literature, individuals who have immigrant acquaintances are less likely to express negative attitudes (Model M3). Actual and meaningful contact with immigrants is associated with less prejudice, indicating that negative attitudes probably reflect a fear of the unknown or poorly known rather than calculated responses to a known threat. If the variable on acquaintances in the model shown here is replaced with one that asks about immigrant friends, the predicted reduction of anti-immigrant attitudes is larger and the model fit is increased substantially (Supplement S5). With the variable on immigrant friends, however, it may be argued that individuals have positive attitudes to immigrants first and then become friends with immigrantsquestions of self-selection and causality.
The associations between attitudes and being in a vulnerable position or having contact with immigrants are robust against the inclusion of a range of different variables, as apparent in Figure 1 where the coefficients are shown graphically. The figure presents the point estimates of the regression models as points with thin and thick lines indicating 1 and 2 standard errors. Each model is provided in a different shade, and dots of the same shade and shape belong to the same model. Starting from the left with a model including relative deprivation and personal contact (black, squares), additional variables are introduced. The second model (dark blue, circles) includes two variables to capture personality types and political ideology (see also Table 1). Authoritarian personality is approached with a question whether respondents support the death penalty for murder. Support for the death penalty is associated with negative attitudes to immigrants: Individuals with more authoritarian personality tend to express more negative attitudes to immigrantsvisible by the corresponding coefficients above the dashed zero line in Figure 1. Sociological liberalism is approached with a question on whether gay and lesbian people should be free to live as they want. Individuals who think that such different lifestyles are wrongindividuals who do not exhibit sociological liberalismare more opposed to immigration (Supplement S6).
In the data at hand, there is evidence that mobilisation plays a significant role for negative attitudes to immigration in South Africa. The intuition followed here is that individuals more likely to be participated in politics are more likely to be exposed to political messages (Ruedin 2007), which would make them susceptible to be mobilised against immigration (Claassen 2014). This is the case for any of the variables available in the dataset: contacting a politician (Table 1), contacting a traditional leader, and participation in a protest or demonstration (Supplement S7). The sign of the association is in all cases as predicted with substantial differences in the regression models. Research on media messages in Europe (Klingeren et al. 2015), for example, suggests that this line of research merits further investigationprobably with variables that better capture mobilisation and exposure to incriminating messages. At the same time, mobilisation may play a significant role not so much in shaping attitudes, but when it comes to turning negative attitudes into discriminatory behaviour and violencesomething the data at hand do not cover.
The rightmost models in Table 1 and Figure 1 show that the outlined associations are robust against the inclusion of a range of variables at the individual level: nationalism, the level of education, age, gender, race, and political ideology (liberal-conservative). None of these additional variables influences the associations with relative deprivation, contact, and personality substantively. These variables include alternative explanations for negative attitudes in the case of nationalism (as social identity), and conservative ideology, and standard controls that may affect how the question on attitudes to immigrants is answered in the case of age, gender, race, and educationa variable often highlighted in Western Figure 1. Negative Attitudes to Immigrants, Coefficients of Regression Models. Notes: Outcome variable: negative attitudes to immigrants (scale), data: SASAS 2013. Given are point estimates of the coefficients as dots, with 1 and 2 standard errors as thick and thin lines. Each of the three models is presented in a different shade and shape. The leftmost (black, square) dot and line for insufficient income presents a coefficient of 0.12 with standard error 0.03. The second model from the left (dark blue, circle) includes 'personality', while the rightmost model (light red, diamond) includes additional socio-economic control variables, and control variables for basic 'liberalconservative' position. These are models M3, E1, and E3 in Table 1.
Europe and North America, despite probably capturing different mechanisms (Hello, Scheepers, and Gijsberts 2002). Higher levels of education tend to be associated with being self-monitoring and trying to control prejudices (Janus 2010;Kunstman et al. 2013). Increased age is similarly associated with a reduced tendency to control prejudices due to social and cognitive reasons (Henry, von Hippel, and Baynes 2009;Radvansky, Copeland, and von Hippel 2010).

