Covering the crisis: Media coverage of the economic crisis and citizens economic expectations

Original Article Covering the crisis: Media coverage of the economic crisis and citizens economic expectations Hajo G. Boomgaarden*, Joost van Spanje, Rens Vliegenthart and Claes H. de Vreese Amsterdam School of Communication Research, University of Amsterdam, Kloveniersburgwal 48, 1012 CX Amsterdam, The Netherlands. E-mail: H.Boomgaarden@uva.nl *Corresponding author. Abstract The 2008 2009 worldwide economic crisis serves as a backdrop to this study of the dynamics of citizens economic expectations. Economic expectations are identified as crucial for a range of political attitudes. This study is the first to consider how information affects evaluations in times of a severe crisis, as prior research of information effects on economic evaluations took place in more stable economic times. It links citizens news exposure and the content of economic news coverage with changes in prospective economic assessments. Drawing on a three-wave panel study and on a media content analysis between the panel waves, we thus provide a dynamic assessment of media influences on changes in economic evaluations. The results demonstrate that media exposure strongly affected expectations regarding the future development of the national economic situation, while being largely unrelated to personal economic expectations. We furthermore show that media dependency increases the magnitude of the media effect. We discuss the disconnect between personal and national economic evaluations with regard to mass-mediated economic information. Acta Politica (2011) 46, 353 379. doi:10.1057/ap.2011.18; published online 8 July 2011 Keywords: economic evaluations; media effects; economic news; media dependency Introduction The economic crisis that took its start in 2008 was the most severe crisis in Europe since World War II. Common wisdom and research (for example, Weatherford, 1978) suggest that such an economic downturn has considerable consequences in the political and economic realm. Citizens economic expectations are susceptible to such real-world developments. The expectations that citizens hold regarding the future development of the economic situation www.palgrave-journals.com/ap/

Boomgaarden et al can affect, for instance, their consumption patterns (for example, Van Raaij, 1989), their voting behaviour (Lewis-Beck, 1988; Duch, 2007), their support for European integration (Gabel, 1998; De Vreese and Boomgaarden, 2005) or immigration and welfare policy support (for example, Kluegel and Smith, 1986; Burns and Gimpel, 2000; Mayda, 2006). Citizens economic expectations are therefore a natural starting point for research on the political and economic consequences of an economic malaise. This study considers dynamics of citizens economic expectations during an unprecedented economic crisis. Citizens economic expectations are (partly) based on information about the economy (Kiewiet, 1983). There are (at least) three ways in which citizens economic expectations may be affected through personal experience, interpersonal communication and through the mass media (Mutz, 1992). In this article, we focus on the latter, information about the economy from the media. We assess to what degree media coverage of economic prospects has an impact on changing economic expectations in times of crisis. We furthermore address the question of which kinds of economic expectations are affected by the mass media. Does new information influence perceptions of one s own economic situation or only that of the entire country? (Mutz, 1992; see also, for example, Kinder and Sears (1981) for different outcomes of these types of economic assessments) As the media mostly cover the economic health of a nation or of sizable social groups within the nation, we follow prior studies by focussing on the influence of the media on assessments of the national economy (Mutz, 1992; Shah et al, 1999). We proceed by comparing this to media influence on assessments of personal economic expectations. We pose that evaluations of the national economy are largely influenced by mediated information. By contrast, citizens personal experiences, not media coverage, may matter more to evaluations of personal economic expectations. The study explains changes in evaluations, thereby conservatively estimating the influence of information between two points in time. Finally, we take from literature in political communication that media effects are likely to differ in strength for different segments of the audience (see, McLeod et al, 2002). By closely linking indicators of media coverage to different types of economic assessments through an integration of media content and survey data, and by considering a potential moderation of media effects through news dependency, this study refines the understanding of the relationship between economic media coverage and citizens evaluations. We address these questions drawing on data from an original three-wave panel survey study and a media content analysis in the Netherlands. With its relatively open, internationalized economy, the Netherlands was greatly affected by the economic downturn and is therefore a valuable case to learn from for other post-industrial economies. The Dutch Gross Domestic Product (GDP) contracted by 4 per cent in 2009, 1 characterizing the most dramatic 354

