The media component of left-right ideology
Examining the media agenda effect on macroideology
Author: Tamas Erkelens. 0542334
Thesis Supervisor: Catherine de Vries
Second Reader: Daniel Mügge
# of Words: (ex. Tables, Footnotes, Abstract & Appendices): 8650
Abstract
Aggregated trends on the left-right scale are used by many scholars and pundits to describe societal and political changes. Nevertheless, they often fail to acknowledge that the meaning of left and right changes constantly over time. This study comes up with an alternative explanation of macroideological trends, by introducing a media component of left-right ideology which links the media agenda to ideology at the micro-level. It is examined whether changes in the media component, drive mean and standard deviation changes of ideology at the macro-level. Furthermore, relations between macroideology, media issue salience and macropartisanship will be investigated, aiming to draw a more complete picture of aggregate dynamics. The ARFIMA and VAR approaches are used to create weekly and monthly models, using data from the Netherlands between 1993 and 2000. Although all models failed to yield reliable results, due to a high noise component in the series, important insights can be drawn. Firstly, theories of agenda-setting and ideology formation are synthesized; making it plausible that one’s ideology is driven by indicators outside of conventional voter and party relations, and offering possibilities to further investigate this on the micro-level. Secondly, fractional integration proved to be a valid technique to deconstruct short- and long-term parts of macroideology and media issue salience, providing an innovation in the study of media agenda-setting. Finally, this research recommends future scholars to differentiate between groups and to use quarterly or annual indicators to measure macroideology more precisely, and encourages scholars to explain its causes.
1. Introduction
A couple of years ago left-right divisions in the Netherlands were judged as outdated.1
Yet in 2011, politicians criticize ‘leftist hobbies’, speak of a ‘rightist government’, and ideologically coloured news programs appear on national television.2 Pundits and
scholars have mistaken these trends as a measure of societal conflict. Recently, for example, it was argued that the country has become more divided3. Similarly, when there
are, for example, more right-wing citizens, a country is arguably becoming more conservative. However, analysing aggregated trends, pundits and scholars have forgotten that left and right are politically constructed; their meanings change slowly but constant over time (van der Eijk, Binder & Schmitt 2005). Rather than more or less societal conflict or changed opinions, the terms left and right may signify different concepts. In this paper I will theorize how the media agenda influences the definition of left and right. It is expected that changes at the micro-level are translated to the aggregate level, driving macroideological change.
Conventional research has explained macroideological waves in conflict intensity or mean ideology through changes in party and voters positions (Eisinga, Franses & Ooms 1997; Knutsen 1997; Klingemann et al. 2006), patterns of government alternation (Fiorina 1988) or party elite strategies (Carmines and Stimson 1989). Supplementing this narrow political view, I postulate that media signal opinion change at it’s most early stage. Traditionally, left-right placement is divided in a social, partisan and a value component that explain one’s position on the left-right scale (Inglehart & Klingemann 1976, Freire 2008). This study introduces a fourth media component, connecting the media agenda to
left-right ideology. Specifically, I argue that attention for issues in media structures the ingredients by which citizens define themselves ideology.
Modelling the varying strength of issue saliency effects over time, I ultimately aim to measure the changing influence of the media agenda on macroideology. If media issue salience accounts for the variation in mean ideology or conflict intensity, evidence is presented that more individuals use this issue to structure their left-right position. A recent example of this mechanism is how the issue of immigration and integration changed the media component of left-right ideology, resulting in a more rightist electorate. Due to increased media coverage in 2001, immigration and integration became one of the most prominent citizens concerns at the 2002 Dutch elections (Boomgaarden & Vliegenthart 2007). A higher number of citizens associated their ideological position to this issue (Pellikaan, de Lange & van der Meer 2003; de Vries, Hakhverdian & Lancée 2011). Since the average Dutch population supported right-wing immigration policies, the average citizen shifted to the right (Pennings & Keman 2003).
Besides introducing a media component of left-right placement, this paper adds to the literature by estimating long- and short-term elements of macro variables. Using a fractional parameter, I measure how some issues are sensitive to events, while others are constantly discussed in media. To my knowledge, this study is the first to use these econometric insights to measure persistence of issues on the media agenda. Furthermore,
1 In a special issue of the magazine Vrij Nederland in 2003 called Left and Right, public opinion leaders
were hesitant to categorize themselves in terms of left or right (see also Bonfait 2010).
2 See for example, the article in de Volkskrant, 28-8-2010 “De jaren zeventig, maar dan andersom; De
terugkeer van links en rechts” (The seventies, but reversed; the comeback of left and right).
I will contribute to the ongoing debate regarding the multidimensional nature of political competition in Western Europe. Recently, scholars have argued that a multidimensional representation explains politics better than seeing the left-right scale as a one-dimensional ‘super issue’ (Pellikaan, de Lange & van der Meer 2007; Kriesi et al. 2008). In this paper, I argue that all issues are discursively related to left and right positions, assuming that macroideology is one-dimensional, but allowing the strength of alignment to vary per issue due to its multidimensional associations.
The meaning of left and right has become increasingly diffuse, complicating the ability of citizens to communicate its preferences to office seekers and holders (Mair 2008). Losing its capacity as “political esperanto” at the individual level, a macro understanding of ideology demonstrates how individual confusion results in one collective orientation. If the electorate as a whole – on aggregate attentive to politics – translates changing issue salience in media to changing left-right preferences, its concerns are represented in one macro indicator. Voters then may be stupid or confused at the individual level, but the ideological current forms an orderly sample of citizen’s concerns at the macro level. Aiming to explain the drivers of this current, this study revaluates the role of media. Media not only facilitate communication between public and mass, but also set the public and political agenda independently in a time frame that is increasingly subject to short-term dynamics of the issue agenda (Green-Pedersen 2007; Bélanger & Meguid 2008). Using data from the Netherlands between July 1992 and December 2000, I will investigate a time frame of ideological convergence and political stability (Aarts & Thomassen 2008). If relations are found in the ARFIMA model, it is likely that the media agenda explains macroideological trends elsewhere and at other times. Monthly and weekly measures are created by aggregating weekly NIPO data (N=760), offering a unique chance to model a finely tuned macroideological thermometer. Media issue salience is gauged through automated content analysis of six salient political issues in Dutch newspapers. Finally, VAR analysis is used to investigate dynamic relations between media issue salience, macroideology and macropartisanship, and to examine the validity of results of the ARFIMA model.
Although the empirical model fails to explain macroideological waves, important insights can be derived. The benefits of cross-validating empirical results are stressed, and fractional integration offers promising prospects to students of media dynamics. Also, future scholars are encouraged to explain macroideological indicators. Understanding how macroideology is driven, measures of conflict intensity or mean ideology can be used to measure public sentiment, both in academic research and in public debate. Below, I will proceed as follows. First, I will discuss the benefits of aggregation, and how this paper brings new insights to the study of macroideology. Then I will explain the causal mechanism in close detail, and derive a macro model from which the hypothesis will be tested. Before presenting the ARFIMA and VAR analyses, data and measures are introduced. Finally, I will discuss the non-findings and whether future research could tackle the hypotheses of this study.
