Effect of Media News on Radicalization of Attitudes to Immigration

Recent years have witnessed a growing aversion to immigration worldwide and, at the same time, radicalization of public opinion on the issue. This paper explores the relationship between media news and individual attitudes to immigration. We run an empirical analysis whereby an index capturing individuals’ pro-immigration attitude, measured in 19 countries, is regressed over indexes capturing the coverage and tone of media news about immigration. We find that pro-immigration attitudes are negatively correlated with media coverage and the negative tone of news. However, this correlation is significant only for those with high trust in the media. In the case of low trust, higher coverage of immigration and a negative news slant make previous preferences and beliefs vis-à-vis immigration more extreme, yielding a lower pro-immigration index for those politically on the right, while the opposite applies to those on the left. The pro-immigration index is constructed by means of fuzzy methods to account for the many aspects defining attitudes to immigration.


Introduction
The recent increase in migratory inflows worldwide has dramatically changed host country attitudes to immigration. Two main facts seem to emerge: a growing aversion to immigration, and increasingly polarized public opinion, with politically left-leaning people more willing to accept immigration and right-leaning citizens strongly opposed to it (Semyonov et al., 2006;Rustenbach, 2010;Halla et al., 2017;Stockemer et al., 2020). The mass media have paid great attention to the migration phenomenon, undoubtedly contributing to putting it at the center of the political debate and to forming the opinion of individuals in this regard (Vergeer et al. 2000;2009;Vliegenthart and Boomgaarden, 2007. To appreciate, albeit only anecdotally, the great influence of the mass media in forming public opinion on immigration, suffice it to think that in September 2015 the mere publishing of a photograph of a Syrian child who died on a beach in Turkey was enough to raise the awareness of the tragedy of asylum seekers from Syria and to demand a political response to the refugee problem. 1 To the memory of Mario Carillo who sadly passed away on Saturday 11 September 2021. We would like to thank O. Cassero, the Editor, and four anonymous referees for very helpful comments and suggestions.
In this paper, we explore to what extent and through which channels the mass media may influence native attitudes to immigration. In analyzing the effects of media news coverage, we consider prior beliefs of individuals as moderating factors of such influence, where prior beliefs are captured by individuals' political orientation and their trust in mass media.
A large body of literature has analyzed how media news may affect the attitudes of individuals to immigration. To a lesser extent it has analyzed how media news interacts with individuals' political orientation (Knoll et al., 2011;Merolla et al., 2013;Della Vigna et al., 2014). We add to this literature by showing that the effects of media news on attitudes to immigration depend on the trust in media and on the political orientation of individuals. Our results indicate that extensive coverage of news on immigration raises concerns regarding immigration only when people have great trust in the media. In the case of little trust, exposure to news on immigration triggers a polarization of public opinion, eroding the proimmigration attitude in the case of right-leaning individuals, while enhancing it in the case of left-leaning individuals.
To the best of our knowledge, this is the first paper highlighting the crucial role played by individuals' trust in media in determining whether exposure to immigration news mitigates differences in pro-immigration attitudes of different groups of individuals, by favoring a convergence in opinions, or exacerbates such differences.
In particular, we run a cross-sectional empirical analysis conducted at individual and country level, where the dependent variable is an index measuring individuals' proimmigration attitude, while the main variables of interest are the numbers of negatively slanted news items on immigration and the number of all news items on immigration, irrespective of content. Both are calculated at country level and over different periods of time.
The pro-immigration attitude index is constructed by means of fuzzy sets theory to take into account attitudes such as trust in foreigners, desirability of ethnic diversity, and preferences for immigration policies, which together combine to define the preferences of individuals vis-à-vis immigration. One of the advantages of this index is to obtain a multidimensional and continuous measure of individual attitudes, rather than a measure based on just one aspect, as is usual in the literature. 2 Data on individual attitudes to immigration were obtained from the World Value Survey (WVS) Wave 2005-2009 for 19 countries. 3 From the WVS, we also retrieved several characteristics of respondents, including their trust in the media and their political orientation. Data for news on immigration were obtained from the Bloomberg data set. Other sources were used to collect country level controls.
The influence of media news on public opinion is widely recognized by the political science and communication literature (McCombs and Shaw, 1972;Erbring et al., 1980;Entman, 1993). The media may affect public opinion through the saliency effect (or prime effect), which refers to the fact that when a topic is extensively covered by the media it becomes very prominent in public opinion (McCombs and Shaw, 1972;Erbring et al., 1980;Hatton, 2017), and through the frame effect (or the tone effect), whereby the tone used by media also influences public opinion (Iyengar and Kinder, 1987;Iyengar, 1991;Entman, 1993, Scheufele, 1999Igartua and Cheng, 2009). 4 Finally, it has been stressed that the influence of media news also depends on the timing of the news release (Hastie and Park, 1986;Chong and Druckman, 2010). In this regard, two cases may be distinguished: a process of continuous information learning, whereby the effect of the media is captured by the accumulation of news received in a given period; the presence of a decay process of information over time, according to which only the relatively recent news is relevant to forming public opinion (Chong and Druckman, 2010).