Covering the crisis economic decline in decades. The first wave of the panel study dates back to November 2008, when Wall Street had just crashed following the bankruptcy of Lehman Brothers Holdings Inc in the United States. Its third, and last, wave was fielded in February 2009, when the crisis had already been ravaging for a number of months. In addition, media content was tracked between the three panel waves. This combination of data offers a unique opportunity to assess media influences on the dynamics of economic perceptions during an economic meltdown, allowing for a careful consideration of whether expectations based on studies that were conducted in relatively stable economic times translate into times of crisis. Our approach allows specifying which type of economic expectations is affected by media coverage. Economic evaluations and deviations from economic reality The literature on citizens economic expectations in established democracies is mostly based on data collected at times when economic developments were relatively stable. Thus, this study is among the first to deal with economic perceptions at times when economic prospects rapidly and unequivocally deteriorated. Talk about the economic crisis is likely to make people uncertain about their economic prospects, and to alleviate this uncertainty they engage in information-seeking behaviour (Berger and Calabrese, 1975). This, in combination with heightened media attention for economically bad times (Fogarty, 2005), potentially renders mass-mediated economic information in times of crisis even more important. Following the vast amount of studies in the area of economic voting, it is important to distinguish between different dimensions of economic evaluations (or assessments or judgements). One dimension relates to the type of situation that is judged, typically distinguishing evaluations of citizens personal economic situation (egocentric evaluations) and those of a nation s economic health (sociotropic evaluations). The second dimension relates to time, and differentiates between judgements of how the economic situation developed in the past (typically over the past year) (retrospective evaluations) or what it will be like in the future (typically in the coming year) (prospective evaluations) (for example, Lewis-Beck and Paldam, 2000). In this study we focus on prospective evaluations (see also Sanders et al, 1993; Nadeau et al, 1999). This is because the widely announced economic malaise immediately started to dominate world news. It is therefore likely that evaluations regarding future developments will overshadow retrospective ones. Moreover, under such circumstances retrospective evaluations have limited relevance and prospective evaluations are more important for political judgements and consumer behaviour. Furthermore, and in line with prior studies (Mutz, 1992), we 355

Boomgaarden et al expect media effects on sociotropic evaluations only, and to ascertain the uniqueness of this effect we compare the impact of information sources on both sociotropic and egocentric prospective assessments. Rational choice theorists point out that it is not rational for citizens to be permanently fully informed about the state of the economy, just as they are rationally ignorant about current affairs in general (Downs, 1957). Yet, there is empirical evidence that many citizens have, in fact, quite some knowledge on economic matters such as their country s unemployment rate, in particular at election times (for example, Paldam and Nannestad, 2000). This is not to say that the perceptions or expectations that citizens have always accurately reflect the state of affairs. There are several reasons why citizens perceptions would deviate from reality (for example, Hetherington, 1996). A first explanation holds that different citizens use different criteria in their evaluation of how well the economy is doing (Kinder et al, 1989). A second explanation is associated with partisanship (for example, Kramer, 1983), holding that citizens perceive economic conditions as more favourable if their preferred party is in power than if it is not (Wlezien et al, 1997; Van der Eijk et al, 2007). A third explanation at the core of this study is that citizens receive varying information from different sources about the state of the economy and form their evaluations accordingly. The role of the mass media as central information providers is discussed next. Mass media and economic assessments The subjective economy, that is, citizens perceptions and evaluations of the economy, features prominently in studies on the political impact of the economy. Early interest in the mediating role of the mass media in the economic voting model (for an overview, see Key and Cummings, 1966; Lewis-Beck and Stegmaier, 2000) stems from the observation that subjective assessments of the economy rather than objective indicators were successful in explaining voting behaviour and/or election outcomes and that therefore there must be a distortion between the subjective and the real economy (for example, Sanders, 2000; Sanders et al, 2001). Hetherington (1996), for instance, showed the US economy to recover before the 1992 US Presidential elections, while mass-mediated information about the state of the economy was continuously negative. On the basis of this information, he argued that Clinton s victory in this election was more based on a perceived economic situation than on reality. Similar observations have been made regarding the 2001 UK General election (Sanders et al, 2001) and the 1998 German Bundestag election (Donsbach, 1999). 2 In his classic account, Kiewiet (1983, p. 22) argues that information about conditions in the nation s economy [y] must be obtained from newspapers 356

Covering the crisis or television. In one of the earliest empirical assessments of the role of the media, Mosley also states that the economy voters see [y] is a mixture of impressions gained from television, radio and the popular press (Mosley, 1984, p. 129). Other studies acknowledge the relevance of mass media for the formation and/or dynamics of subjective economic assessments, without, however, providing empirical evidence (for example, Weatherford, 1983; Conover et al, 1986; Clarke and Stewart, 1994). Research about media influences on economic assessments is mostly based on macro-level analyses. In his time-series study, Mosley (1984) finds media estimates of the economic situation to be a far better predictor of economic assessments than official economic indicators. Relying on a similar design, Sanders et al (1993) and Goidel and Langley (1995) find the tone of economic news coverage to be affecting public assessments, in addition to the impact of real-world factors, in the United Kingdom and United States, respectively. The latter, however, show that only negative news coverage affects public evaluations, which is in line with Soroka s (2006) more recent analysis (see also Ju, 2008). Others come to different conclusions based on macro-level analyses, such as Haller and Norpoth (1997) who find little, or Wu et al (2000) who find no impact of economic news on assessments. Only few investigations address the impact of economic news coverage on the level of the individual citizen. Hetherington (1996), considering the case of the 1992 US Presidential election mentioned above, claims that news coverage of the economy was predominantly negative, and shows that media consumption during the campaign had a highly significant and negative effect on voters retrospective evaluations of the national economy (p. 381). De Boef and Kellstedt (2004) found indirect effects of media coverage of the economy on consumer sentiment, after controlling for actual economic conditions. When news coverage was positive, citizens gave favourable evaluations, leading to more positive sentiments. Some studies go further to explicate the mechanism through which the media can affect economic evaluations. Sheafer suggested that media cues regarding the general direction of the economy should affect perceptions of the economy and eventually voting behaviour. He highlights that one main framing mechanism is the presentation of change in the environment (Sheafer, 2008). It appears that individuals are particularly sensitive to changes in a positive or negative direction (Boettcher, 2004, p. 333), and that attitude change must y be understood as a change in the balance of positive and negative considerations relating to a given issue (Zaller, 1992, p. 118). This idea is based in general on the research of Kahneman and Tversky (1979, 1984), who find that individuals are more sensitive to changes in a certain indicator than to the actual level of the indicator. Based on the above we expect an influence of the tone of 357