2. Theory
2.1.1 Introduction to Left-Right theory
Since the French Revolution, the left-right division has remained relevant in political parlance, facilitating communication between masses and elites all over the world. In an
Economic Theory of voting (1957), Anthony Downs described how voters and parties use the
left-right scale as a tool to make sense of politics, designing a scale of policy preferences. A placement on this ‘ideological shortcut to the vote’ summarizes all issue preferences and is ultimately driven by one question: ‘how much government intervention in the economy should there be?’4 The Downsian model is opposed to the empirical tradition of the
American Voter (1960). Unable to structure ideological preferences coherently, Campbell
et. al. argue that the American voter is an easy target for manipulation and emotional judgments: “We have the portrait of an electorate almost wholly without detailed information about decision making in government.”5 The controversy about voter’s rationality has divided the
discipline until now. Scholars continue to demonstrate how ideology is based on issue preferences (Bélanger & Meguid 2009; Green Pedersen 2007), while others warn to interpret election results as policy mandates (Lewis-Beck et al. 2007).
2.1.2. Benefits of a macro-approach
Bridging the controversy of this eternal debate, students of macro dynamics see the electorate as a whole. Unlike in the American Voter, these scholars postulate that
macroideological indicators reflect more than the sum of its parts. And unlike the Downsian model, not every voter is portrayed as rational. Stimson argues that aggregation gets us closer to some sort of general will: “if just some people pay attention to politics and just some care a little bit about it – never mind all those who don't pay attention and don't care – the average of opinions will predominantly reflect those who pay attention and care. It is precisely this average we see when we see public opinion.”6 Macroideology then presents a fine-grained
measure of popular movements. Attentive citizens who update their ideological profile constantly are heavily represented in macro indicators (Erikson et al. 2002).
2.1.3. Explaining macroideological waves
Substantive research has aimed to explain how aggregated ideological change is driven, particularly in the United States. Capable of driving election results, many scholars have investigated the pendulum swinging from conservatism to liberalism. Such trends in mean ideology7 have been related to changes in economic indicators (Durr 1983),
government alternation patterns (Stimson 2004; Wlezien 1995), events (Carmines and Stimson 1989; Ellis 2011) or parties and politicians’ popularity (Abramson & Ostrom 1992). Furthermore, scholars have investigated conflict intensity, defined as the degree of popular dispersion between the poles of a distribution (Fiorina 1988), ranging from polarization to convergence (literature review, Hetherington 2009). Most European studies also focused on changes in party strategies (Adams, de Vries & Leiter 2011; Oosterwaal & Torenvlied 2010) or citizens’ preferences (Knutsen 1997; Noëlle Neumann 1998). In all these studies, media are regarded as communication platform between citizens and politicians. For example, Carmines and Wagner postulate that: “Party elites
4 Downs (1957), p. 116.
5 Campbell et al. (1960), p. 543
6 Stimson(2004), p.20.
7 Mean ideology is also known as public ideology (Ellis 2011) mass ideology (Nie & Andersen 1974) or
communicate their issue preferences to the mass public mainly through national media coverage.”8 Yet,
media also play an active role highlighting some issues and ignoring others (Kellstedt 2003).
Below, I will explain how the media agenda influences macroideology. First, I will explain the micro foundations of macroideological change. Then, I explain how changes in the media component of left-right placement at the individual level influence macroideological trends. Next, I will clarify my position in the multidimensional debate and discuss how macroideology and the media agenda relate to the partisan component of left-right placement.
2.2. Micro foundations of macroideology: Theorizing a media component 2.2.1. Introducing a media component of left-right placement
Traditionally, individual left-right placement is divided in a social, partisan and value component (Inglehart 1976; Freire 2006; 2008). The social component consists of structural variables such as class and religion that inform citizens’ left-right placement. The partisan component describes the relation between one’s partisan preference and left-right position. Finally, the value component encompasses the major value conflicts, and reflects issue preferences. Thus far, media have been regarded as facilitators of communication between citizens and parties (Knutsen 1995; Freire 2008). However, media also play an increasingly independent role, shaping public opinion by setting the agenda (Hall & Mancini 2004; Green-Pedersen & Stubager 2010). In this paper, a media component of left-right placement is introduced, which is defined as the connection between media issue salience and individual ideology. It is theorized how the media agenda structures ideology, by weighting the salience of issues that are mixed in the value component. This could explain trends in conflict intensity or mean direction of the average citizen, even if policy preferences have remained similar. Elements from agenda-setting and ideological studies will be synthesized in the next paragraphs.
2.2.2. The influence of the media agenda on the public agenda
Extensive research on the mediatization of politics finds that Western-European media have substantive power over citizens (Green Pedersen 2010). According to media malaise theories, media are following market logics (Mazzoleni and Schulz 1999), whereas mobilization theory argues that media expose crucial information to the public (Newton 2006). Since my aim is to expose a mechanism that explains macroideological change with the hotness of issues in public debate, I will bypass this debate, focusing on the effect of media on the public agenda. The media agenda is defined as a platform of public debate, whether this is driven by parties, market logics or public opinion. My contention is summarized as such: “the press may not be successful much of the time in telling people what to think, but it is stunningly successful in telling its readers what to think about."9
2.2.3. The effect of the media agenda on public opinion through the public agenda
Agenda-setting literature suggests that news media influences the public agenda (Mc Combs 1977; Miller & Krosnik 2000; Boomgaarden & de Vreese 2006). If particular issues are more pronounced in media, more individuals see politics through the lens of these issues. Media present a ‘pseudo environment’, which citizens use as a frame of reference to translate their personal experiences into political attitudes (Lippmann 1922;
8 Carmines & Wagner (2006), p.71. 9 Cohen (1963), p. 13.
Mutz 1994). Predicting the future public agenda, Mc Combs describes the media agenda as “the initial stage in the formation of public opinion”10 The literature on agenda-setting has
firmly established the influence of media issue salience on voting behavior and performance evaluations at the micro-level (see literature reviews of Kinder 1999; Scheufele 2001). Yet, ample research investigated media effects on ideology (Jost 2005). 2.2.4. Durability of Ideology on the micro-level
Seeing the left-right scale as an intermediary between policy preferences and the vote (Downs 1957), one could expect similar effects of media issue salience on ideology as on the vote. However, ideology is no shortcut to a binary choice (Klingemann 1979; Knutsen 1995). In political science literature, the durability of ideology is stressed, for example by authors who define ideology as a worldview (Hinich and Munger 1994) or as a social identity which is learned from childhood (Campbell et al. 1960; Green, Palmquist & Schickler 2002). Furthermore, recent psychological studies suggest that ideology reflects cognitive dispositions, such as anxiety that predicts more conservative preferences (literature review, see Jost 2009).