In exploring empirically the effects of media news on individual attitudes to immigration, we take all the above aspects into account, considering both the saliency effect and the tone effect as well as the interval of time in which news are released. Indeed, in order to consider possible heterogeneous effects due to the timing of the news release, we define different indexes of news, two with shorter time Footnote 1 (continued) view in The Independent of September 3, 2015, that "anyone who saw those images during the night could not help but be moved and, as a father, I was deeply moved by the sight of that boy on a beach in Turkey." He added that "Britain is a moral nation and we will fulfill our moral responsibilities." German Chancellor Angela Merkel declared that the policy of "open door" towards refugees, recently announced, appeared morally and politically valid. US President Obama and the Canadian Prime Minister Justin Trudeau immediately announced plans to resettle Syrian refugees and the EU approved a controversial plan to relocate 120,000 migrants across the continent. For a review of the political and social consequences of publishing the photo, see Adler-Nissen et al. (2020). lags, and one that accumulates news over a longer period of time, finding that the news indexes with 1-year and 2-year lag exert greater effects.
As already pointed out, we find that the mass media, by simply raising the saliency of the migration phenomenon, may cause a reduction in positive attitudes to immigration. Furthermore, when the news about immigration is negatively slanted, the reduction is amplified. However, this effect materializes only when people have high trust in the mass media; in the event of low trust, news on immigration causes increased polarization of individuals' preferences according to their political orientations. Indeed, higher exposure to news on immigration is associated to higher aversion only in politically right-leaning individuals, but the opposite effect arises in their left-leaning counterparts. 5 This indicates that the media coverage and its context (frame) interact with prior beliefs of individuals in such a way as to further radicalize prior policy preferences of individuals on immigration. Such results highlight the crucial role of media trustworthiness, and are indirectly corroborated by the findings of a recent literature focusing on the effects of slanted political information provided by social media on political orientation of individuals (Alcott and Gentzkow, 2017;Zhuravskaya et al., 2020;Barrera et al., 2020).
Additional interesting results come from heterogeneity analysis, revealing that highly educated people, in forming their attitudes to immigration, attach less importance to news on immigration, relying more on their political orientation, whereas less educated individuals are more exposed to the "saliency effect" of news on immigration, especially when news has a negative tone. However, also in this case, individuals' political orientation continues to exert its influence.
The paper is structured as follows: after a review of the literature and the main motivations of the paper discussed in the "Literature and Motivations" section, the variables used in the empirical analysis are described in "The Model." "News and Immigration Indexes" introduces the model and descriptive statistics. "Empirical Analysis" discusses the empirical results and contains a heterogeneity analysis, focusing on the moderating role of individuals' level of education and countries' economic development. Finally, "Conclusions" offers some final remarks.
In this literature, the most widely accepted hypothesis is that greater media coverage, by boosting the saliency of the immigration issue, may create alarmism and therefore can also erode pro-immigration attitudes. As regards the frame effect, it is argued that the negative (positive) tone of the news, by creating more aversion (empathy), can enhance (reduce) the negative effect of media coverage (Iyengar and Kinder, 1987;Entman, 1993). Despite the general agreement on such hypotheses, empirical analyses have not yet reached unambiguous results. While some (Hericourt and Spielvogel, 2014;Schemer, 2012;Facchini et al., 2017;Benesch et al., 2019) found that raising the saliency of the immigration issue erodes pro-immigration attitudes, others (Vergeer et al., 2000;Boomgaarden and Vliegenthart, 2007and Meltzer et al., 2020 have reached the opposite conclusion. As part of the literature has emphasized (for a review, see Meltzer et al. (2017)), such inconclusive empirical evidence can be explained by considerable individual heterogeneity in reactions to media news arising from differing economic conditions, culture, beliefs, and political orientations. For instance, Schuller (2016), in analyzing whether the terrorist attacks of 9/11 changed attitudes to immigrants, found that media news affects attitudes on immigration negatively only for individuals with a low level of education, while Schemer (2012) found that negative news on immigration affects only individuals with little issue-specific knowledge. Others focus on the moderating role of group-threat and economic conditions (Vergeer et al., 2000), and of political orientation (Knoll et al., 2011 andMerolla et al., 2013). Our paper contributes to this literature by exploring the moderating role of individual political orientation and trust in media upon the relationship between media news and attitudes to immigration. Even if we fail to solve the endogeneity problem, which naturally arises in this context, and thus offer no causal interpretation for our results, 6 we show that trustworthiness in the media is crucial, as individuals are responsive to media content only when they have high trust in the media; on the contrary, what drives their attitudes to immigration is their political orientations. This is especially true in the case of highly educated people, while for those with low education levels, media news still exerts an influence, especially when such news is negatively framed.