Boomgaarden et al economic media reports on economic assessments (above and beyond the impact of real economic indicators). We proceed to argue that these media effects are only likely to occur for assessments of the national economy (sociotropic evaluations), and not for assessments of the personal economic situation (egocentric evaluations). The economy is partly an obtrusive subject (Zucker, 1978) a doorstep issue (Haller and Norpoth, 1997) in that citizens may notice that others in the immediate surrounding lose their jobs, that the cost of living increases, that they themselves lose their job, or that shops in their neighbourhood go bankrupt. However, most developments of national economic scope are not all that easy to observe first-hand; here the economy is an unobtrusive issue. Apart from via the media or from talking to others, how should citizens learn about, for instance, changes in the GDP? Mutz (1992) considers how problem perceptions of unemployment are influenced by media coverage. Distinguishing different information sources and different types of problem assessments, she finds that personal experience with unemployment significantly contributed to perceptions of unemployment as a personal problem, whereas exposure to newspaper coverage about unemployment uniquely identifies perceptions of unemployment as a social problem (p. 496). In more general terms, mass media influences are expected to occur with regard to evaluations of the national economic situation (sociotropic) but not concerning egocentric assessments. We contrast media influences on sociotropic versus egocentric expectations, while also controlling for personal economics, such as the household income or whether someone is currently employed or not. Finally, we focus on differences in the potential effects of positive versus negative economic news coverage. Psychological studies have repeatedly shown that people pay greater attention to negative information as compared to positive information (for example, Fiske, 1980; Pratto and John, 1991), which makes them more likely to draw on this information when forming an opinion. In fact, prior research in the field of political communication confirms that negative information (also in the form of threat or risk frames) tends to have stronger effects on public opinion dynamics compared to positive information (for example, Schuck and de Vreese, 2006; de Vreese et al, 2011). These observations square well with the prevalence of threats and negativity in classic persuasion studies (for example, Hovland et al, 1953). Accordingly, and paying tribute to recent studies in the field of media and the economy (for example, Soroka, 2006) that argue that people are more likely to pick up negative information about the economy, we expect the impact of negative economic news to weigh stronger than that of positive news. This study contributes to the literature by relying on a dynamic design considering changes in economic expectations at the individual level. We are 358

Covering the crisis able to link in a direct manner the economic information that citizens were exposed to between panel survey waves to changes in their economic expectations. In order to do so, we ensure a strong link between the content analytical indicators and citizen evaluations, by specifically coding news assessments of expectations of future economic developments, and linking these specifically to public expectations (prospective assessments). Moreover, we differentiate between prospective evaluations of the personal versus the national economic situation and highlight the importance of a proper specification of the dependent variable. Finally, we pay attention to differential effects of positive and negative economic news coverage. The study furthermore, in line with developments in media and public opinion research, takes the conditionality of media effects into account, as elaborated in the next section. The contingency of the media impact The assumption that mass media content affects all citizens equally is somewhat naı ve (for example, Delli Carpini, 2004) and political communication research pays increasing attention to the contingency of media effects (McLeod et al, 2002; Walgrave and Van Aelst, 2006). We pursue this approach in our study of media effects on economic assessments. The degree to which individuals rely on the media for understanding and interpreting their surroundings is a key component of understanding the media s impact on opinions and evaluations. Media system dependency theory (Loges and Ball-Rokeach, 1993), abbreviated as MD, argues that the degree to which people depend on the news for informational goals matters for whether news coverage affects their perceptions (Morton and Duck, 2001). MD theory thus takes its starting point in the observation that the impact of media messages on audiences is a function of how dependent audiences are on (mass) media: the higher the dependency, the larger the effects. Initial support for the theory (for example, Ball-Rokeach et al, 1984) has been corroborated by experimental research on health communication effects. Morton and Duck (2001) found that the impact of mass communication on both personal and impersonal perceptions was moderated by self-reported dependency on mass-mediated information. MD theory and the argument why it should moderate media effects is based on motivational factors and thus has conceptual links with uses and gratifications research (for example Blumler, 1979) and later work on the notions of news elaboration and news surveillance introduced in the cognitive mediation model outlining how motivations affect media use and subsequent effects (for example, Eveland, 2001; Eveland et al, 2003). The uses and 359

