Welfare states, media ownership and attitudes towards redistribution

ABSTRACT A sizable literature has examined the relationship between welfare state regimes and attitudes towards redistribution, generally expecting that citizens in more generous welfare states are more supportive of redistribution. However, empirical findings are inherently mixed. I argue that the role of the media has been neglected in this research. By providing cues to individuals, the media plays a strong role in shaping individual-level attitudes. These cues can be expected to vary according to the characteristics of national media systems. I test these claims with multilevel regression analysis of 18 European countries. The results show that redistribution support is lower where media ownership is more concentrated. Accounting for media ownership concentration helps to explain variation in attitudes towards redistribution, besides established determinants related to welfare state regimes. I outline implications for future research relating to the characteristics and regulation of media systems, news coverage, and the politics of inequality and redistribution.


Introduction
Based on the work by Esping-Andersen (1990), a large body of research has examined the associations between welfare state regimes and preferences and attitudes towards redistribution, expecting individual-level support for redistribution to matter for the politics of welfare state reform. Scholars generally expected support to be higher in the more universal and generous welfare states (Korpi & Palme, 1998). However, empirical evidence is inherently mixed. Various scholars have found a patterning of preferences which does not conform to these expectations (e.g., Jaeger, 2009;Schmidt-Catran, 2016). In addition, a substantial amount of cross-national variation in redistribution preferences remains unexplained.
The role of the media, and more specifically, the media ownership structure, constitutes an important neglected factor in this research. The more concentrated media ownership is, the more likely it is that news coverage will be biased for political or profit-related reasons (Anderson & McLaren, 2012). As a consequence, news coverage related to inequality and redistribution might receive relatively little attention or might be portrayed in a disadvantageous light, leaving its imprint on individual-level attitudes. The claim that the mass media may affect public opinion is not new. A large, and largely experimental literature on framing effects has shown that news framing in the media has an impact on individual attitudes (Hopmann et al., 2017;Slothuus, 2007). Yet, this literature to date has hardly addressed to what extent media frames may actually vary across countries, and to what extent variation in the characteristics of the mass media may systematically, and beyond experimental settings, affect individual-level attitudes.
I evaluate the relevance of media ownership concentration for attitudes towards redistribution applying multilevel regression analysis of data from the European Social Survey (ESS, 2016) and the Media Pluralism Monitor (MPM, 2016). The results show that more concentrated media ownership structures are tightly associated with lower support for redistribution. This finding holds when taking into account the role of welfare state regimes and income inequality. My findings imply that media ownership structures deserve more attention for the study of issues related to regulation of media systems, news coverage of social policy and social inequality, attitudes towards redistribution, democratic accountability, and the politics of inequality and redistribution.
The following sections discuss existing research on the associations between welfare states and redistribution preferences, media framing, and the role of media ownership structures to elaborate the claim that we should take into account the impact of media ownership concentration when studying individual-level attitudes towards redistribution. The empirical results provide support for this line of reasoning. I conclude by outlining the implications of my findings and by providing an outlook for future research.

Welfare states, media ownership and attitudes towards redistribution
Following Esping-Andersen (1990) claim in his Three Worlds of Welfare Capitalism that welfare state regimes shape the politics of further welfare state reform, many scholars have examined how welfare state regimes feed back into attitudes towards the welfare state and redistribution (for an overview, see Kumlin and Stadelmann-Steffen (2014)). Most scholars started from the assumption that support for redistribution should be highest in the most generous social-democratic welfare states, lowest in the market-oriented liberal welfare states, and in between in the conservative welfare states. The more universal and the more generous welfare states are, the more they should bolster support for continued redistributive social policies by making people adjust to established welfare provision (affecting self-interest) and by shaping the normative evaluations of welfare states (Korpi & Palme, 1998;Kumlin & Stadelmann-Steffen, 2014).