2.2.5. Short-term elements of Ideology on the micro-level: the media component Despite these durable elements, most authors stress that ideology is politically constructed and can therefore change (Sani and Sartori 1983; Popkin 1991; van der Eijk et al. 2005). According to this view, the value component is not only related to ‘major value conflicts’. The value component encompasses all political issues, ‘regardless of how many cleavage and/or identification dimensions exist.’11 In other words: positions on almost any
political issue can to some extent be referred to as being leftist or rightist. Left and right positions are defined at a specific time (Inglehart 1984; Fuchs and Klingemann 1990; Knutsen 1995) and place (Huber 1989; Inglehart 1990; Knutsen 1995), and are slowly but constant changing in meaning. How do media help constructing this changing meaning? 2.2.6. Synthesizing Agenda Setting and Ideology Theories: The media component Conventional research has explained changes in the meaning of ideology through changes in party and voters positions (Eisinga, Franses & Ooms 1997; Knutsen 1997; Klingemann et al. 2006). However, the meaning of ideology is also structured by the weight of issues in one’s ideology equation. Salient issues on the public agenda have a bigger impact on one’s ideology. (Budge & Farlie 1983; Knutsen 1995; Petrocik 1996; Lachat 2011). Iyengar (1990) has called this the accessibility bias, “the general tendency of individuals to attach greater weight to considerations that are, for whatever reason, momentarily prominent or salient.”12 This provides leeway to theorize an indirect media agenda setting
effect on individual ideology. After all, if the media agenda is correlated to the public agenda, and the latter structures one’s ideology, it is very plausible that the media agenda influences the weight of the ingredients of ideology too.
2.3. Macroideological consequences 2.3.1. What happens under aggregation?
Studying macroideological change, it is crucial to understand how individual change plays through at the macro-level. Using media as their political information outlets, attentive individuals update their ideology more often when the salience of issues in media changes (Mac Kuen 1983). Although this well-informed group is least vulnerable to
10 McCombs (2004) p.3.
media impact at the individual level (Converse et al. 1992; Zaller 1992), attentive individuals have the strongest influence on macroideology (Erikson et al. 2002). Attentive citizens transmit media messages to a less politically sophisticated public by discussing the news with their environment (Popkin 1991). And secondly, left-right positions of an uninformed public cancel each other out, when they are aggregated into macroideology. Therefore, it is expected that small changes in macroideology reflect meaningful shifts. This premise is not beyond dispute. Some scholars state that the opinions of the inattentive public are not filtered out under aggregation (Conover & Feldman 1981; Jacoby 1995). This is because mass ideology consists of two parts: Operational ideology, the sum of policy preferences, and symbolic ideology, the ideological label one identifies with. Especially in the United States these two parts are very different. Symbolic ideology is not expected to influence macroideology strongly in this study for three reasons: Firstly, left-right ideology is strongly related to policy positions in the Netherlands (Van der Eijk et al. 2005), and a weaker social identity than in the United States. Also, left-right placement is associated to a scale in multiparty systems, in which parties are distinguished in the degree of being left or right (Knutsen 1995). Finally, it is suggested that lower educated Dutch citizens display little value coherence (Achterberg &Houtman 2009). 2.3.2. The causal mechanism: how the media agenda drives macroideology
Summarizing, if the media agenda influences individual left-right placement, it may not only change the meaning of the left-right divide, but also drive macroideological patterns. Since policy preferences on political issues are not evenly distributed, we would expect the media component of left-right placement to influence conflict intensity and/or mean ideology. In graph 1, I present a hypothetical example of how this process is taking place.
Graph 1: Effect of Media Agenda on Macroideology through Public agenda and Ideology
Firstly, there is more attention for an issue (say that immigration is Issue A) in the news, and the weight of immigration in the media agenda (1) increases with Δ t>1. More
individuals then see immigration as an important problem: an increase of issue A on the Public Agenda (2). Citizens associate immigration more often to their ideologies, affecting individual left-right placement (3). The meaning left and right increasingly associated to immigration, particularly because immigration has a stronger weight in the value component, but also because parties who own the immigration issue have become more popular. Since especially attentive citizens, that have a stronger weight on the macro-level change their ideology accordingly, this influences the distribution of macroideology (4), affecting conflict intensity and mean ideology. Because the policy distribution of immigration is more skewed than of other issues in this example, a change in the media agenda (1) has a polarizing and rightist effect on macroideology (4). Macroideological change can thus be twofold: In direction, ranging from left to right, and
in conflict intensity, ranging from convergence to polarization.
2.3.3. Hypothesis
Studies on macroideology are typically based on large cross-sectional datasets, examining changes between one or more years (Inglehart & Klingemann 1984; Knutsen 1995; Stimson 2004; Knutsen & Kumlin 2005; Freire 2006). This study investigates macroideological trends in a shorter time span, weeks and months, aiming to construct a precise measure of macroideology. Expectations on lag length are thus not supported by previous research. Given the stability of ideology, especially on the macro-level, I expect that a delay takes place before interaction the media and public agenda results in macroideological change, but it is expected to be not more than a year:
(H1) Changes in media issue salience precede changes in macroideology, taking place
over multiple time lags.
By estimating effects of media issue salience on macroideological change, I aim to constructe a measure of the impact of the media agenda on macroideology. Looking at the variation of significant effects over time, one could then model the changing impact. Alternatively, if no effects would be found in the model, it indicates that macroideology is no approximation of a general will that responds orderly to changes in media issue salience. Alternative explanations will be discussed referring to the dynamics of the series in the analysis section.
2.4. The multidimensional meanings of left and right
Currently, scholarly debate is raging between proponents and opponents of a multiple dimensional representation of politics. Some scholars argue that we should not speak of one issue continuum, because there are different dimensions with distinct meanings of left and right. These authors typically find evidence in factor analysis for a strong one-dimensional socio-economic left-right scale (Inglehart & Klingemann 1976), and a less important and orthogonally aligning second dimension, whether this is a cultural (Kriesi et al. 2008; van der Brug & van Spanje 2009), a post-materialist (Inglehart & Klingemann 1984), religious (Aarts & Thomassen 2008), or authoritarian/legal (Kitschelt 2004). Granting that such cross-cutting cleavages exist for voters, I argue that a one dimensional representation of left and right still remains relevant. According to ‘persistence theory’ (Kitschelt & Hellemans 1990) the left-right scale includes all political issues, as is validated by Stimson’s common-sense test: “Can I say which side of the issue is liberal and which
side is conservative?”13 Substituting liberal and conservative with left and right, the test
underlines that issues from multiple dimensions can also be defined in terms of left and right. Nevertheless, some policy issues are stronger related to one’s left-right ideological position than others (Inglehart 1990; Knutsen 1995; Freire 2008), and if aggregated, to macroideology.