Our results are corroborated by recent findings on a related topic: how people process information they receive, the role of their prior beliefs in this process, and its effects on policy preferences on a particular issue and political orientation (Lenz, 2009;Gleaser and Sustein, 2009;Della Vigna et al., 2014;Alesina et al. 2018;Couttenier et al., 2019;Barrera et al., 2020;Freddi, 2020). Lenz (2009), in analyzing the effects of media news on some salient issues, finds that great exposure to news on these issues does not change the preferences of individuals, but only conveys information on positions of their favorite candidates, which are adopted by individuals as their own. If this is the case, media news should polarize public opinion. Similar conclusions are reached by Della Vigna et al. (2014) who, on analyzing the effects of exposure to nationalistic content of Serbian public radio on the conflict between the Serbian and Croatian groups, find that such exposure triggered the ethnic conflict and polarization of political views. In both cases, the exposure to media news works as a prime for an issue: by making the latter more salient, it activates stereotypes and prejudices. Alesina et al. (2018), focusing on political preferences on immigration and redistribution, find results in line with this conclusion. Through an experiment, the authors obtained in a first step information regarding participants' views about immigration, while in a second step, they asked the participants questions about their preferences regarding a redistribution policy. Of great relevance to the issue in hand, when researchers, in asking participants for their opinion on redistribution policies, made the issue of immigration more salient, respondents with a more negative view on immigration also showed less support for such policies. By contrast, when researchers posed the same questions without giving saliency to immigration, this differentiation in preferences for redistribution policies disappeared. The above findings suggest that prior beliefs of individuals, political orientation, and prejudices do matter for understanding the way in which individuals process information. Similar conclusions are reached by Barrera et al. (2020) who, by exploring the rationality of misleading statements used by politicians, find that misinformation on immigration produces higher support for right-wing parties. This effect remains even when individuals are informed about the true facts. According to the authors, misleading news, by giving high saliency to immigration, induces a sort of "disconfirmation bias" (Edwards and Smith, 1996), for which, when there are strong opposing views about an issue, individuals become more extreme in their beliefs. Finally, Couttenier et al. (2019), in exploring how news on immigrant criminality affected voting at the referendum on a minaret ban in Switzerland, find that the strong negative bias of news about immigrant criminality triggered the support for the minaret ban.
Our paper is also related to the literature on the determinants of individual attitudes to immigration (for a comprehensive survey, see Hainmuller and Hopkins (2014)). A major contribution of this type of literature concerns the distinction between economic and non-economic factors. Among the former is the labor market competition hypothesis, according to which native attitudes depend on whether host country citizens compete with immigrants on the labor market. Within this strand are papers by Mayda (2006), Gang et al. (2013), Facchini and Mayda (2008), and Scheve and Slaughter (2001). All the above papers find that unskilled workers have lower pro-immigration attitudes with respect to skilled workers, since immigration is, for the most part, formed by unskilled workers. Mayda (2006) finds that individual skill is positively correlated with proimmigration attitudes in countries where immigrants are unskilled, whereas the opposite holds in countries where immigrants are more skilled than natives. Reaching similar conclusions, Facchini and Mayda (2008) find that older people and unskilled workers also have a higher aversion to immigration since such categories are in competition with immigrants for welfare services.
Other authors shift the focus onto the role played by non-economic factors, such as culture, beliefs, political orientation, religion, racial orientation, and cultural distance (Quillian, 1995;Gang et al., 2002;Scheepers et al., 2002;Kunovich, 2004;O'Rourke and Sinnott, 2006;Hainmuller andHiscox, 2007 andDustman and Preston, 2007;Dustman et al., 2011;Schuller, 2016;Brunner and Kuhn, 2018). The first important papers suggesting the relevance of non-economic factors were Quillian (1995), Kunovich (2004), and Scheepers et al. (2002), who provide evidence that opposition to immigration is rooted in perceived group-threats and in ethnic prejudices against immigrants considered as out-of-group people. Dustman and Preston (2007) explicitly model the influence of racially driven concerns in forming views about immigration, and establish that racial/cultural prejudice is the main underlying channel through which overall attitudes are driven, largely overcoming welfare and labor market concerns. O' Rourke and Sinnott (2006) include nationalist sentiment in their analysis of the main drivers of attitudes to immigration. Culture also plays a relevant role in explaining individual attitudes to immigration. Hainmuller and Hiscox (2010) find that better educated people also show a higher pro-immigration attitude, as they have a lower level of ethnocentrism, higher preferences for cultural diversity, and more optimistic expectations about the economic impact of immigration (see also Citrin et al., 1997;Chandler and Tsai, 2001).
Our paper contributes to this literature by highlighting the role of cultural factors, such as the reactiveness to media news and further exploring the role of political orientation and trust in media, not widely analyzed hitherto.