Boomgaarden et al gratification perspective highlights surveillance as one of the most important reasons for using news media (Becker, 1979). As noted by Eveland et al (2003, p. 363) surveillance motivations are represented in the information transmission function of the news; that is, they represent a desire to learn from the news as part of a more basic need to monitor the environment. In our view, the central role of motivations for staying informed is strongly related to self-reported dependency on the media for information that is at the core of MD theory. Both express a fundamental cognitive motivation (Blumler, 1979) for media use and we expect that this dependency on the media for keeping up to date should extend beyond health communication or political learning and also apply to changes in attitudes in general, thus also for dynamics in economic expectations. This study brings together the two perspectives outlined above. It is expected that the more someone reports to depend on the mass media for functioning and being informed, the more likely he or she is to be affected by the contents of such media. 3 Hypotheses This study focuses on media influences on changes in citizens economic assessments. We are in particular interested, following the existing literature (for example, Soroka, 2006), in the potential impact of the tone of reporting, in this case mass-mediated future economic expectations, on prospective, sociotropic economic assessments. In accordance with the literature reviewed above we formulate the following hypotheses: Hypothesis 1 (H1): Hypothesis 2 (H2): Hypothesis 3 (H3): Hypothesis 4 (H4): Exposure to economic news leads to changes in sociotropic economic expectations, so that negative economic news leads to more negative expectations and positive news to more positive expectations. Exposure to negative economic news has a stronger effect on sociotropic economic expectations than exposure to positive economic news. The influence of economic news on sociotropic expectations is moderated by media dependency. Citizens who depend more on media for information monitoring are more strongly affected by economic news coverage. Egocentric economic expectations are not affected by exposure to economic news. 360

Covering the crisis Methods We rely on three-wave panel survey data and a news content analysis between the panel waves to address the expectations formulated above. Respondents reported whether and to what degree they were exposed to outlets that were subject to the content analysis, and media content and survey data were integrated. Relying on such a design allows assessing what kind of economic information respondents were exposed to between different time points and how this information exposure affects changes in their economic outlook. Content analysis The unit of coding is an article in a newspaper or a news item in a TV newscast. A TV news item is defined by its topic. It consists of all story elements (films, interviews, statements and so on) that are presented as belonging to the same topic. 4 We have coded the content of each item of the two main news broadcasts in the Netherlands, the NOS Journaal (20h00 daily) and the RTL 4 Nieuws (19h30 daily) during the 3-week period between wave 1 (w1) and wave 2 (w2) of the panel. Weather forecasts and specific sections devoted to sports news were not coded. In addition, we have analysed the content of nine national newspapers, both quality papers and (free) tabloids (De Volkskrant, NRC Handelsblad, Trouw, Algemeen Dagblad, De Telegraaf, Metro, Sp!ts, NRC Next and De Pers) in the period between w1 and w2. The front pages of each newspaper were coded in their entirety. For the period between wave 2 and wave 3 (w3), we relied on a more confined sample of news outlets, and coded the front pages of all newspapers mentioned above except for De Pers. The four coders were asked to code whether or not the story provided or suggested a prediction, assumption or expectation for the economic situation either in the Netherlands or for the Dutch people during approximately the next year. The specific coding instructions were: Does the story provide or suggest an expectation, assumption or prediction of the personal economic situation of the Dutch people/the Dutch economy regarding the period of the next 12 months (approximately)? If so, is this a positive or a negative expectation, assumption or prediction? Each of the selected items could be coded 2 (negative), 1 (rather negative), 0 (balanced), þ 1 (rather positive) or þ 2 (positive) for either national or people-specific economic coverage. Inter-coder reliability for the country s mediated economic expectations was Cohen s Kappa ¼ 0.76 and for predictions of the personal economic situation of the Dutch citizens it was Cohen s Kappa ¼ 0.88. 5 Both reliability scores are satisfactory. 6 News items occasionally made references that pertained to both 361

Boomgaarden et al variables. The correlation between the two variables is 0.66 (between w1 and w2) and 0.76 (between w2 and 3). Owing to practical constraints, we relied on a more limited selection of media outlets for the second period. However, descriptive results demonstrate considerable similarity across outlets for that period. For instance, two of the three initially most negative outlets remain most negative in the second period (Metro period 1: 2, NRC Handelsblad period 1: 1.78, Metro period 2: 2, NRC Handelsblad period 2: 1.69). Moreover, in the first period nine of the eleven outlets range between 1 and 2 in terms of national economic evaluations, and this goes for seven out of the eight outlets analysed in the second period. Furthermore, a replication of the analyses for the first period that incorporates only those outlets that were also available in the second period yields highly similar results. This makes us confident that our (still broad, but more confined) selection of outlets for the second period suffices. Survey data We test the hypotheses on the basis of data from a three-wave panel survey of Dutch citizens eligible to vote. From an online panel of 143 809 citizens, 2400 persons over 17 years old were randomly selected, and invited to fill out an online questionnaire. Of these persons, 1394 completed the questionnaire, 7 which yields a response rate (RR1) of 58 per cent. When compared to the census data from the Dutch electorate, groups underrepresented in our sample are men (47.0 per cent versus 49.4 per cent), young citizens (30.8 per cent versus 34.2 per cent), and those who had intermediate vocational education (31.3 per cent versus 48.0 per cent). After the invitation in November 2008, all respondents received a participation request again 3 weeks later, in December 2008. All 1394 initial participants were re-contacted, of whom 1127 cooperated. This means a sample loss of (1394 1127)/1394 ¼ 19 per cent. 8 In addition, a fresh sample was drawn based on gender, age, education, household size and region of residence. This sample of 285 additional respondents was invited to fill out the w2 questionnaire, of whom 166 participated (AAPOR RR1 ¼ 58 per cent). In February 2009, the third wave took place. 9 The full sample of 2400 was re-contacted and invited to take part in w3. Of the 2400 persons in the sample, 1174 (49 per cent) completed the third questionnaire. 10 In light of the participation rate of these respondents in the first two samples, this means that the w2 w3 attrition rate was (1293 976)/1293 ¼ 25 per cent, and the overall (w1 w3) sample loss (1394 976)/1394 ¼ 30 per cent. The underrepresentation of men reported for the w1 sample is somewhat smaller (48.6 per cent versus 362