As in most other Western countries (Kitschelt 2004), socio-economic issues in the Netherlands load stronger on a one-dimensional left-right scale than issues from alternative dimensions (Eisinga, Franses & Ooms 1997; Aarts & Thomassen 2008). Socio-economic issues have been salient on the political agenda since the very beginning of the Dutch party system (Mari 2008). Socio-economic issues have been linked to the terms left and right for decades, and are described as thé most dominant value conflict in Dutch politics (van der Eijk & Niemöller 1992; Aarts & Thomassen 2008). Contrary, abortion has not structured ideology to a strong extent, and is only moderately explaining ideology at the micro-level (Green-Pedersen 2007a). Differences in ideological linkages should result in different effects per issue of media issue salience on macroideology. If the salience of an issue that is strongly related to the terms left and right, it is more likely that an individual changes his ideological placement:
(H2): Issues from the socio-economic dimension are expected to have a stronger effect on macroideological change than issues from alternative dimensions.
2.4. A complete picture of macroideology
Left-right ideology is shaped in the messy causal triangle between media, citizens and parties (Walgrave et al. 2009; Box-Steffensmeier, Darmofal, Farrell 2009; Green-Pedersen & Stubager 2010). In order to draw a complete picture of macroideological formation, the relations between media issue salience, macroideology and macropartisanship will be further specified, relaxing the causal assumptions of the previous two hypotheses.
2.4.1. Media issue salience and Macroideology
As I have argued above, changes in media issue salience should precede changes in macroideology, resulting in an assumption for hypothesis 1. This unidirectional causality is questioned by some authors (Soroka 2002; Uscinski 2009). Neumann & Fryling (1985), for example, found bidirectional effects between media issue salience and public opinion. They argue that media follow public concerns, and that their reports amplify these concerns, resulting in feedback loops. Also, in ‘bottom-up agenda-setting’, media follow public opinion (Kleinnijenhuis and de Ridder 1999). Uscinski (2011) found for example that media cover more rightist issues, if right-wing parties are more popular. Confirmation of endogenous relations or bottom-up agenda-setting, as is formalized below, would bring the validity of hypothesis 1 into question.
(H3) Media issue salience is affected by macroideology 2.4.2. Macroideology and Macropartisanship
Parties play a prominent role in the construction of ideology, as is demonstrated by the strong role of the partisan component of left-right placement (Inglehart & Klingemann 1979; Freire 2006, 2008). Stimson has explained how parties help to construct a consistent ideology: “We get used to the idea that certain positions are advocated together by the same people at the same times and places and come to believe that they must be logically tied.”14 Given the
13 Stimson (2004), p.70. 14 Stimson (2004), p.68
strong link between partisanship and ideology, it is expected that macroideology is more strongly related to macropartisanship than media issue salience. After all, the latter effect is indirect, mediated by the public agenda.
(H4a) Correlation H4 > Correlation of H3
Despite the prominence of ideology and partisanship in political science, few studies have investigated their relations at the macro-level (Box-Steffensmeier, Knight & Sigelman 1998). In the United States, it was concluded that macropartisanship, aggregated partisanship, and mean macroideology, were only related for sophisticated voters (Box-Steffensmeier & de Boef 2001). In other words, macro indicators contain preferences of so many citizens, that a systematic correlation between aggregate partisanship and left-right ideology is only immanent with a group of citizen’s that pays attention to politics. Contrary to the expectation of Stimson (2004) and Erikson et al. (2002), which is a premise of my study, Box-Steffensmeier et alumni demonstrated that the influence of this attentive group is not translated to ideology at the macro level. No correlation in the VAR model would falsify Stimson’s theory on aggregation dynamics. Contrary to the American evidence, Eisinga, Franses and Ooms (1997) argued that in the Netherlands aggregated party ideology and macroideological change are related, finding a stronger effect of ideology on partisanship than vice versa. Following a two-step vote choice model (LA chat 2011), they argued that voters first think of their ideology, and secondly select a party which comes most close to their ideal position. Eisinga, Frances and Ooms found that differences between left and right converged from 1978 until 1995, and that this resulted in smaller policy differences between parties. Although they computed ideological measures for each partisan group rather than investigating the electorate as a whole, I expect to draw similar conclusions as Eisinga et al., since I am investigating dynamics in the Netherlands.
(H4b) Macroideology and macropartisanship are interdependent processes, with the former being stronger than the latter.
2.4.3. Media issue salience and Macropartisanship
Thirdly, the relationship between macropartisanship and media issue salience is investigated, examining whether parties benefit from changes in media agenda. Issue ownership theory (Budge & Farlie 1983; Petrocik 1996) predicts that parties' popularity increases when media report more about an issue on which they hold a competitive advantage (Boomgaarden & Vliegenthart 2007; Kriesi et al. 2008.) Many studies have found such effects (Walgrave et al. 2009), so the hypothesis also checks whether the translation from media agenda to political opinions goes as I have explicated in the causal mechanism. No reciprocal effects are expected, although Uscinski (2011) found that macropartisanship predicts coverage of issues that are owned by parties. This is not plausible in the Netherlands, because media systems less ideologically colored than in the United States (Hall and Mancini 2004), and issue ownership dynamics vary more (Walgrave et al. 2009).
(H5) The popularity of issue owning parties increases when media salience of their issue increases.
3. Data & Operationalization
3.1. Measuring macroideology and macropartisanship in the NIPO dataset
In order to measure macroideological change, weekly data on left-right positioning between July 6th 1992 and December 23rd 2000 from the NIPO dataset is used15. The
NIPO dataset provides a unique opportunity to investigate weekly and monthly changes, but unfortunately the biggest transition in modern Dutch politics, the sudden entrance of the populist LPF in 2002, is not covered by the study. Instead, a time frame of political stability is investigated. From the start in September 1994, the two centrist Purple government coalitions, of VVD (Right-liberal) D66 (Centre-liberal) and PvdA (Social-democrats) had continuous popular support. Also, the two parliamentary elections in 1994 and 1998 showed little volatility (Aarts & Semetko 1999). Since few individuals are expected to change their ideology, the time frame offers a chance for a conservative test of my hypotheses. Every week a representative sample of an average 760 respondents was asked for their preference on a 7 point left-right scale16. The item asks for policy
direction gauging operational ideology rather than symbolic ideology (Ellis 2011). Next, mean and standard deviation of left-right placement were aggregated, creating time series of 441 weekly and 102 monthly scores. The mean captures the changing direction of macroideology, whereas the standard deviation measures conflict intensity, ranging from polarization to convergence. Secondly, macropartisanship was measured by asking respondents every week about their vote intention. Only the five biggest parties are included in this study, since other parties had too little observations to create a reliable weekly indicator. PvdA, VVD, CDA, D66 and GroenLinks reflected more than 90% of the preferences of voters in the Parliamentary elections of 1994 and 1998 (Aarts & Thomassen 2008) and present an almost complete political landscape.