Finally, our article is also related to the literature on trust in the media (Kiousis, 2001;Kohring and Matthes, 2007;Lee, 2010;Ardèvol-Abreu and De Zúñiga, 2017). This strand of the literature has shown that the trust of individuals in the media depends on both contextual and individual factors: among the former are factors such as the proliferation of sources, the type of issue dealt with, whether general or specific, and the media bias 7 (or media "slant") (Groseclose and Milyio, 2005). Among the individual factors, an important role is played by emotional and psychological trust in the government and in fellow citizens and, above all, by political ideology and the partisanship of individuals (Lee, 2010). We contribute to this literature by exploring the indirect effects of degree of trust in the media on attitudes to immigration. Furthermore, this literature lends further empirical support to our results, since it shows that trust in the media interacts with political ideology and predicts the use of news by individuals (Ardèvol-Abreu and De Zúñiga, 2017), findings which are both consistent with ours.

The Model
In its most general specification, the model proposed takes the following form: Subscripts i and j refer to individual and country level factors, respectively. In order to obtain consistent results, we use the Huber robust regression approach (Huber, 1973;Maronna et al., 2006;Rousseeuw and Leroy, 1987).
The pro-immigration index is discussed in the "Pro-immigration Index" section. With regard to the right-hand side of the equation, News Index j is our main variable of interest. The usual assumption is that i,j is iid but this is clearly violated in many cases. A natural generalization is to assume "clustered errors," i.e., that observations within group j (within each country) are correlated in some unknown way, inducing correlation in i,j within j, but that groups j and k do not have correlated errors. In the presence of clustered errors, OLS estimates are still unbiased but standard errors may be quite wrong, leading to incorrect inference in a surprisingly high proportion of finite samples. In order to overcome this problem, we implement a pooled OLS regression with standard errors clustered by country.
The coverage effect (or saliency effect), as discussed above, can have negative effects. Extensive coverage can create fear as the phenomenon may be perceived as being out of control. The tone (frame) is a different channel through which media news influences public opinion (Iyengar and Kinder, 1987;Entman, 1993). The argument is that public opinion is greatly influenced by the tone associated to reported news, which can inspire empathy or aversion for the issue. In order to account for such effects, we considered, in turn, both the total quantity of news items and negative news items on immigration. 8 Furthermore, we extend the analysis by considering the moderating role of individuals' prior beliefs, captured by their political orientation, and their trust in the media. The variables Right i,j and Lef t i,j capture respectively right-and left-wing political orientation of individuals, by means of dummy variables, 9 while News Right i,j and News Left i,j are the interaction terms between news exposure and individuals' political orientation which vary both for countries and individuals. Y i and W j are vectors of individual ( Tolerance i , Altruism i , Trust i , Education i , Town Size i , Employment i , Age i , Individual Income i , and Gender i ) and country level ( Unemployment Rate j and GDP Growth j ) control variables, respectively. Low (reference group), medium, and (1) The effects of media bias have been extensively explored elsewhere. Mullainathan and Shleifer (2005) theoretically justify the reasons for the media shaping news in order to meet consumer expectations. Groseclose and Milyio (2005) find empirical evidence that major media in the USA have a liberal bias and propose an objective measure of the slant of news toward a particular political ideology. It has been shown that media bias can have a significant effect on political attitudes (Stromberg, 2004;Gentozkow and Shapiro, 2004, 2006Della Vigna and Kaplan, 2007). high income correspond to < 15,000, 30,000 − 35,000, and > 58,000 Euros per year, respectively; in our case, we use disposable income (after tax) at individual level. Low and high education stand for completed primary school and secondary or higher degree, respectively. Small and big towns refer to < 2000 and > 100,000 inhabitants, respectively. Our analysis considers, in turn, respondents stating they have high trust or no trust in media news.
Finally, we also investigate the effect of the time structure of news. The rationale is that news impacts on public opinion differently depending on the news release lag with respect to the survey release. Political scientists (Hastie and Park, 1986;Chong and Druckman, 2010) suggest that the media effect duration depends on the way people process the information they have been exposed to. The main distinction is between individuals who engage in on-line processing and individuals who use memory-based processing. On-line processors routinely integrate considerations about a specific issue. Hence, the information has longer time effects and an opinion is formed by using all the news received over a certain period of time, whereas memory-based processing individuals use only information they can remember the most and are thus more influenced by recent news. The difference between the two is that on-line processing, using a wider spectrum of information, is likely to be less sensitive to media exposure.

News and Immigration Indexes
In order to study the impact of news on the attitude of natives to immigrants, we present in this section the main variables of our regression models. In particular, in the "Media News Index" section, we present the media news index (our main variable of interest) whereas the pro-immigration index (our dependent variable) is discussed in the "Pro-immigration Index" section.