Empirical findings are inconsistent. While some studies have found support for these expectations (Linos & West, 2003;Svallfors, 1997), others have found substantial mismatches in the associations between welfare state regimes and redistribution preferences (Brady & Bostic, 2015;Gelissen, 2000;Jaeger, 2009;Schmidt-Catran, 2016). For example, based on a range of indicators of welfare state characteristics, Jaeger (2009) finds support for redistribution to be higher in the conservative than in the social-democratic welfare states. Much of the earlier research relied on analyzing a small number of countries as representatives of the different welfare state regimes, or including welfare state ideal types as explanatory variables, limiting the validity and generalizability and contributing to some extent to the inconsistencies of the findings (Jaeger, 2006).
More recently, scholars have begun to examine feedback effects of welfare states on individual attitudes for specific indicators of welfare state characteristics (Brady & Bostic, 2015;Jaeger, 2006Jaeger, , 2009, and they have taken into account more seriously contextual factors potentially shaping the demand for redistribution (Dallinger, 2010;Dimick et al., 2018;Finseraas, 2008;Jaeger, 2013). For instance, Jaeger (2006) finds levels of public social spending to be positively associated with public support for redistribution whereas this association does not apply to various other indicators of welfare state generosity. Among the demand-side factors, income inequality has turned out to have a strong positive impact on redistribution support in European welfare states, with income inequality generally being higher in less generous welfare states (e.g., Dallinger, 2010;Jaeger, 2013). 1 Despite these advancements, we still face difficulties in making sense of the large extent of cross-national variation in welfare state attitudes which cannot be attributed to welfare state regimes as should be expected. I argue that cross-national variation in the characteristics of media systems plays an important role in affecting attitudes, independent of the impact of welfare state regimes. While welfare states should leave their imprint on individual-level attitudes because individuals experience the functions of the welfare state, e.g., when becoming old, sick, or unemployed (Soss & Schram, 2007), direct contact with the welfare state is not the only factor shaping attitudes. The way the media provides information can be considered as an important factor shaping support for redistribution and the welfare state (Larsen, 2013).
Numerous, largely experimental studies have shown that media coverage of social policy and redistributive issues leaves its imprint on individual-level attitudes (e.g., Barnes & Hicks, 2018;Hopmann et al., 2017;Kneafsey & Regan, 2020;Slothuus, 2007). When welfare state recipients are portrayed as undeserving, support for welfare state retrenchment increases (Hopmann et al., 2017;Slothuus, 2007). Barnes and Hicks (2018) similarly show that how austerity was framed in the British media after the financial crisis had an impact on individual-level support for deficit reduction. However, while the strength of many of those experimental studies is their high internal validity, with regard to external validity we hardly know what drives actual variation in media coverage and how relevant it is in explaining variation in redistributive attitudes across countries. 2 Real-world media coverage of social inequality and redistribution tends to deviate from what one could expect based on socio-economic developments or the material interest of ordinary citizens. The media may pay disproportional attention to economic news which are relevant for more affluent individuals to the detriment of issues relevant to those with low incomes (Jacobs et al., 2021). If the media reports about inequality, it sometimes frames individual economic hardship as a matter of personal responsibility, which tunes down the responsibility of the government in combating inequality (Epp & Jennings, 2021). In a detailed analysis of news coverage of social inequality in four European countries, Grisold and Preston (2020) find media coverage of issues related to social inequality to be highly incomplete. Media reporting on social inequality tends to downplay existing power relations and conflicts of interest and frequently stays silent with respect to concrete political measures that could be implemented to counteract rising levels of social inequality. Despite these findings, however, empirical work trying to map and explain variation in media coverage on inequality and redistribution across countries is highly limited (cf. the discussion in Grisold & Preston, 2020).