3.1.1. The benefits of Time Series Analysis
Time series analysis offers unique possibilities to understand the functioning of democracy. Firstly, this type of analysis specifies time order, enabling to make strong causal claims. Secondly, time series analysis allows explaining general trends in public opinion (Wu 2002; Vliegenthart 2009). As Vliegenthart argues: “It might not help us to understand why a certain individual changes his or her political opinions, but it does give insight in why general public opinion sways in one direction or the other.”17Aggregation, furthermore, results in
the reduction of all noise (Inglehart 1985), as was described earlier in this paper. Despite the advantages, caution is also required. Robinson (1950) points at the ecological fallacy problem: Observed effects at the macro-level may not be valid at the micro-level. In this paper, it is emphasized that all expected effects take place at the macro-level, and I have only aimed to make plausible in the theoretical section that some individuals indeed change their ideology when media issue salience changes.
3.2. Measuring issue salience in media
Six prominent political issues in the Netherlands were selected to gauge issue salience: Unemployment, Social benefits, Law & Order, Immigration /Integration, Environment and Euthanasia/Abortion. All issues have been salient for some time18 and represent a
15 Quarterly data may be a suitable time frame too, but results in too little observations for TSA (30) 16 In the NIPO questionnaire the following question was asked: “You see 7 boxes here between the words left and
right. Could you indicate on this scale to what extent your personal political direction is placed? The numbers have no meaning? Which box should I note for your political direction?”
17 Vliegenthart (2009) p.5.
18 The issues were mentioned by at least 50 people as most important problem facing the Netherlands in
sub dimension of the left-right scale19 (Aarts & Thomassen 2008). Newspapers present a
snapshot of salient issues at a certain point in time and therefore are a valid indicator of change in the media agenda (Helbling & Tresch 2011).20 Acknowledging that issue
salience in the media is driven by parties, events, and editorial decisions, the sum of these drivers results in an independent indicator of media issue salience (Helbling & Tresch 2011). All articles between 6-7-1992 and 23-12-2000 were searched, since from this date more than one newspaper is available in Lexis Nexis. Parool, Algemeen Dagblad, Trouw and NRC Handelsblad are of different political affiliation, presenting a representative selection of newspapers.21 Automated content analysis with Lexis Nexis was used to
count the number of newspaper articles in which an issue was featured, using well specified search strings. Below, I will explain the criteria that were used to create the wording of the search strings (See Appendix A, for exact wordings): Insensitivity to trends, a potential impact on Dutch political debate and reliability.
3.2.2. Criteria for search strings
Firstly, in order to make my issue indicators less sensitive to events and seasonality, I aimed to measure several parts of one issue dimension. For example, euthanasia and abortion are guided by a Christian versus a liberal moral position (Aarts & Thomassen 2008). Although speaking of issues, I am thus actually referring to issue dimensions. Furthermore, I have checked whether one part of the search strings is not overrepresented compared to other issues that belong to the same dimension. This would be the case if one keyword, say abortion, would generate 70% of the variation in ethical issues. Finally, the search strings are resistant to trending words. For example,
gastarbeider and allochtoon, and WAZ and Wajong are words that signify the same concept at
a different time, and are therefore all included.
Secondly, the issues should have a potential impact on Dutch public debate. Assuming that the meaning of left and right is tied to a specific place (van der Eijk, Binder & Schmitt 2005), I aim to explain macroideological waves in the Netherlands. The target of my search strings is thus to gauge issues that potentially affect Dutch public opinion. Whether international news may influence the meaning of left and right depends on the issue at hand. For example, news about international immigration may alter the salience of immigration, but news about American economic growth would normally not influence the salience of economic issue preferences in the Netherlands. If included, such international news would systematically distort the measure of issue salience in the Netherlands. The search strings thus have different specifications and exclusionary terms, directed to measure issue salience in Dutch public debate.
Thirdly, the search string should be reliable, as measured by Holsti’s formula for intercoder reliability (Holsti 1969). The percent agreement of two coders, in this case the researcher and automated content analysis, is compared. For 30 randomly selected articles, I have coded whether the issue content could influence Dutch political debate. Usually, search strings are very broad (see for example Boomgaarden & Vliegenthart
19 Unemployment is referring to the economic dimension (Economic) Social benefits to the social
dimension (Social), law and order to the authoritarian dimension (Laworder), immigration and integration to the cultural dimension (Immigration) environmental issues to the postmodern dimension (Environment) and euthanasia and abortion to the religious dimension (Euthabortion). (Variable labels in parentheses).
20 Weighting the audience size of the newspaper is not relevant here, since the aim of this study is to
measure salience of issues in public debate in general, and not to differentiate between newspapers.
2007; Green-Pedersen 2007), assuming that all the errors are distributed randomly. In my opinion, such research underestimates the threats of systematic overrepresentation. This threat is even more pronounced here because of the limited number of articles per week, the lowest on average 10. Careful attention to reliability resulted in an average percent agreement of 0.92, with law and order as issue with the lowest score of 0.85 (See Appendix Table A1). This is considered well above the acceptable standards (Boomgaarden & Vliegenthart 2007).
3.2.3. Creating a standardized relational indicator of media salience
After measuring issue salience in newspapers, standardized relational scores were constructed. Since the total number of articles is highly dependent on the words in the search strings, an absolute measure is not reliable. First, the number of articles in which an issue is mentioned was counted, creating absolute scores per week, which were then standardized in index numbers. Say, for example that in week 1 there are 12 articles in which the environment is featured, this is 2 more than the average of 10, resulting in a standardized index number of 120. Assuming that all issues have – on average – an equal weight in public debate, I have divided this standardized score to the total of standardized scores, creating relational scores per week per issue. I also assumed that the issue capacity is similar for each week. Although these assumptions are crude, it is the only way to control for seasonality. This is crucial for newspapers that print fewer articles in the summer than in the spring and often publish extra sections at special occasions (Helbling & Tresch 2011). As a result, there may be more news on the environment in week 1 than on average (standardized =120), but when this goes for all issues (Sum of standardized issues >6*120), the relational score would still be lower than its average of 1/6. For example, a relational score of 0.12 would indicate that 12% of the issue capacity is taken by the environment. Secondly, using relational scores is the only way to control for multicollinearity and news about other issues. All statements in this paper will thus only be about how 6 issues influence changes in macroideology.
4. Models & Methods
The hypotheses will be examined using the ARFIMA and VAR approach. In the first descriptive model, I will explore the dynamics of the series. In order to model the impact of the media agenda on macroideology, a number of conditions should be met. These conditions will be examined in the subsequent models, of which the results will be described in the Analysis section.