Covering the crisis 49.4 per cent) among those who completed all three waves. By contrast, the underrepresentation of young voters (28.0 per cent versus 34.2 per cent) and those with intermediate vocational education (30.9 per cent versus 48.0 per cent) is slightly larger among the 976 three-time participants than in the original sample. Our dependent variables were measured in each of the three panel waves, by means of the following questions. Sociotropic prospective economic evaluations were measured by asking respondents: Looking at the economic situation in the Netherlands, do you think the situation will be better or worse 12 months from now? For measuring egocentric evaluations we asked: Looking at your personal economic situation, do you think the situation will be better or worse 12 months from now? Respondents were asked to indicate their view on a scale ranging from much worse (1) to much better (7) (see results section for descriptive information). Our main independent variable is exposure to economic news. We asked respondents to report how many days per week they would use each of the news outlets that were subject to the content analysis. The question wording was How many days in a typical week do you watch the following TV programmes and/or read the following newspapers? (all question wordings are translations by the authors from the original Dutch questionnaire). Responses were given on 8-point scales, from 0 to 7 days per week. We integrated the media content data (see previous section) and the survey data according to the following procedure: for each respondent we matched the media data with reported media exposure in such a way that the respondent was assigned a score based on a multiplication of the reported frequency of use per outlet and the aggregated mean of the relevant content characteristics of that outlet in the period between the interviews (for a similar procedure, see De Vreese and Semetko, 2004). Before integrating the content data, we in line with Soroka (2006) weighed the negative news twice. People are argued to be more likely to pick up negative rather than positive information, which is reflected in this simple weighting procedure. The index measure then is a weighted measure of exposure to positive and negative economic news. We use two different indices for the two different dependent variables, in order to have a close match between the information that is provided by mass media and the expectations this information may affect. That means we have one variable that integrates the tone of national economic expectations in the media and exposure, and one that links the tone of the economic expectations for the Dutch people in the media with exposure. The resulting scores are mean-centred before being included in the analysis. Three items in our models reflect the personal economic situation and serve as control variables. Respondents reported their gross net household income in euro per year (classified into 27 categories from low to high, 363

Boomgaarden et al Median ¼ 15 E 34 000 45 000, inter-quartile range ¼ 3 categories, E 23 000 28 000 to E45 000 56 000), their occupational status (recoded into a dummy variable distinguishing between those respondents who are currently employed versus those who were in education, who had retired, or were unemployed at the time of the survey, M ¼ 0.68, SD ¼ 0.47), and their highest level of education, derived from responses to the question: what is your highest educational level, regardless of whether completed or not? (in six categories from low to high, Median ¼ MBO (third lowest category), inter-quartile range ¼ 3 categories, from the second lowest to second highest). Media dependency was measured by a battery of five items. These items tap the motivational factors for media use and are adapted from Morton and Duck (2001) and Eveland et al (2003). Respondents positioned themselves on a scale ranging from strongly disagree (1) to strongly agree (7) concerning the following five statements: (i) I use the news to see how politicians stand on issues, (ii) I use the news to keep up with important political issues, (iii) I use the news to help me make up my mind about things, (iv) I often try to relate what I see on TV news or read in the newspaper to my own personal experiences, and (v) I often try to think about how what I see on TV news or read in the newspaper relates to other things I know. A factor analysis with principal component extraction shows respondents answers to these questions loading on one factor (Eigenvalue of first component ¼ 3.38, of other componentso1, explained variance 68 per cent). A scale variable including all five items has been constructed, which is highly reliable (Cronbach s a ¼ 0.88). In accordance with the relevant literature (see above), we call this component media dependency. The scaling and reliability analyses reveal that our assumption of a close relationship between the concepts of dependency and motivation is empirically supported. The variable was mean-centred before including it in the analysis. The interaction term between exposure to economic news and media dependency is composed by multiplying the mean-centred version of both variables. In addition to considering influences of exposure to economic news coverage and personal economics, the models control for citizens assessments of the national government: How satisfied or how dissatisfied in general are you with the current government? Answers were given on a 7-point scale ranging from very dissatisfied (1) to very satisfied (7). Recent research suggests that people who approve of the government are more likely to have a positive economic outlook (van der Eijk et al, 2007). Both crosssectional (Wlezien et al, 1997; Van der Eijk et al, 2007) and dynamic analyses (Anderson et al, 2004; Evans and Andersen, 2006) show that voters perceptions of the economy are affected by their political preferences. This holds for both retrospective and prospective economic evaluations (Anderson et al, 2004). In particular during the 2008 2009 economic crisis 364