That said, various studies have documented that media systems vary across time and countries along various characteristics (Brüggemann et al., 2014;Hallin & Mancini, 2004). Media systems vary along aspects such as the development of media markets, journalists' engagement in political advocacy along partisan lines (political parallelism), the degree of journalistic professionalism, or with respect to the role of the state, and this variation is expected to coincide with variation in news reporting (Hallin & Mancini, 2004). Building on the notion of political parallelism, various studies have examined partisan media bias at the level of media outlets and countries (Castro-Herrero et al., 2016;see Puglisi and Snyder (2015) and Lelkes (2020) for recent overviews). This research has generally shown that news outlets report more positively about parties that share a political orientation similar to the news outlet, and vice versa for parties with different political orientations. However, limitations of the existing studies on media bias are that most of them are not comparative and focus only on single countries (Lelkes, 2020;Puglisi & Snyder, 2015). As a consequence, the linkages between media system characteristics and actual media coverage have hardly been assessed. Finally, the focus on studying the alignment between parties and media outlets neglects that media bias can also reflect the interests of media owners, which can lead to an additional, independent source of bias in the coverage of issues related to social inequality and redistribution.
To some extent, variation in media systems coincides with countries' clustering into different welfare state regimes. Causation leading to this overlap between media systems and welfare states may run in both directions (Enli et al., 2018). Media systems should affect welfare state development, e.g., by adopting or not a positive view of politics and egalitarian ideas. Vice versa, the intervention of governments in media systems can be considered as one element of welfare state interventionism (Ahva et al., 2017;Enli et al., 2018), implying that the evolution of welfare states should leave its imprint in media landscapes. In line with this perspective, Larsen (2013) shows how varying media frames of poverty in the US, the UK, Sweden and Denmark are associated with different public perceptions of the poor in those countries. Despite such initial findings, there is nevertheless a substantial degree of unexplained variation in the characteristics of media systems within different country clusters (Brüggemann et al., 2014(Brüggemann et al., , p. 1057. Thus, irrespective of the role of welfare state institutions, we do not know to what extent variation in the characteristics of media systems within and across welfare state regimes might affect individual-level attitudes towards inequality and redistribution. A growing literature in economics has theorized about how media characteristics should affect news coverage and politics (Anderson & McLaren, 2012;Besley & Prat, 2006;Di Gioacchino & Verashchagina, 2020;Kennedy & Prat, 2019). A central claim of this literature, which rests to a large extent on formal economic modeling (e.g., Anderson & McLaren, 2012;Besley & Prat, 2006), is that relevant actors in business and politics have incentives to exert control over news coverage because this allows them to influence public opinion. Di Gioacchino and Verashchagina (2020) is, to my knowledge, the first study examining empirically the association between national media characteristics and individual-level attitudes. They show that where the risk to market plurality in the media landscape is higher, individuals tend to devote less importance to the equality of opportunities.
In this respect, patterns of media ownership can be considered as a crucial aspect of the characteristics of media systems. Increasingly concentrated ownership structures, which are partly a result of mergers and acquisitions, have been one of the most dramatic changes in media landscapes over the past decades (Anderson & McLaren, 2012;Kennedy & Prat, 2019;Noam, 2016). According to numbers from a recent comprehensive data collection of ownership concentration in 30 countries around the globe, on average, the top four media companies in each country had a market share of 65 per cent in 2012 (excluding online media and platforms), reflecting an oligopolistic market structure (Noam, 2016(Noam, , pp. 1306(Noam, -1308. Between 2004 and 2012, the market shares of these top four companies increased by an annual 1.35 per cent (Noam, 2016(Noam, , p. 1309) and this increase has been particularly pronounced in those countries previously characterized by relatively low levels of concentration (Noam, 2016(Noam, , p. 1317. High and increasing media ownership concentration possibly poses one of the greatest risks to pluralism in the mass media because it is likely to strengthen media bias (Besley & Prat, 2006;Di Gioacchino & Verashchagina, 2020;Hughes & Prado, 2011;MPM, 2016;Noam, 2016). This may be for political and profit-related reasons.