4.1. Models
4.1.1. Examining Hypotheses 1 & 2 in Baseline Model
In model 2, the “baseline model”, Hypotheses 1 is tested: Changes in media issue salience affect macroideology over time, assuming that the population distribution of issues
varies. Unfortunately, no data is available that summarizes these distributions per month or even year. Even the DPES studies do not provide indicators for all 6 issues. However, from other studies we can learn that citizens’ preferences were different per issue (Kleijnneijenhuis 1998; Knutsen & Kumlin 2005; Aarts & Thomassen 2008; Adams, de Vries & Leitner 2011). Explaining conflict intensity and mean ideology at t (week 1≤t≥week 52), I will test first whether changes in the media agenda at t≥0. Furthermore,
it is expected that socio-economic issues have a stronger effect on macroideological changes than issues from alternative dimensions (H2).
4.1.2. Examining Hypotheses 3/5 using the VAR approach
In order to model the changing impact of the media agenda on macroideology, it is firstly required that at least three22 of the issues have an effect on changes in macroideology in
the ARFIMA model. Secondly, changes in media issue salience should precede changes in macroideology. Endogeneity will be examined by Granger causality tests in model 3, If these two requirements are met, the changing impact of media on macroideology can then be modeled. If the conditions fail, model 3 and the next models will be used to explain why the hypotheses are not met. In model 4, I will explore dynamics between partisan preferences and macroideology, testing hypothesis 4 using the VAR approach. If no effects are found in model 3 and 4, monthly macroideology is an unreliable indicator and its over-time dynamics as such are meaningless. Finally, an issue ownership VAR model is run to investigate to what extent parties benefit from changes in the media agenda. If no issue ownership effects are found, it is likely that the controls for seasonality, multicollinearity and confounding variables have resulted in false measurement of media issue salience.
4.2. Methods
Before examining the hypotheses, I will introduce the ARFIMA and VAR approach, and inspect the dynamics of the series.
4.2.1. Stationarity in the ARIMA approach
ARIMA modelling examines relationships over time, allowing to model varying effects of media issue salience on macroideology. In an ARIMA (n,d,q) model, dynamics of each variable are first carefully modelled, for example by including the right number of Autoregressive (n), and Moving Average (q) terms, so that the residuals are white noise. This is the stochastic process that drives a time series, and means that there is no explained variance in previous residuals, both in variance and in mean (McCleary et al. 1980).
4.2.2. Fractional integration in the ARFIMA model
As is visible in graphs 2a and 2b, volatility per month is high, which alarms that macroideology carries a large random component. This could complicate the analysis at later stages. Firstly, however, we should assess stationarity, “if its mean and all autocovariances (including the variance) are finite and unaffected by a change of time origin.”23
Graph 2a and 2b: Direction (Mean) and Conflict intensity (s.d.) of ideology (1-7 point scale)
22 Two significant effects would lead to a comparison of two issues, rather than describing the effect of a
representative sample of issues on left-right ideology. Three presents the minimum requirement.
Using Robinsons (1995) semi-parametric routine, I will estimate a long-term d parameter,
a measure for the persistence of shocks, ranging between 0 and 1: The higher d, the
longer a shock is influential at later lags (Box-Steffensmeier & Smith 1996; 1998). If d=1, it means that the series are a unit root, and shocks accumulate perfectly over time. If d=0, variation from week to week is random. Tests indicate that both macroideological series are long range dependent, and no unit root (table A1 & B1 Appendix). The long-term parameter (0<d>1) is used to transform the series, resulting in a loss of 12 observations.2425
Traditionally, the ADF test functions as a criterion to examine stationarity (Mc Cleary et al. 1980; Vliegenthart & Hollanders 2007). If I would have only used the ADF test, this would have resulted in wrongly accepting the assumption of stationarity, or in a false recommendation to difference, subtracting all observations from the previous, based on the assumption that the series is a unit root. Macroideology and media issue salience have long range elements that need to be modelled. According to Clarke, Lebo and Walker (2000) aggregation introduces fractional dynamics, because some individuals change ideology often and others keep the same ideology for decades. This also makes methodological sense: “The essence of Granger’s argument is that the aggregate series is generated by different micro-level autoregressive and moving average processes among individuals, thereby introducing fractional dynamics.”26
By fractional integration of media issue salience, an important innovation to the study of media dynamics is presented. The fractional parameter measures the volatility of issues, that is in line with theories on the media agenda (Hilgartner & Bosk 1988). Issue salience in media consists of a short-term component, sensitive to events, and a long-term component which includes news about for example law and order if the topic is not on the media agenda. In table B1 (Appendix), the estimates of d and the t-values for the test
of no long range dependency are described. Issues with no constant appearance in the news, such as abortion and euthanasia, have the lowest long-term parameter. 27 Future
studies could explore the fractional dynamics of media issue salience more carefully. 4.2.3. Solving Autocorrelation and Heteroskedasticity
After fractionally integrating all variables, univariate models of conflict intensity and mean ideology are run. Inspection of the (partial) autocorrelation functions (McCleary et al. 1980) showed that autocorrelation, when for example mean ideology in December is explained by mean ideology in November, was present. After including one or more lagged dependent variables, the problem was solved: The residuals of the monthly and weekly models were distributed randomly. Moving Average processes were absent, marking the stability of macroideology from abrupt shocks. After testing models with various autoregressive specifications, the model with the best goodness of fit was selected, using the Aikakke Information Criterion (AIC) (Burnham & Anderson 2004). Ljungbox Q post-estimation tests confirmed the univariate monthly models and the weekly directional model have White Noise Residuals. Models of the weekly conflict
24 Fractional integration uses 12 prior observations. Although valuable observations are lost, the goodness
of fit of the models improved significantly (AIC improvement >20).
25 Dickey Fuller tests indicated that none of the fractionally integrated series were unit root (p<0,001). 26 Clarke, Lebo & Walker (2000), p.33.
27 News about abortion and euthanasia has the weakest long-term component, with no long range
dependency rejected at the 0.05 level. No definite conclusions can be drawn for this since the salience indicators were not created to compare the long- and short-term components of issue salience in media.
intensity series could not be well specified,28 demonstrating that weekly conflict intensity
changes are random.
4.2.4. Monthly ARFIMA models
After specifying the univariate models appropriately, bivariate models were run, examining dynamic relations between media issue salience and macroideology. For each issue, 12 models29 were run separately, since the relational independent variables are
relational. The ARFIMA models with significant effects and White Noise residuals are described below30. Changes in macroideology are best described by monthly intervals,
because no effects were found in the weekly model of mean ideology31.