Covering the crisis in which the government took an active role, it is important to control for this measure. 11 In addition to the mass media, economic information may stem from interpersonal discussions. Our data do not, however, include a detailed measure of the frequency and tone of interpersonal talks about the economy. We use the frequency of political discussion as a proxy measure: How often do you talk about political issues with friends, family or colleagues?, measured on a 7-point scale from (1) never to (7) very often. Finally the models include a control variable for respondents who do report to not consume any news at all. Our weighted exposure measure includes a broad, but nevertheless limited selection of newspapers, and there are obviously other outlets that people can turn to for news. Including age, education and income as control variables in the models also means that we control for the best part of the most important predictors of news consumption patterns (for example, Lauf, 2001; Elvestad and Blekesaune, 2008), thereby reducing the chance of self-selectivity in news consumption to influence our results. Moreover, our panel design with a relative short period between panel waves circumvents problems of selfselectivity that are more common in studies relying on cross-sectional survey data. Analysis In order to assess the hypotheses, we perform multiple OLS regression analyses. We present four models for sociotropic and four models for egocentric expectations, regressing economic expectations at t2 (second wave) and t3 (third wave) on the variables introduced above, while controlling for economic expectations at t1 (first wave) or t2 (second wave), respectively. The use of a lagged variable puts a focus on change of the dependent variable between two panel waves (for example, Markus, 1979). Moreover, controlling for prior economic assessments decreases the likelihood of our models being underspecified, as predictors of static economic evaluations are taken account of by including the lagged term. 12 Results We start with a short section of descriptive results of the content analysis before discussing explanatory models of economic expectations. The percentage of news items that qualifies as evaluative economic news varies across the outlets that were examined during the first analysis period (w1 w2). The daily NRC Next (2 per cent) devoted relatively the least attention to economic issues 365

Boomgaarden et al on the front page, and Algemeen Dagblad (12 per cent) the most. On average, 6 per cent of the news stories included references to the economy. The average number of stories is 151 per outlet, and varies from 75 (Sp!ts) to 229 (Volkskrant). Figure 1 shows descriptive analyses of our data concerning the tone of economic news coverage regarding expectations about the future development of the Dutch economy. The bars indicate the mean tone per type of outlet TV newscasts (NOS and RTL), quality newspapers (De Volkskrant, NRC Handelsblad, Trouw and NRC Next) and sensationalist tabloid papers (Algemeen Dagblad, De Telegraaf, Metro, Sp!ts and De Pers). Although economic news was, not surprisingly, overall negative, the average tone varies by outlet type. Moreover, we find differences in tone between the two periods from w1 to w2 and from w2 to w3. These differences allow us to estimate the effect of the tone of the economic news consumed by the respondents. Quality newspapers overall had the most negative coverage of the economic future in both periods. For these outlets on this aggregate level we hardly see any change in coverage between November 2008 and February 2009. Tabloid newspapers were also very negative about the future of the national economy in the first period, but remarkably less so in the second period. TV news coverage was on average equally negative as newspaper coverage was in the first period. Also within types of news outlets we find variation in tone, with overall the broadsheet NRC Handelsblad being the most pessimistic of the newspapers and the sensationalist de Telegraaf being the least pessimistic. We furthermore see a strong overlap with media assessments of the future of the personal economic situation of the Dutch people (not shown here). Also here we have a uniform negative picture in both periods, again with 2 1.5 1 0.5 0-0.5-1 -1.5-2 Wave 1-2 Wave 2-3 TV news Quality newspapers Tabloid newspapers Figure 1: Tone economic news coverage by type of outlet. Note: Values are mean scores of evaluations given in the different outlet types on a scale from 2 to2. 366

Covering the crisis some differences between different types of media (results can be obtained from the authors upon request). Having demonstrated that the overall tone in the media regarding future economic developments was rather negative, but varied across outlets, we now turn to the survey data. How did sociotropic and egocentric expectations develop over the three time points? The aggregate trend of both measures is illustrated in Figure 2. Interestingly, the mean scores of sociotropic economic expectations clearly decrease over time in an almost linear fashion, whereas those of egocentric economic expectations remain remarkably stable between November 2008 and February 2009. The average score on national economic expectations drops from M ¼ 3.21 (SD ¼ 1.06) in November 2008 to M ¼ 2.52 (SD ¼ 1.17) in February 2009, a drop of 0.69 points on a 1 7 scale (N ¼ 976). By contrast, the personal expectations (N ¼ 976) are about as high in November (M ¼ 3.52, SD ¼ 1.01) and December (M ¼ 3.40, SD ¼ 1.06) as 2 months later (M ¼ 3.40, SD ¼ 1.07). Overall, the data show that 52 per cent of respondents changed their egocentric economic expectations between w1 and w3, whereas 67 per cent changed their sociotropic expectations. In the following, we will assess to what extent the variation in both variables may be explained by news media content. We first turn to explaining expectations regarding the development of the national economy as shown in Table 1. Models A (for the first time period) and C (for the second time period) show that exposure to news about the future of the national economy is significantly related to changes in prospective sociotropic evaluations. This Figure 2: Trends in aggregated economic expectations. Note: Values are mean scores of prospective economic evaluations on a 7-point scale from much worse (1) to much better (7). To illustrate the differential trends we omit parts of the scale. No measurement point in January 09. 367