Media News Index
We use the Bloomberg News platform to retrieve news headlines using machine-learning news analytics based on linguistic pattern recognition. Our underlying assumption is that the more relevant an event is to immigration, the wider the media coverage would be. Hence, the frequency of news reported by the media is used as an indicator of the importance of such events. Bloomberg collects news reported in the language in which it is published and provides a translation in English.
By means of an extensive search string, we first counted all news stories (see Boomgaarden and Vliegenthart, 2009;and Arin et al., 2017) containing the words "immigration," "migrants," and "race," which are immigration-related but have no negative connotation. Then, we counted the number of stories that included the above words together with words implying a negative connotation, namely crime, discrimination, illegal immigration, hate, race, tension, and violence. The former would feed the News All Index whereas the latter feed our News Negative Index.
The main variable of interest, News, is expected to capture two different aspects of the potential news effects: coverage (i.e., the extent and frequency of a particular topic covered by the media with no distinction between the positive, negative, or neutral tone associated to it) and the tone effect (i.e., the effect of the tone, whether positive or negative). The two indexes, labeled respectively as News Index All j and News Index Negative j , are constructed as follows: where j indicates the number of news items per country. The data for the News Index All and News Index Negative are collected from Bloomberg where news coverage is proxied by story headline counts.
The above index has been widely used in the literature (Arin et al., 2017;Caporale et al., 2017 and and possesses some clear advantages. It allows smoothing of the raw number of news items, given by the count of all news and negative news reported by the media on this subject (termed News All j and News Negative j , respectively ). Skewness toward large values would then be smoothed out across our sample. This is particularly important given the large heterogeneous set of countries considered in our empirical analysis. News indexes are constructed for each country.

Pro-immigration Index
The attitude of natives to immigration, like many socioeconomic phenomena, is characterized by many factors (e.g., social and/or political tolerance of immigrants) that should be considered in constructing an indicator of attitudes to immigration. Accordingly, we construct a pro-immigration index by means of fuzzy theory which varies across country and individuals. The index is continuous and takes into account the whole spectrum of preferences, opinions, and beliefs that are expected to form native attitudes to immigrants. Classifying people as pro-or anti-immigration in dichotomous terms is a hyper-simplification of reality that may entail a sizeable loss of information. By contrast, fuzzy sets theory overcomes this issue. Furthermore, it allows both qualitative variables and non-ordinal classes to be considered.
In order to apply fuzzy theory in constructing the proimmigration index, the following four steps have to be taken: (1) the variables that define the pro-immigration (2) News Index All j Negative j = ln e + News All j Negative j attitude have to be chosen; (2) the membership function (MF) must be constructed; (3) the weights associated to each MF have to be calculated; and finally, (4) the MF needs to be aggregated. We identify eight categories aimed at determining attitudes to immigrants. Six questions from the World Values Survey Database are involved in building the index (for more details, see Appendix 1). For each of these (questions) variables, we construct the associated MF. In accordance with Cheli and Lemmi (1995), our MF is a A (x) function that takes values in the interval [0; 1]. In particular, A (x) = 1 identifies full achievement of the target (a resident very inclined to host an immigrant), A (x) = 0 denotes poor achievement (a resident little inclined to host an immigrant), and 0 < A (x) < 1 denotes a situation in between the two extremes. The notion of frequency is considered to define the membership function (for more details, see Lelli (2001)).
Finally, for any individual the six specific MFs calculated (one MF per question) are combined by means of the weighted arithmetic mean (for more details see Cheli and Lemmi 1995;Zani et al. 2010Zani et al. , 2011. Therefore, a weight sensitive to the fuzzy membership is assigned to each variable. Zani et al. (2011) suggest comparing the solutions obtained by different weighting criteria, "in order to gain insight into the stability of the pattern highlighted by the different methods." We compute the fuzzy composite index considering three distinct weighting criteria: • w1-equal weight for each variable; • w2-normalized weights as inverse functions of the fuzzy proportion of each variable; • w3-normalized factor loadings applying principal component analysis (PCA) on the rank correlation matrix (Zani et al. 2010;OECD, 2008). This criterion will be considered only if the first component explains more than 30% of total variability (Zani et al. 2010 and. 10 Table 1 shows the correlation coefficients among fuzzy indicators obtained using different weighting criteria. Pairwise correlations appear to be very high (always greater than 0.86) regardless of the weighting criteria used. Therefore, only results using w3 are reported (weights are reported in Table 12 in Appendix 1).

Control Variables
Identification of a causal impact of media coverage on natives' attitudes is challenging for several reasons. First, attitudes and preferences concerning immigration are due to several unobservable individual characteristics such as altruism, psychological attitudes to new and different people, fear, and the degree of self-confidence. These unobservable components are also correlated with the probability that individuals read particular journals and magazines, and are sources of self-selection. Moreover, there is reverse causality: newspapers report what their readers look for; hence, the attitudes of individuals affect the news reported in the newspapers that individuals read. We do not address all these sources of endogeneity concerns, which is why our results are simple correlations. However, we mitigate the potential bias deriving from self-selection and unobservable covariates, controlling for a number of variables at individual and country level that can be confounding factors.