Ownership concentration may lead to biased information out of political reasons because highly affluent owners may try to promote their own political agenda (Anderson & McLaren, 2012;Corneo, 2006;Gilens & Hertzman, 2000;Hughes & Prado, 2011). Concentrated ownership should be detrimental to individual support for redistribution because owners, on the one hand, may try to influence news reporting in favor of their own material interests, and, on the other, may try to promote certain worldviews in news reporting which are legitimizing the absence of strong government intervention in the income distribution. For the case of Latin American countries, Hughes and Prado (2011) show how highly concentrated ownership patterns narrow down media coverage of issues related to social inequality and social policy, instead priming crime and public security, often with a sensationalist framing. 3 While the influence of ownership on news reporting can be expected to be particularly high in the Latin American countries (Hughes & Prado, 2011), similar arguments have been made for the more egalitarian Western countries. Larsen (2013) has illustrated the importance of how the news media deals with crime in relation to poverty in four highly developed Western countries. Gilens and Hertzman (2000) have shown for the US that media owners indeed seem to have been successful in influencing news reporting in line with their economic interests.
High ownership concentration is not to say that powerful owners will always try to shape news reporting related to inequality and redistribution. Owners may well have progressive world views which might be reflected in news coverage, or they may value the independence of their editorial offices. However, with increasing ownership concentration, the incentives and opportunities for owners to influence news reporting increase. If ownership is concentrated in the hands of a few wealthy owners, the difference in the economic interests between owners and average citizens becomes highly pronounced (Corneo, 2006). Under concentrated ownership, low market shares of more independent media outlets will be insufficient to counteract bias in news coverage. In such situations, market-dominating media outlets can afford biased coverage because the risk of losses in profits and market shares due to their audiences potentially defecting to competitors is relatively low (Puglisi & Snyder, 2015). 4 Biased news coverage may also arise out of more narrow profit-related reasons. Owners may try to maximize sales by appealing in their news reporting to their key audiences, which are often not representative of the larger population (Anderson & McLaren, 2012). From an organizational perspective, integrated news corporations have incentives to materialize economies of scale by relying on contracts with specific news agencies and by exchanging material between their editorial offices. Beckers et al. (2019) and Vogler et al. (2020) provide supportive evidence of news content diversity having decreased over time in Belgium, and Switzerland, respectively, two countries characterized by trends of rising ownership concentration. In addition, with one single management decision potentially affecting news reporting of various outlets of a news corporation, a highly concentrated ownership structure contributes to a decreasing diversity of news coverage (Beckers et al., 2019). With fewer alternative providers being available in countries with highly concentrated ownership structures, the risk increases that informational bias of an individual media outlet translates to biased information at the aggregate level. Thus, increasingly concentrated ownership structures are likely to amplify already existing tendencies towards only limited news reporting on inequality and redistribution. As the experimental literature on media framing effects suggests (e.g., Hopmann et al., 2017), this biased media coverage is likely to have adverse consequences for public support for redistribution.
Hypothesis 1: Where media ownership is more concentrated, support for redistribution is lower.

Data and methods
To evaluate the claims elaborated above, I examine the empirical association between media ownership concentration and attitudes towards redistribution. The analysis is based on multilevel random intercept regression models of a sample of 31,354 individuals nested in 18 European countries, for which the relevant data is available. 5 Individual-level data is taken from the ESS (2016). I use the standard measure of attitudes towards redistribution which is used by the majority of the studies discussed above, i.e., agreement or disagreement to the statement: The government should take measures to reduce differences in income levels Responses are measured on a five-point likert-scale and range from 'agree strongly' to 'disagree strongly'. I recoded the variable so that higher values indicate stronger support for redistribution. I include individual-level control variables similar to several other studies on the determinants of attitudes towards redistribution (e.g., Jaeger, 2006). I control for age, gender, employment status, and social class based on the social class scheme developed by Oesch (2006). As a robustness check, I include in additional model specifications individual levels of news consumption and their interaction with media ownership concentration. 6 For data on the characteristics of media systems, I use the Media Pluralism Monitor (MPM) (2016). The MPM assesses the risk to media pluralism across European countries based on the assessments of national experts. Expert responses are coded into 200 variables which are aggregated into a set of twenty indicators measuring four dimensions of risks to media pluralism (basic protection, market plurality, political independence, and social inclusiveness; MPM, 2016, p. 6). The analysis in this contribution relies on the measure of horizontal concentration of media ownership as part of the broader dimension of risk to market plurality. This measure indicates how much control one or a few media owners may exercise in a specific sector of the media system, including audiovisual media, radio, newspapers, and online media. It takes into account national regulation designed to prevent ownership concentration, and how this regulation is implemented (MPM, 2016, p. 27). 7 Risk to media pluralism arising from horizontal ownership concentration is assessed on a scale from 0 to 100 where scores below 33 are considered low risk, scores from 34 to 66 are medium risk, and scores above 67 are high risk.