Boomgaarden et al Table 1: Explaining change in sociotropic economic expectations Sociotropic economic expectations at t2 (model 1) Sociotropic economic expectations at t3 (model 2) Model A Model B Model C Model D B Beta SE B Beta SE B Beta SE B Beta SE Intercept 1.058** 0.308 1.015** 0.313 1.413*** 0.334 1.236*** 0.341 Sociotropic economic expectations at t1 0.492*** 0.476 0.031 0.495*** 0.479 0.031 0.404*** 0.375 0.038 0.404*** 0.374 0.038 (model 3)/t2 (model 4) Age 0.000 0.001 0.003 0.000 0.003 0.003 0.005 0.063 0.004 0.004 0.052 0.004 Household income 0.012 0.031 0.013 0.013 0.032 0.013 0.006 0.013 0.015 0.006 0.014 0.015 Employed (dummy) 0.085 0.037 0.104 0.084 0.037 0.104 0.085 0.035 0.126 0.095 0.039 0.126 Education 0.038 0.055 0.023 0.038 0.056 0.023 0.000 0.000 0.027 0.001 0.002 0.027 Government evaluations 0.085*** 0.111 0.024 0.085*** 0.111 0.024 0.092** 0.116 0.028 0.092** 0.116 0.028 Frequency political discussion 0.028 0.036 0.024 0.029 0.037 0.027 0.053 0.065 0.029 0.013 0.016 0.033 No news exposure (dummy) 1.878** 0.248 0.565 1.667** 0.220 0.575 0.484 0.059 0.282 0.488 0.059 0.281 Exposure tone of economic news coverage 0.909** 0.220 0.307 0.864** 0.209 0.308 0.169** 0.108 0.054 0.161** 0.103 0.054 Media dependency 0.050 0.044 0.047 0.115* 0.096 0.047 Interaction exposure tone of news and media dependency 0.184* 0.074 0.094 0.053 0.034 0.053 Adjusted R 2 0.292 0.303# 0.193 0.199# F 38.76 32.14 28.31 23.08 N 825 825 724 724 *Po0.05; **Po0.01; ***Po0.001 (all one-tailed); # R 2 increase significant at Po0.005. Note: Cell entries are unstandardized and standardized OLS regression coefficients and Standard Errors. 368

Covering the crisis media effect holds for both periods of analysis, but is somewhat more pronounced in the first period. The more positive the news is, and the more one is exposed to such positive news, the more likely it is that expectations become more positive, and vice versa (H1). Additional analyses showed that weighting negative news more heavily than positive news (as shown in Table 1) yielded only marginally stronger effects and better model fit compared to a non-weighted news index, in which positive and negative news are assumed to have the same substantial effects (H2). 13 It appears that in an economic climate that is characterized by strong negativity, weighting negative news more heavily than positive news is largely inconsequential. We conclude that the inclusion of the main independent variable, tone of economic media coverage, has the predicted positive effect. Keeping all other variables constant, in the first period the news variable caused an increase of 0.91 on the 7-point sociotropic evaluation scale when moving from the most negative to the most positive news coverage to which respondents are exposed. Interpreting this effect, one does have to keep in mind that coverage was consistently negative throughout all the outlets, making variation in tone of the coverage someone is exposed to relatively small. 14 The dummy variable indicating no news exposure yields a noteworthy negative effect on expectations, but only in the first period. It means that those people who did not read or watched any of the outlets we analysed scored 1.878 points lower on their economic expectations compared to people who did read or watch any of the outlets. Furthermore, the lagged terms have strong impacts on the two dependent variables and they are, by far, the strongest predictors in the models. Finally, government approval is positively related to economic expectations in both periods. The more respondents approve of the government the more positive, on average, their economic expectations. Variables related to personal economic circumstances were not related to changes in sociotropic expectations. Neither did the frequency of political discussion show a significant relationship with the dependent variable. Models B and D in Table 1 test the assumption that respondents with higher media dependency were affected to a larger degree by economic media coverage than those with lower media dependency (H3). The effect of the tone of economic news is indeed more pronounced among those with high media dependency, but only in the first period. In the second period, we see a direct effect of news dependency. In the first period, a point increase on the media dependency variable results in a 0.184 increase in the effect of the tone of the consumed economic news. We now turn to explaining the variable that, on the aggregate level, has been largely constant over the period of study respondents personal economic expectations. We estimate models that do not and that do include media variables for both time periods. As shown in Table 2 (Models E and G), most 369