A first set of control variables, at individual level, captures personal attitudes, beliefs, and ideological positions. In particular, we control for the degree of tolerance, trust in others, and degree of altruism. Moreover, we also account for age and gender (a dummy where male is equal to one) since some authors (Facchini and Mayda, 2008) find that pro-immigration attitudes attenuate with age, while males have a lower pro-immigration attitude. Individual socioeconomic background is measured by working status (which is a dummy variable equal to one if employed, and zero otherwise), and the skills level of the respondent, in order to test the hypothesis that the unemployed and less educated workers have lower pro-immigrant attitudes. Finally, we account for income (those with lower income may feel they are competing with immigrants for welfare state services), respondent geographical location (size of the town where respondents live), and individual country macroeconomic conditions, measured by GDP per capita growth, and unemployment (countries with more solid macro fundamentals are expected to be more open to immigration). The choice of the proper number of principal components takes place on the basis of three criteria which take into account their explanatory power. First we consider a number of principal components which take into account at least 95% of the variance of each of the k initial variables, which imposes a minimal threshold; second, we retain all the principal components whose eigenvalue is larger than 1; third, we observe the screen plot of the eigenvalues as a function of the number of principal components; as eigenvalues are obtained in decreasing order, the graph will show a decreasing curve, with a kink in correspondence to the proper number of principal components. In particular, on the basis of the results of the analysis, we choose only three principal components.

Empirical Analysis
We

Empirical Results
The empirical results are reported in Tables 2 and 3. In Model I, in order to estimate the effect of news, we use the 1-year lag news index 12 (i.e., the index is constructed with news reported on the year before the survey was released); in Model II, we use the 2-year lag news index, constructed with news reported 2 years before; and finally in Model III, we use a cumulative news index measured by news issued from 1 to 6 years before the survey was released. The time lags of the news index can also help reduce endogeneity concerns.
The results obtained show that there is a difference between people with high trust in media news and those with low trust in media news. In the former case the coverage of news on immigration (captured by all news) shows a negative and statistically significant correlation with the pro-immigrant index (see the first three columns of Table 2). The same occurs when we consider the tone of news: the coefficient is negative and significant, even if the negative tone shows a larger coefficient with respect to news coverage (see the first three columns of Table 3). The coefficients of interaction terms, News Right i,j and News Left i,j , , are respectively negative and positive, but not statistically significant. The political orientation variables show that right-wing individuals have a negative propensity towards immigration, whereas their left-wing counterparts have a higher pro-immigration attitude. 13 The negative sign of media coverage is consistent with the hypothesis that a wide coverage may create alarmism, in the sense that the phenomenon is believed to be out of control. Obviously if the tone of the news is negative, this effect is amplified. By contrast, prior beliefs do not seem to modify this correlation.
The picture changes when we analyze individuals who do not trust the media. In this case, media coverage and the tone effect are not statistically significant, even if both coefficients have a negative sign, as in the previous case. By contrast, the interaction terms, News Right i,j and News Left i,j , show statistically significant coefficients, with respectively negative and positive sign. These findings are consistent with the hypothesis that, when individuals do not rely on the media for information, high exposure to immigration news increases the saliency of this issue in the minds of individuals who react by reinforcing their previous beliefs. 14 Finally, the time structure of news reveals that all indexes, 1-and 2-year lag indexes as well as the cumulative index, influence pro-immigration attitudes. However, the main effects occur when 1-year and 2-year lag news is considered, indicating a dominance of memory-based individuals.