In the analysis section, the correlation between residuals of the univariate model and independent variables will be inspected at different times to analyze the dynamics. When the residuals in the cross-correlation functions have a consistent positive or negative correlation with variation in the independent variables from t=0 until the significant lag, the effects will be interpreted (McCleary et al. 1980). For example, if changes in media issue salience lead to a more rightist public four months later, it is unlikely if the months before showed a sign of a leftist trend. Also, the AIC will be used to assess whether variance in media issue salience reduces noise compared to the univariate model: A lower AIC: a negative Δ in the table.32Finally, it is stressed that when discussing issue saliency
effects, I actually refer to the attention for one issue in relation to 5 other issues. 4.3. Relaxing the causality assumption in VAR approach
After the baseline ARFIMA model is run, the Vector Autoregression (VAR) approach (Brandt & Williams 2007) will be used to examine relations between media issue salience, macroideology and macropartisanship and to cross-validate the results of the ARFIMA model. All macropartisan indicators had long memory patterns and were fractionally integrated (see Appendix Table B2). VAR does not need theoretical expectations on the direction of causality; after selecting the appropriate number of lags, the direction of causality will be examined using the Granger Causality Wald test. If conclusions of the ARFRIMA model are supported, a VAR model will be run. Then impulse response functions are computed, which are necessary to measure the size of the effect of fractionally integrated variables33. When effects of the ARFIMA model are confirmed in
the VAR model, the size of the effects is thus interpreted in the VAR model.
5. Analysis
5.1.1. Explaining conflict intensity with changes in the media agenda
28 McCleary et al. (1980) use a decreasing pattern in the residuals of the PACF and ACF as evidence for
proper specification. This condition was not met: Weekly variation looked random.
29 Since media agenda effects on macroideology should take place within a year, 12 models were run: One
model per lag.
30 For the univariate models, Ljungbox Q test indicates that autocorrelation (s.d.: P=0,09. mean P=0,17)and
heteroskedasticity (s.d.: P=0,999. mean: P=0,745) in the residuals is rejected (α=0,05). All bivariate models reported also had White Noise residuals.
31 This is due to the higher noise factor of weekly series, resulting in a more difficult estimation of effects. 32 Since a lower number of observations results in lower AIC’s (Burnham & Anderson 2004), models with
less than 90 observations were run to create standards, in order to measure change in AIC correctly.
The effect of media issue salience on conflict intensity was examined in ARFIMA models with one lagged dependent variable. Models with significant coefficients of the independent variables are described below:
Table 1: Monthly effects of Media issue salience on Conflict intensity
Coëfficiënt AR(1) Constant Δ AIC N Xcorr
Univariate FI s.d. 0.823*** (0.054) 0.285*** (0.017) (345) 0 90 Bivariate; FI issues Immigration t-3 0.234*** (0.061) 0.775*** (0.065) 0.267*** (0.014) +9 87 criss-cross Immigration t-4 -0.265*(0.073) 0.797*** (0.063) 0.288*** (0.016) +9 86 criss-cross Immigration t-5 0.199** (0.077) 0.732*** (0.068) 0.266*** (0.013) +8 85 criss-cross Immigration t-9 0.265*(0.121) 0.599***(0.086) 0.256***(0.011) +13 81 criss-cross Economy t=0 0.192** (0.063) 0.822***(0.058) 0.278***(0.017) +5 90 Consistent Economy t-10 0.222** (0.084) 0.799***(0.057) 0.247***(0.019) +5 90 Consistent Social t-5 0.160*(0.078) 0.757***(0.064) 0.275***(0.013) +8 85 criss-cross Euthabortion t-5 -0.187***(0.051) 0.795***(0.057) 0.286***(0.014) +15 85 criss-cross Euthabortion t-6 0.134*(0.060) 0.706***(0.068) 0.267***(0.011) +11 84 criss-cross Laworder t-10 -0.165*(0.081) 0.615*** (0.091) 0.274***(0.009) +9 80 criss-cross Laworder t-11 0.153*(0.076) 0.623***(0.085) 0.259***(0.010) +8 79 criss-cross
Standard Errors in parentheses. P that coefficient =0: *p < 0.05, **p < 0.01, ***p < 0.001
As is visible in the cross-correlation graph of model 1 in the Appendix (Graph B2), only the salience of economic issues is consistently correlated with the residuals of the univariate model. News about the economy has an immediate and a long-term significant polarizing effect on conflict intensity. Unfortunately, the rest of the issues do hot have consistent correlation functions. Caution is required: The goodness of fit measured by the AIC did not decrease compared to the univariate model. Besides helping to interpret the effect, validation in the VAR models is needed to infer from these results.
5.1.2. Explaining Directional effect of media issue salience
In the next model, with a second order autoregressive specification, the effect of media issue salience on mean ideology is examined.
Table 2: Monthly effect of media issue salience on mean ideology
Coefficient X AR(2) Constant Δ AIC N Xcorr
Univariate FI mean 0.917*** (0.038) 0.316*** (0.043) (272) 90
Bivariate; FI issues
Immigration t-1 0.279** (0.102) 0.905*** (0.040) 0.300*** (0.040) +2 89 Consistent
Immigration t-2 0.302* (0.150) 0.889*** (0.046) 0.288*** (0.032) +5 88 Consistent
Environment t=0 -0.215* (0.098) 0.920*** (0.038) 0.353*** (0.048) +2 90 criss cross
Environment t-7 0.272* (0.123) 0.790*** (0.074) 0.266*** (0.020) +11 83 criss cross
Environment t-9 -0.275* (0.109) 0.789*** (0.079) 0.274*** (0.020) +12 81 criss cross
Laworder t-7 -0.370*** (0.084) 0.816*** (0.067) 0.289*** (0.021) +17 83 criss cross
Laworder t-9 0.299** (0.100) 0.789*** (0.075) 0.255*** (0.020) +14 81 criss cross
Economy t-4 0.293** (0.112) 0.840*** (0.053) 0.276*** (0.024) +8 86 criss cross
Economy t-6 -0.220* (0.106) 0.829*** (0.065) 0.284*** (0.024) +9 84 criss cross
Economy t-8 0.273* (0.112) 0.753*** (0.084) 0.259*** (0.019) +12 82 criss cross
Unfortunately, a likely pattern in the cross-correlation function (see Appendix, graph B3) was only present for the issue of immigration and integration, with high peaks at the first two lags and randomly distributed correlations at later lags. As we see in table 2, an increase in news about immigration and integration results in more rightist ideological preferences the next month. This confirms that the average citizen has a more right-wing position on issues of immigration and integration than on other issues (van der Brug & van Spanje 2009). However, again the goodness of fit has not decreased.
5.1.3. Lag length, multidimensionality and media component
Despite the instability of the data, provisional conclusions are drawn. It was expected that some time would pass before issues are so ingrained in the average citizens’ mind that they influence macroideology. News about immigration and integration had an over-time effect on mean macroideology (lag 1 and 2). However, news about the economy resulted in macroideological polarization within the same month and over time (lag 0 and lag 10). The assumption of over time effects was thus accepted in three of the four cases. According to hypothesis 2, attention for socio-economic left-right issues has a stronger effect on macroideological change than alternative issues. In the conflict intensity model, news about the economy was the only significant and consistent effect. Yet, in the second model, an issue belonging to an alternative dimension than the socio-economic left-right dimension turned out to have a significant effect on mean macroideology. Issues from an alternative dimension do not have a stronger impact on macroideology than issues from the socio-economic dimension. .