Robustness with World Value Survey data and education
To ascertain the robustness of the findings, I replicated the analysis with data from the World Value Survey 2013 as far as possible. Supplement S1 outlines the variables used, while Supplements S3 and S6 show that for most explanatory and control variables, the sign of the coefficients is the same; and the same variables tend to be substantively important. In particular, individuals who feel part of the local community are less likely to express anti-immigrant views. As in the models above, individuals who are not satisfied with the financial situation of their household express more negative views about immigrants. To further test the robustness, I also used a question on immigrant neighbours as an alternative outcome variable with the World Value Survey data, even though the concepts are not necessarily equivalent. Probit models predicting answers on not wanting immigrant neighbours are substantively in line with the results reported (Supplements S3, S6).
An important difference to the models using the SASAS data is that education remains a substantive correlate in some models using the World Value Survey. This is likely a reflection of differences in the coding, which leads to a slightly more skewed distribution in the SASAS data compared to the operationalisation with the WVS data. Additional analysis in Supplement S8 shows that this association can also be found in the SASAS data when differentiating university graduates from everyone else. University graduates are clearly more positive to immigrants than individuals with lower levels of education. The other variables in the models are not substantively affected by the substitution of the educational variable, but I have refrained from modifying the models reported above because this exploration was not motivated by theoretical considerations but by a simple scatter plot. Future research is necessary in this regard. With significant differences between university graduates and others, the results for South Africa are in line with findings in Western Europe and North America in that individuals with higher levels of education are less likely to express negative attitudes to immigrants in surveys. At the same time, it should be highlighted that the statistical evidence for this association between education and attitudes is mixed in the limited research on attitudes to foreigners outside Western Europe and North America, perhaps highlighting that education is capturing different causal mechanisms that may be easier to disentangle in countries where the purported liberalising effect of education and labour-market competition can be differentiated empirically rather than just conceptually. Such research may also explore the interaction between levels of education and other variables in the models.

Discussion and conclusion
In this article, I examined the attitudes of South African citizens to immigration. This allowed me to test existing theories developed in Europe and the United States on a fresh case, an important test of validity. In the absence of evidence to the contrary, I assumed that those findings from other places also apply to South Africa because the underlying mechanisms are assumed to be the same. Contrary to some commentators on xenophobia in South Africa (Mattes et al. 1999;Harris 2002;Solomon and Kosaka 2013) there are clear patterns in who tends to be more positive or negative about immigration. Once the concepts were applied to the historical and local specificities of the case, the general findings from Europe and the United States also apply to South Africa.
Individuals in vulnerable economic situations and those lacking community support are more likely to oppose immigration. The subjective feeling that one is vulnerable and finds it hard to make ends meet is associated with negative attitudes. It is not objective competition with immigrants and poverty, but perceived threat: Many objective measures that capture poverty and vulnerability are not substantively associated with negative attitudes to immigrants, but the feeling of being in a vulnerable situation by not having sufficient income is. The importance of perceived threat is reminiscent of studies in Europe and the United States where the influence of the size of the immigrant out-group on attitudes to immigrants is examined (Lahav 2004;Pottie-Sherman and Wilkes 2017).
Individuals who feel they can rely on their neighbours when it matters, more generally, are less likely to express negative attitudes to immigrants. This association can be found across society, not just those in vulnerable economic positions. In line with much of the literature (Semyonov et al. 2004;Ben-Nun Bloom, Arikan, and Lahav 2015), this points to perceptions of vulnerability and social isolation as important factorsnot objective competition with immigrantsalthough the two are unlikely to be completely disconnected.