Control variables at individual level, such as tolerance, altruism, and trust, are all statistically significant and have a positive sign as expected. Since employment also has a positive sign, labor market competition can reduce the propensity for immigration, a finding which is in line with the literature (Mayda, 2006;O'Rourke and Sinnot, 2006). Age shows a negative sign; this implies that older people have a lower pro-immigration index, which also agrees with findings reported elsewhere (Mayda, 2006). Variables capturing income and education show that people with higher incomes (≥ Euros 58,000) have a higher pro-immigration attitude compared to those with low (Euros 15,000) and medium (Euros 30,000-35,000) incomes. A similar pattern emerges with respect to a higher level of education (secondary school or degree) which has a greater effect on pro-immigration attitudes compared to low education (primary school). These results confirm those obtained by Mayda (2006) and Hainmueller and Hiscox (2007). Finally, we found that individuals living in large towns (population ≥ 100,000) have a higher pro-immigration attitude. This result confirms the greater "openness" of those living in more urbanized areas already found elsewhere. Overall, our results confirm the important role of some individual variables, although it is not always clear whether this Table 2 Regression results-pro-immigration index and all news *** , **, and * denote rejection of the null hypothesis at 1%, 5%, and 10% levels respectively. Standard errors (reported in brackets) were clustered by country. The News index is equal to ln[e + all news story counts]. Models I and II use the news index based on the total number of news items released 1 year and 2 years before the survey was released, respectively. Model III refers to the total number of news items issued in the 5-year interval before the survey was released. Respondents are grouped according to whether they stated they had high trust and no trust in media news. Income low, medium, and high are equal to < 15,000, 30,000-35,000, and > 58,000 Euros per year, respectively. Education low and high stand for completed primary school and secondary or higher degree, respectively. Small and big towns refer to < 2000 and > 100,000 inhabitants, respectively depends on cultural or economic motives given the strong interrelationship between the two. Control variables at country level, unemployment, and GDP per capita are statistically significant, with the estimated coefficients indicating respectively a negative (unemployment) and positive (GDP per capita) effect, as one would expect. The effect of unemployment confirms the importance of labor market channels: in the case of high unemployment, people do not prefer high immigration levels because of market competition, while a high level of development makes countries more open and favorable to immigration given not only the greater economic opportunities but also the different effects that the level of development has on the skills composition of workers. However, the level of GDP per capita growth can also capture other characteristics of countries not Table 3 Regression results-pro-immigration index and negative news *** , **, and * denote rejection of the null hypothesis at 1%, 5%, and 10% levels, respectively. Standard errors (reported in brackets) were clustered by country. The News index is equal to ln[e + negative news story counts]. Models I and II use the news index based on the total number of negative news items released 1 year and 2 years before the survey was released, respectively. Model III refers to the total number of negative news items issued in the 5-year interval before the survey was released. Respondents are grouped according to whether they declared having high trust and no trust in media news. Income low, medium, and high are equal to < 15,000, 30,000-35,000, and > 58,000 Euros per year, respectively. Education low and high stand for completed primary school and secondary or higher degree, respectively. Small and big towns refer to < 2000 and > 100,000 habitants, respectively completely limited to labor market factors, such as the level of social capital and a lower level of criminality. However, we were unable to disentangle the different types of motivations.

Heterogeneity and Robustness
In this section, we analyze whether there are differences in the reactions to news on immigration driven by some characteristics at country and individual level. In particular, we consider countries' economic development and individuals'  level of education as possible sources of heterogeneous reactions to news on immigration. Some authors (Facchini et al., 2008;Chang and Kang, 2018) find that the native population in more developed countries shows a higher pro-immigrant attitude, triggered by better economic conditions and higher trust in institutions (Chang and Kang, 2018). If this is true, it can be argued that people living in developed countries show a less adverse reaction to news on immigration. In order to test such a hypothesis, we run our estimations for two separate groups, one made of high income (developed) countries and another by low income (developing) countries. We use the   Table 8 Regression results-pro-immigration index and all news for people with high education *** , **, and * denote rejection of the null hypothesis at 1%, 5%, and 10% levels, respectively. Standard errors (reported in brackets) were clustered by country. Respondents with a secondary or higher qualification are classified as people with high education. Individual and country level controls are included in the estimation. Please refer to Table 2 Tables 4, 5, 6, and 7, do not corroborate this hypothesis as findings for both groups do not appear to differ substantially from one another and are in line with those obtained for the whole sample. The level of education may affect both attitudes to immigration (Scheve and Slaughter, 2001;Mayda, 2006;Hiscox, 2007 andSchuller, 2016) and the way in which individuals process information from the media. Therefore, we also investigate whether the level Table 9 Regression resultspro-immigration index and all news for people with low education *** , **, and * denote rejection of the null hypothesis at 1%, 5%, and 10% levels, respectively. Standard errors (reported in brackets) were clustered by country. Respondents with a primary education degree are classified as people with low education. Individual and country level controls are included in the estimation. Please refer to Table 2 for the set of control variables with the exclusion of the education variable High trust in media news No trust in media news Model  I  II  III  I  II  III All news News − 0.0221*** − 0.0879*** − 0.0312*** − 0.0214*** − 0.0623*** − 0.0655*** (0.  Table 10 Regression resultspro-immigration index and negative news for people with high education *** , **, and * denote rejection of the null hypothesis at 1%, 5%, and 10% levels, respectively. Standard errors (reported in brackets) were clustered by country. Respondents with a secondary or higher qualification are classified as people with high education. Individual and country level controls are included in the estimation. Please refer to Table 2  of education moderates the reaction to news on immigration. To this end, we split the sample into two sub-samples according to individuals' level of education (low or high). 16 The results, reported in Tables 8, 9, 10, and 11, are consistent with our previous findings. However, it appears that in the case of people educated to a low level, even for those with no trust in the media, the coefficients associated to All News, as well as Negative News, variables are always negative and statistically significant even for those with no trust in the media, while in the case of the highly educated, those coefficients are still negative (with a smaller magnitude in absolute value) and not statistically significant. We interpret these results as evidence that education does affect reactions to news on immigration, as the highly educated, in forming their attitudes to immigration, rely more on their prior beliefs. In this respect, our results are consistent with findings of Schuller (2016) who, in analyzing the effects on pro-immigration attitude of news related to the September 11, 2001, terrorist attacks, shows that educated people are less influenced by negative news.