In the empirical analysis, weak evidence for an effect of the media component of left-right orientation was found. The changing changing impact of the media agenda on macroideology cannot be modelled, since the precondition of 3 significant issues was not met. Furthermore, the AIC showed no significant reductions in goodness of fit after including an independent variable. Validation in the VAR analysis is needed to infer from these results, and to explain why the indicators are unreliable.
5.2. VAR analysis: Testing ARFIMA validity and relation with macropartisanship 5.2.1. Macroideology and media issue salience
Using the VAR approach, the assumption of hypothesis 1 and the ARFIMA model, that changes in media issue salience affect macroideology, was examined. Although the analysis confirms this direction of causality in most cases, the analysis is not reliable, since no consistent effect from the ARFIMA model was validated in the VAR analysis. Firstly, a different number of lags were often recommended. Secondly, the direction of causality of the consistent effects in the ARFIMA model was rejected in most cases, as is visible in the table below, which reports only the significant results of the Granger Test for causal relationships. The only consistent effect that passed this test, the rightist effect of immigration and integration, had, contrary to the ARIMA model, a leftist effect on mean ideology in the VAR model. Furthermore, contrary effects of the VAR regression analysis and the impulse response function provide further evidence for weakness of the data.
Concluding, the VAR analysis showed fundamentally different results than the ARFIMA model, rendering the conclusions of both models useless. Unfortunately, I can thus not
reliably assess hypothesis 3.34 The models confirm the descriptive observation in graphs
1a and 1b that all macro indicators of this study have a large noise component. Conflict intensity and mean ideology thus differ too little per month systematically, to be able to explain the indirect effect of media issue salience. Furthermore, there may be problems with the reliability of the media issue salience is also questionable, if we see its high volatility (Appendix; Graph B1). Given the doubts on the validity and reliability of the indicator, the next models will not only be used to examine the next hypotheses, but also to investigate why the series of media issue salience and macroideology display so little correlation. A comparison with effects on macropartisanship is made in order to discuss the weakness of the indicators.
Table 3: Granger Wald Test for Causality (H0 = No causal Relationship)
X Y Lags P>X2 H1/H3 Confirms ARFIMA effect?
Mean Immigration Mean 4 0.023 H1 Causality confirmed, but different direction Environment Mean 10 0.018 H1 Different causality than ARFIMA model Mean Environment 10 0.000 H3 Different causality than ARFIMA model Economy Mean 5 0.042 H1 Different causality than ARFIMA model Conflict
intensity Laworder s.d. 1 0.023 H1 Different causality than ARFIMA model Economy s.d. 12 0.000 H1 Different causality than ARFIMA model
5.2.2. Macroideology and macropartisanship
Secondly, it was examined whether macroideology is a stronger predictor of macropartisanship than vice versa (H4b). As in the American study of Box Steffensmeier et al. (2001), no consistent significant effects were found35 in the VAR analysis. The
findings of Eisinga et al., which were based on left-right estimates for partisans on the group level, can thus not be transferred to the macro-level. VAR analysis makes clear that there is no strong relationship over time between macropartisanship and macroideology. Furthermore, the counterfactual, that the partisan component and the value component of left-right orientations are so related that changes are taking place within the same month, is also rejected. As is visible in table 4, a longer lag length than 0 is recommended in all cases.
We can conclude that explaining differences in macroideology as a characteristic of the electorate as a whole (Erikson et al. 2002) is problematic using the short time spans in this study. If macropartisanship is not even related to macroideology, an effect that was expected to be much stronger than the effect of the media agenda, the monthly measure of macroideology in this study is inaccurate. Unfortunately, the monthly macro-indicator of ideology reflects a meaningless average since its variance cannot be explained.
Table 4: Advised lag length (AIC criterion) in VAR model of macroideology and macropartisanship
CDA VVD PvdA D66 GL CDA VVD PvdA D66 GL
mean 3 3 4 4 3 s.d. 3 3 8 3 3
34 In table 3 the results of the VAR analysis are nonetheless described. If the analysis would have been
reliable, I would have concluded that media have a stronger effect on macroideology than vice versa, although feedback effects are present for the environmental issue.
35 Two inconsistent significant effects were found (VVD: positive lag 1 and 4, negative lag 2 & PvdA,
5.2.3. Macropartisanship and Media issue salience
An issue ownership model was run to investigate whether partisan popularity is guided by changes in media issue salience. No consistent significant effects were found, juxtaposed to most findings in the field (Walgrave et al. 2009), and a similar study in the same era (Kleijnnijenhuis 1998). The non-result may be due to the failure of aggregation to obtain reliable measures. Yet, macropartisanship is much less volatile (See Graph B4; Appendix) than media issue salience and macroideology. The media agenda does not influence party popularity. A plausible alternative explanation is that the different results of this study are due to different measurement of the media agenda. In this study it was assumed that the media agenda has a fixed issue capacity, andthat attention for one of the five issues relational to 5 other issues. Future scholars are advised to control for seasonality and multicollinearity in other ways. For example by including seasonal components in the time series framework, modelling each issue separately, or including more issues on the media agenda.
5.3. Conclusion
From the results, we can conclude the influence of media issue salience is too minor to explain political preferences at the macro-level. If average public opinion is so stable that on aggregate level even partisanship and ideology are not related, it is not surprising that differences in macroideology in months cannot be explained. Stimson’s view on aggregation, that more attentive citizens should influence macroideological indicators significantly, is rejected for these monthly indicators. Monthly macroideological shifts do not reflect collective meaningful shifts. Future scholars are advised to use more precise indicators of aggregate opinion. Differentiating groups by sophistication level may help how particular groups in society are influenced by changing salience of issues in media. Highly educated construct their ideologies differently (Converse 1964; Box-Steffensmeier & Smith 2001). Unfortunately, the NIPO dataset only allows differentiating between income and religious groups36. Furthermore, the failure to find effects with the media
issue salience scores may be due to false operationalization of the media agenda.
The rejection of all hypotheses, stresses that we cannot see macroideological change as a current that drives policies and elections results in the Netherlands. This may also be due to the diverse party systems in the Netherlands, and the presence of more cross-cutting party loyalties and issue preferences than in the United States. Individual variation produces a differentiated average. Perhaps, cross-sectional studies are the best available datasets to investigate ideological change over time, since the relations between all these influential dimensions can be taken into account. Changes in the short-term component are too little to influence macroideology dramatically. The durable ideological components also form a check on ideological change. A quarterly indicator of macroideology and macropartisanship can probably reduce the noise level to an acceptable standard for datasets with bigger time frames. Also, if data are available, tests on an individual- or group level could be conducted.
Concluding, considering the weaknesses of the data, we cannot reject that the media agenda influences macroideology. Future scholars are advised to investigate the influence of changing impact of the media agenda on macroideology with a more precise focus, than seeing the electorate as a whole, and a bigger time span.