People with more authoritarian personalities are more likely to oppose immigrants in South Africa, while those showing sociological liberalism (Crouch 1999) give more positive responses in questions on immigrants. With this, there is support for personality types shaping reactions to immigration. These findings are congruent with more universal approaches to the role of personal values in political attitudes (Schwartz et al. 2014), operationalised with more generic variables on authoritarianism, conservatism, and sociological liberalism available in the data used.
Like in Western countries, contact with immigrants is associated with more positive attitudes to immigration in South Africa. Individuals with immigrant acquaintanceslet alone immigrant friendsare much less likely to oppose immigration. It may be that contact with immigrants alleviates perceptions of competition. Put differently, in the absence of contact, attitudes to immigrants may reflect fear of the unknown, reminiscent of studies in Europe and the United States that highlight individuals concerned about threats to their society rather than their own situation (Hainmueller and Hopkins 2014)often by individuals with the least contact (Newman 2014). By contrast, contrary to findings in Europe and the United States, the number of years of formal education was not consistently associated with more positive attitudes in South Africa (see also Whitaker and Giersch 2015, although they focus on country differences). This is in line with evidence from Russia and Asian countries where no systematic effect of education is reported (Alexseev 2015;Barceló 2016;Gordon 2017; see also Weil 1985 who highlighted liberal democracies). Data exploration suggests that in South Africa the distinction between university graduates (more positive to immigration) and other groups in society is relevant.
Further research is necessary to understand the mechanisms that drive an association in Europe and the United States, but not necessarily elsewhere, or perhaps not in the same way. Pecoraro and Ruedin (2016) suggest that individual values and beliefs may be the reason for different associations, pointing to variables interpreted as indicators of authoritarian personality in this article. In some contexts, education may also pick up vulnerable positions in society (Gorodzeisky and Semyonov 2016). Better theory and a more careful operationalisation of relevant concepts may be enlightening.
In the case of South Africa, nationalism is sometimes cited (Peberdy 2001;Gordon 2017), which by implication points to social identity theory. The argument is that as South Africans construct a new inclusive post-apartheid identity, negative attitudes to immigrants as the out-group ensue naturally. While positive identity without a denunciation of an out-group is possible, denunciating members of the out-group is more effective and thus commonplace (Abrams and Hogg 2010). At the individual level, there is evidence that individuals with stronger attachment to South African identity have more negative attitudes to immigrants, but this does not appear to be the dominant mechanism for negative attitudes to immigrants. Similarly, mobilisationoperationalised through political participationdoes appear to affect attitudes. Mobilisation not only seems to play an important role when it comes to the association between negative attitudes and behaviour, such as participation in violence against immigrants (Claassen 2014), but already when it comes to the formulation of negative attitudes. More research in this area is warranted, especially since research from Europe and the United States suggests that messages in the media are able to influence attitudes (Ben-Nun Bloom, Arikan, and Lahav 2015;Klingeren et al. 2015)perhaps by crystallising vague notions of unease and providing narratives to express these attitudes. Such research should also disentangle political mobilisation from possible underlying causes that may affect both mobilisation and attitudes, such as personality types.
In sum, despite claims to the contrary (e.g. Harris 2002; Solomon and Kosaka 2013), there are systematic patterns in who is more likely to favour or oppose immigration in South Africa. Indeed, it seems that the same mechanisms that shape attitudes to immigrants in Western countries are at work in South Africa: Vulnerable economic and social positions, personality, and lack of contact with immigrants are associated with negative attitudes to immigrants (compare Pettigrew 2016 for a review). This suggests that findings from other contexts are robust and generally apply. Of course historical and contextual factors always play a role how these mechanisms are played out and which variables may be best capturing theseparticularly in the absence of standardised questions across surveys as they are sometimes available in Europe and the United Statesbut there is no reason not to draw on the extensive literature on attitudes to immigrants in Western countries. At the same time, cases like South Africa allow to test the generalisability of previous findings, and potentially refine them. This may particularly be the case for testing the relationship between negative attitudes and violence against immigrantsa phenomenon not of lesser importance in Western countries, but more commonly observed in South Africa.