Finally, to check the robustness of our whole sample results, we also estimate the model with countries' fixed effects (see Tables 15, 16, 17, 18 in Appendix 2). The fixed effect specification does not allow us to include unemployment, GDP per capita growth, or the News indicator as the above variables are country-specific. The results, however, as far as the interaction terms are concerned, are in line with Table 11 Regression resultspro-immigration index and negative news for people with low education *** , **, and * denote rejection of the null hypothesis at 1%, 5%, and 10% levels, respectively. Standard errors (reported in brackets) were clustered by country. Respondents with a primary degree are classified as people with low education. Individual and country level controls are included in the estimation. Please refer to Table 2

Conclusions
This paper analyzed the effects of media coverage on individuals' pro-immigration attitudes. The novel pro-immigration attitude index proposed was calculated by means of a fuzzy approach. The results can be summarized as follows: both the coverage and negative tone of news attenuate proimmigration attitudes, albeit only for those with high trust in the media. By contrast, for individuals with no trust in the media, news on immigration strengthens the effects of political orientations, further eroding the pro-immigration attitude in the case of individuals with negative prior preferences (right-leaning), while increasing the pro-immigration attitude in the case of individuals with positive prior preferences (left-leaning). Therefore, in the latter case, media news radicalizes individuals' prior preferences on immigration, captured by their political orientation.  Table 13 Descriptive statistics The pro-immigration index is calculated according to fuzzy set theory described in the "Pro-immigration Index" section. Right and Left measure political orientation. The moderate political orientation is omitted because it represents the reference class. For this reason, the sum of individuals who claim to vote for left-wing and right-wing parties is not equal to the number of observations related to the immigration index. C.V. stands for the coefficient variation (or relative standard deviation). This is reported instead of the standard second moment as mean values are substantially different across countries Weights associated to each member function are reported in Table 12 Data Descriptive Statistics A first visual summary of the pro-immigrant index distributions due to political orientation is provided in Fig. 1. It can be seen from this figure that the distributions are quite similar for people who vote right and moderate. In particular, we note that the modal value of the indicator is about 0.4. By contrast, the distribution of the indicator for people who vote for left-wing parties shifts to the right, showing greater propensity toward immigration. Moreover, the distribution is bimodal, with the higher mode characterized by lower propensity toward immigration (with a value slightly higher than 0.4); hence, the left-leaning people show a higher heterogeneity Furthermore, we report the pro-immigration index distributions conditional upon the degree of trust in news (high trust and no trust). The unconditional distribution is also shown as the benchmark distribution. Figure 2 shows the distributions in this case.
We observe that the distribution of people with high confidence in news is more concentrated around the median value, with tails lower than those of the distribution of people with low confidence. We interpret this behavior of the distribution as evidence of a process of convergence in opinion in case of high trust in media. The distribution of people with no trust in news is very similar to the benchmark distribution except for the extreme values. With respect to the distribution of people with high trust in news, it is less concentrated around the median value and has the right tail fatter. We interpret this behavior of the distribution as evidence of a higher heterogeneity in opinion which is more polarized. Reassuringly, in case of high trust in media, it seems that news convey towards a convergence of opinions, by reducing the effects of prior beliefs. Tables 13 and 14 present a set of descriptive statistics for the variables of interest. The pro-immigration index average is particularly high in the cases of Finland, Norway, Sweden, and Switzerland, albeit experiencing higher volatility. Georgia has the lowest (0.317), whereas the remaining 14 countries considered range between 0.408 (Poland) and 0.499 (USA). On average, individual respondents who stated they had a political orientation to the right or left were almost equally split, with ten countries out of 19 showing a higher percentage associated to a right-wing political orientation. Descriptive statistics on news are also very informative. Countries with the most exposure to news on immigration are Germany, Italy, Spain, and the USA, whereas eastern European countries appear to be the least exposed. A similar pattern is observed when we look at countries highly exposed to negative news, whereas Bulgaria, Cyprus, Finland, Moldova, and Serbia appear to be the least exposed to negative news. Appendix 2 Table 15  Table 16  Table 17  Table 18  Table 16 Fixed effects regression results-pro-immigration index and all news for developing countries *** , **, and * denote rejection of the null hypothesis at 1%, 5%, and 10% levels, respectively. Standard errors (reported in brackets) were clustered by country. Low income (  Data Availability Data are available on request from the authors.

Conflict of Interest
The authors declare no competing interests.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.