This measure of horizontal media ownership concentration is particularly suited for this study. As discussed above, whether ownership is concentrated or dispersed should have a strong influence on the extent of biased information and on the availability of competing standpoints regarding inequality and redistribution. 8 On average, the measure of horizontal ownership concentration has received the highest score of risk to media pluralism of all indicators collected in the MPM (MPM, 2016, p. 3). The average risk to media pluralism arising from horizontal ownership concentration is 73 for the countries included in this study with a standard deviation of 16. While a high risk score does not imply that ownership concentration necessarily presents a threat to media pluralism, the very high levels of risk assessed in the MPM suggest that ownership concentration can be plausibly expected to actually have an impact on news reporting, even in the advanced European countries.
The indicator on horizontal ownership concentration focuses on ownership patterns of private market participants. Politically independent and robustly funded public broadcasters may present a counterweight to highly concentrated private media markets (Brüggemann et al., 2014;MPM, 2016, p. 29). The existence of independent and well-funded public broadcasters and of generous public press subsidies has been identified as a characteristic feature of media markets in particular in the Nordic countries (Brüggemann et al., 2014). To examine whether such types of media policy may counteract the influence of ownership concentration, I use another item from the MPM measuring the risk to political independence of public service media governance and funding. While this indicator only captures some aspects of public media policy intervening in media markets, it is nevertheless helpful as a robustness check to assess whether the effects of concentrated ownership on redistribution support hold when accounting for the role of independent public broadcasters.
As an additional consistency check, I include an indicator on commercial and owner influence over editorial content from the MPM and interact it with ownership concentration. This additional indicator assesses through various items the degree of journalists' independence from owners and from the editorial line of their news outlets, and the amount of influence of commercial interests on media content (see Table A.2 in the online appendix for further detail). The rationale for this additional step of the analysis is that ownership concentration should be particularly consequential for news coverage and individual-level attitudes, if owners are relatively well positioned to actually influence news coverage.
I use various measures to capture the role of welfare states and socio-economic context as further independent variables. As a proxy measure of welfare state generosity, I use total public social expenditure (OECD, 2020). Although this measure is controversial -'it is difficult to imagine that anyone struggled for spending per se' (Esping-Andersen, 1990, p. 21; emphasis in original)it has been found to be positively associated with support for redistribution (Jaeger, 2006). To be able to compare my results to previous research relying on welfare state ideal types, I include in additional models dummy variables for the different welfare state regimes. I use the regime-clustering by Ferrera (1996) because it covers a larger number of countries than Esping-Andersen's (1990) original analysis and it allows capturing potential differences between Continental and Southern European welfare states. To widen the scope of the analysis and to increase sample size, I also include the Eastern European countries as an additional category (Dallinger, 2010). 9 I run models with and without including the indicators of welfare state characteristics and media ownership concentration. This allows me to examine how the effect estimates of the welfare state variables change when adding ownership concentration, and whether the effects of ownership concentration hold when accounting for the potential associations between the characteristics of welfare states and media systems (Ahva et al., 2017;Enli et al., 2018;Larsen, 2013).