Boomgaarden et al Table 2: Explaining change in egocentric economic expectations Egocentric economic expectations at t2 (model 3) Egocentric economic expectations at t3 (model 4) Model E Model F Model G Model H B Beta SE B Beta SE B Beta SE B Beta SE Intercept 0.587* 0.242 0.698** 0.248 1.257*** 0.265 1.285*** 0.276 Egocentric economic expectations at t1 0.583*** 0.554 0.030 0.578*** 0.549 0.030 0.520*** 0.526 0.032 0.525*** 0.531 0.032 (model 3)/t2 (model 4) Age 0.001 0.021 0.003 0.001 0.017 0.003 0.006 0.086 0.003 0.005 0.080 0.003 Household income 0.012 0.033 0.011 0.013 0.035 0.011 0.035** 0.096 0.012 0.037** 0.099 0.012 Employed (dummy) 0.142 0.065 0.090 0.149 0.068 0.090 0.015 0.007 0.102 0.026 0.012 0.102 Education 0.027 0.042 0.019 0.020 0.031 0.020 0.001 0.002 0.021 0.011 0.016 0.022 Government evaluations 0.122*** 0.169 0.021 0.120*** 0.165 0.021 0.039 0.056 0.023 0.039 0.055 0.023 Frequency political discussion 0.000 0.000 0.021 0.034 0.047 0.023 No news exposure 0.370 0.052 0.197 0.277 0.038 0.224 Exposure tone of economic news coverage 0.357* 0.060 0.167 0.053 0.011 0.148 Adjusted R 2 0.410 0.413 0.337 0.338 F 116.12 97.66 73.79 61.42 N 825 825 724 724 *Po0.05; **Po0.01; ***Po0.001. Note: Cell entries are unstandardized and standardized OLS regression coefficients and Standard Errors. 370

Covering the crisis of the variation in economic expectations is explained by prior expectations, the lagged terms in the models. The comparatively high coefficients for the lagged term are in line with the relative stability of the dependent variable between panel waves (see Figure 2). Moreover, approving of the government leads to a more positive outlook, but only in the first period. The personal economic situation matters as well, but only in the second period. Having a high household income is positively related to better expectations regarding the personal economic situation in the second period. Turning to Models F and H in Table 2, exposure to economic media coverage does only weakly add to explaining changes in egocentric economic expectations (H4). 15 We find a significant effect of our news exposure variable for the first period, but the effect is clearly weaker than for sociotropic expectations. Neither consuming news nor the frequency of political discussion is related to changing personal economic expectations. A remaining concern in such studies of media effects is the issue of selfselective exposure to certain types of news contents that are in line with personal predispositions (Sears and Freedman, 1967). Translated to our study, this would mean that prior economic expectations should to some degree structure exposure to economic news, in that more positive expectations should lead people to expose themselves to more favourable news outlets. When estimating models that predict our exposure variable, controlling for age, gender, education and income, we find no significant effects of prior economic expectations for any of the two time periods. This leads us to conclude that selection bias is unlikely to contaminate our results in such a way that it would lead us to draw incorrect inferences with regard to our hypotheses (additional analyses not shown here, but are available from authors upon request). Conclusion What affects people s economic expectations in times of an economic crisis? We expected changes in expectations regarding the national economy to be affected by the tone of the media coverage of the economy. Our results demonstrated that the media have a considerable influence on people s expectations of the course of the national economy. The more negative the prospective evaluations of the economy in the news were to which a citizen was exposed, the more negative his or her sociotropic expectations became (supporting H1). Exposure to economic news also affected egocentric expectations; however, rather weakly and only in the first period, providing mixed evidence regarding H4. This also suggests that in times of crisis, citizens are able to discriminate between different types of economic expectations in a 371

Boomgaarden et al meaningful way, and that news coverage indeed, by and large, affects expectations regarding the national situation (Mutz, 1992). Only at the point when the economic crisis came to full swing in our first period was news coverage sufficiently pervading to also marginally impact personal concerns. The results showed a substantial negative effect of no news consumption. This low score of no news consumers may be due to the fact that the entire information climate regarding the economy was very salient and negative in the first period, which meant that negative information was picked up through sources other than news outlets. Furthermore, and interestingly, our results show that whereas personal economic circumstances did not affect changes in sociotropic expectations, household income was to some degree related to changes in egocentric expectations. Household income can be considered a proxy measure for information about peoples personal economic situation. The findings thus provide some suggestive evidence that citizens rely on different information sources (personal experience versus the mass media) when forming the different types of economic expectations. The egocentric expectations on the aggregate level did not significantly change during the time of our study. In the midst of the worst crisis in over half a century, people s outlook on their personal financial situations was, on average, unaffected. Most people still believed that their personal financial situation would not be endangered, even if they predicted that the national economy would go down the drain within the coming year. We do not only see such a pattern for the Dutch case, but also more generally. All over Europe it seems that assessments of personal and national economics have diverged (Eurobarometer 71). Perhaps, this is a result of a lack of experience with a severe economic crisis. Perhaps, this is a more general psychological phenomenon of people s disbelief that any major misfortune might happen to them. Perhaps, this is because the crisis takes time to eventually affect a mass of individuals. Especially the latter explanation might have extra mileage because economists agree that it takes time from national economic developments to translate into individual pocket book considerations. In any case, it seems that citizens generally do not link the abstract world of macroeconomic figures to their personal situation. This disconnection of ego- and sociotropic economic expectations allowed finding strong media effects on one, but hardly an effect on the other type of judgement. Information about the national economy is obtained from the mass media. The little variation in egocentric economic prospects that was there, however, was explained to a small degree by personal economics (see also Mutz, 1992). We furthermore believe that it is important to have a close match between economic media coverage and the type of economic assessment. We specifically analysed references to the future development of the national 372