To control for the possibility that demand-driven support for redistribution biases the results, I include the Gini index of inequality in household market income (Solt, 2020). In additional models, I take into account additional potential confounders related to the country-specific socio-economic context. I control for inflation, GDP growth and unemployment rates. I also control for population size because economies of scale might contribute to higher ownership concentration in smaller countries. The operationalization of the variables is described in more detail in Table A.2 in the online appendix.
A limitation of the analysis is that it is unable to test the full postulated causal chain, expecting ownership concentration to increase media bias which should in turn reduce individual-level support for redistribution. As discussed above, evidence exists for several steps of this causal chain with studies examining ownership concentration patterns (e.g., Noam, 2016), media bias (e.g., Lelkes, 2020;Puglisi & Snyder, 2015), or media framing effects in experimental settings (e.g., Hopmann et al., 2017). Largely due to the difficulties of collecting media content across different countries and languages, these different research fields to date have hardly been linked to each other (Puglisi & Snyder, 2015;Van Aelst et al., 2017). While this study is unable to fully address this research gap, it advances the existing literature by for the first time empirically linking media ownership concentration and individual redistribution support as two central elements of the causal chain outlined above. Table 1 shows the results from the multilevel random intercept models. Model 1 includes ownership concentration as the only macro-level variable, Model 2 adds income inequality, Models 3 to 8 add welfare state indicators (public social spending in Models 3 and 4, and welfare state regime dummies in Model 5 to 8). To be able to compare the influence of media ownership concentration and welfare states, each of these model specifications is estimated with (Models 3, 5, and 7) and without (Models 4, 6, and 8) the variable of ownership concentration. Individual-level controls are included in all models but not shown to facilitate the presentation of the results. The full results are shown in Table A.3 in the online appendix.

Results
As Model 1 shows, a higher level of media ownership concentration is negatively associated with support for redistribution, but, when included as the only macro-level variable, this association is statistically insignificant. Yet, as bivariate correlations show, ownership concentration is positively correlated with income inequality (r=0.28, p=0.25), which in turn tends to lead to stronger support for redistribution (Dimick et al., 2018;Finseraas, 2008). When income inequality is added in Model 2, the effect of ownership concentration increases in size and becomes statistically significant. As expected, higher  Table A.3 in the online appendix. Notes: *p < .10, **p < .05, ***p < .01. levels of inequality are associated with stronger support for redistribution. Both effect estimates are substantial, and roughly similar in size. Calculating average marginal effects based on the regression coefficients shows that moving from the lowest to the highest level of ownership concentration reduces support for redistribution by 0.41 scale points (the change is 0.44 scale points when moving from the lowest to the highest level of inequality). The magnitude of these effects amounts to 1.41 and 1.52, respectively, standard deviations in redistribution support (cf. Table A.1 in the online appendix). Thus, the more concentrated media ownership is, the less supportive are voters of redistributing income.
Model 3 adds total public social spending to the model. The effect estimates of ownership concentration and income inequality remain basically unchanged. Contrary to earlier findings by Jaeger (2006), social spending is not statistically significant, and it is even negatively signed. When ownership concentration is dropped from the model (Model 4), the sign of public social spending turns positive, but the effect remains insignificant. The comparison of Models 3 and 4 furthermore demonstrates that including media ownership concentration as a determinant of attitudes towards redistribution improves substantially the explanatory power of the model. The share of the variance at the random intercept level which is explained by the variables included in the model increases by 16 percentage points (from 0.195 to 0.355) when including ownership concentration.
Using welfare state regime dummies in Models 5 and 7 leads to a similar picture with regard to the role of media ownership concentration. Higher levels of ownership concentration are associated with lower support for redistribution. In Model 5, the significance of ownership concentration slightly fails to reach the 5 per cent threshold. Inspection of the data reveals that Czechia is an outlier regarding its association between ownership concentration and redistribution attitudes (Table A.1 in the online appendix). 10 If Czechia is dropped from the sample, the effect of ownership concentration is statistically significant (Model 7).
Regarding regime effects, not accounting for media ownership concentration, support for redistribution is significantly higher in the Southern European countries compared to the reference category of the liberal welfare states (Models 6 and 8). In Model 8, redistribution support is also higher in the Eastern European countries. These effects are in line with earlier findings (e.g., Dallinger, 2010). The other regime dummies are statistically insignificant. However, when including ownership concentration in Models 5 and 7, the effect of the Southern European countries disappears. The effect for the Eastern European countries remains significant in Model 7 but decreases in size. 11 The share of explained variance increases to 0.478 (from 0.384) when adding ownership concentration in Model 5 (and to 0.734 (from 0.612) in Model 7). Taken together, including a measure of media ownership concentration contributes substantially to explaining crossnational variation in support for redistribution, and to some extent it cancels out (the partly unexpected) welfare state regime effects found in earlier research.
The effects of ownership concentration are robust to alternative model specifications (Table A.6 in the online appendix). Including availability of politically independent and adequately funded public broadcasters as an additional control variable does not affect attitudes towards redistribution, and it does not alter the effect of ownership concentration. The same applies to the different additional control variables for the socio-economic situation of a country (inflation, GDP growth, and unemployment), population size, and individual news consumption (Tables A.6 and A.7 and Figure A.2 in the online appendix). Irrespective of these additional control variables, higher concentration of media ownership is associated with lower support for redistribution. It is also important to note that levels of ownership concentration are among the highest in the Nordic countries (Finland and Sweden; Table A.1 in the online appendix) for which Larsen (2013) has identified a relatively favorable coverage of issues related to poverty and inequality. Dropping Finland and Sweden from the analysis does not change the results (Table A.6, Model 5). In sum, across model specifications, higher levels of ownership concentration are significantly associated with lower support for redistribution.
As a final consistency check, I include another indicator from the MPM capturing owners' control over editorial content and interact it with ownership concentration. To the extent that increasing ownership concentration affects individual-level support for redistribution by increasing (unobserved) media bias, this effect should be most pronounced where actual influence of owners over news content is strong. Predicted probabilities show that the effect of ownership concentration is indeed particularly pronounced when ownership control over news content is above average (Figure A.1 in the online appendix). The marginal effects of changes in ownership concentration are significant at the 5 per cent level at high levels of ownership control (at the 75th percentile of ownership control), but not when ownership control over editorial content is negligible (at the 25th percentile of ownership control). This finding lends further confidence to the claim that the effects of ownership concentration on individual redistribution support are mediated by owners' influence on news coverage.

Conclusion
In this contribution, I argue and show empirically that characteristics of the mass media, more specifically, the degree to which media ownership structures are concentrated, should be taken into account when studying attitudes towards redistribution. Media reporting has the power to shape fundamentally the way individuals interpret matters related to inequality and redistribution. With concentrated ownership structures, the risk of low or unfavorable coverage of such issues increases, with possible consequences for individual-level attitude formation. In the following, I outline the wider implications of these claims.
Acknowledging the role of the mass media has important implications for welfare state research. Existing research on the role of welfare state regimes and social inequality as macro-level explanatory factors of attitudes towards redistribution has produced inconsistent findings and has left substantial variation in attitudes unexplained. Adding media ownership concentration to these established indicators reveals an independent and substantial effect of media ownership on redistribution attitudes. The more concentrated ownership is, the lower is the support for redistribution. Including a measure of media ownership increases substantially the explanatory power of the regression models, and reduces the explanatory power of welfare state regimes. While evidence from survey experiments has indicated that such media framing effects on public attitudes could exist (e.g., Hopmann et al., 2017), this study is, to the best of my knowledge, the first that shows empirically that national media ownership structures are systematically associated with levels of public support for redistribution.
The findings of negative effects of media ownership on individual redistribution support are also relevant for broader research in comparative political economy and economic sociology. Future research could examine to what extent media characteristics and news coverage contribute or not to a dominant discourse with respect to economic growth and economic policies (Baccaro & Pontusson, 2016;Ferrara et al., 2021;Kneafsey & Regan, 2020), to perceptions of wealth and preferences regarding its taxation (Beckert & Arndt, 2017;Piketty & Cantante, 2018), and to various other related issues.
Taking seriously the role of the media in opinion formation also highlights the importance of media regulation. In principle, governments have instruments available to counteract processes of increasing ownership concentration (Kennedy & Prat, 2019;MPM, 2016). Studying why governments might refrain from implementing and enforcing regulation to avoid increasing concentration of media providers becomes a promising research question. With only very limited research being available on cross-national variation in news coverage of social inequality and redistribution (cf. Grisold & Preston, 2020), future research should also analyze in more detail the causal links of how media structures shape news coverage, and how news coverage ultimately affects individual attitudes and empowers or disempowers individuals in holding their governments accountable.
Finally, the elephant in the room in this study, not scrutinized here due to data limitations, is the change that is brought about by digitalization.
Digitalization has dramatic consequences for ownership concentration patterns, and news coverage and consumption, given the prominence of social media usage and widespread possibilities to consume news online (Zhuravskaya et al. 2020). Studying the consequences of digitalization for media systems, news consumption, and public opinion provides an exciting and important avenue for future research. Notes 1. The association between inequality and support for redistribution has been found to vary across countries with evidence of a negative association between inequality and redistribution support in the US and other liberal welfare states (Dallinger, 2010;Kelly & Enns, 2010). 2. Researchers have started to address the limitations in external validity. King et al. (2017) demonstrate the existence of framing effects in a more realistic natural experiment. Lecheler et al. (2015) show that framing effects can persist over time beyond the immediate experimental setting. 3. Evidence from social psychology illustrates the underlying logic of how media coverage of crime may affect individual attitudes towards redistribution. Experiencing feelings of existential threat increases individuals' priority for safety and security, thereby making individuals more conservative, more supportive of the status quo and less open to social and political change (Jost et al., 2009, p. 322). 4. Lack of market competition is considered to be particularly pronounced for certain local media markets (Van Aelst et al., 2017, p. 11) where economies of scale often magnify tendencies of increasing concentration that may lead to local monopolies. 5. Austria, Belgium, Czechia, Germany, Estonia, Spain, Finland, France, United Kingdom, Hungary, Ireland, Italy, Lithuania, Netherlands, Poland, Portugal, Sweden, and Slovenia. 6. It is difficult to specify clear expectations regarding the role of the individual time spent on news consumption because individuals also receive their information indirectly through social exchange with their family, friends or colleagues (Eveland & Hively, 2009). Nevertheless, findings of negative effects of ownership concentration across levels of news consumption would provide further confidence in the results. 7. See Table A.2 in the online appendix for further details. Data and full documentation are available online (https://cmpf.eui.eu/media-pluralism-monitor/, accessed 04 September 2020). 8. Di Gioacchino and Verashchagina (2020) use the aggregate indicator on risk to market plurality (see Table A.2 in the online appendix for the operationalization of the corresponding sub-indicators). However, for some of the sub-indicators, it is less straightforward to predict how they should be related to media discourse and public attitudes. Additional models including these different subindicators show that horizontal ownership concentration is most tightly associated with attitudes towards redistribution (Table A.4 in the online appendix). 9. The results hold if the Eastern European countries are dropped from the analysis (Table A.5 in the online appendix). 10. Table A.1 shows that in Czechia, the risk of horizontal ownership concentration is slightly below average while redistribution preferences are clearly below average. However, the country scores clearly above average on the remaining indicators of risks to media pluralism. This implies that media characteristics unrelated to horizontal ownership concentration present a risk to media pluralism in the Czech case which could affect attitudes towards redistribution. 11. Marginal effect estimates based on the regression results in Models 7 and 8 indicate that the difference in preferences between the liberal and Eastern European countries is reduced from 0.40 to 0.29 scale points when media ownership concentration is included in the model.