More Than Weighting Cognitive Importance: A Dual‐Process Model of Issue Framing Effects

Abstract

Issue frames in policy discourse and news reporting regularly influence citizens' political opinions. Yet, we only have a limited understanding of how and among whom these framing effects occur. I propose a dual‐process model of issue framing effects arguing that we must understand mediators of framing (the how question) in connection with individual‐level moderators of framing (the whom question). Experimental results show that issue framing affects opinion through different psychological processes depending on who the receiver of the frame is. Among the moderately politically aware or those having weak political values, framing effects were mediated through processes of changing importance of considerations as well as changing content of considerations. Among the highly aware, only the importance change process mediated framing effects, while there were no framing effects among those least aware or those having strong values.

How and when are political messages from policy makers and mass media able to change opinions among citizens? News stories as well as policy debates present and interpret issues and events by emphasizing certain aspects of the issue and downplaying the role of other aspects in order to guide the audiences to “the essence of the issue” (Gamson & Modigliani, 1987, p. 143). In numerous studies, this process, known as issue framing, has shown its ability to influence distributions of opinion, at least if the messages are one‐sided with only a single dominant frame. Based on the empirical evidence from the last decade and a half, Sniderman and Theriault (2004) summarize that by now it is “widely agreed that citizens in large numbers can be readily blown from one side of an issue to the very opposite depending on how the issue is specifically framed” (pp. 133–134; see Chong & Druckman, 2007a; Kinder, 2003, pp. 358–361; de Vreese, 2003, pp. 35–41).

However, we know far less about how issue frames are able to affect opinion—i.e., what mediates issue framing effects?—and about who the more susceptible to such an influence are—i.e., what individual‐level factors moderate issue framing effects? Pointing out some of the gaps in the literature on media effects, Kinder (2003) notes, “we need clarification of the psychological mechanisms that mediate mass communication influence” (p. 378) and, furthermore, although we know that people differ quite a lot in their level of political awareness, “how this difference is implicated in citizen's vulnerability to media influence is unclear” (p. 377).

In this paper, I address both of these questions. By linking the question of mediators with the question of individual‐level moderators, I suggest that a more adequate understanding of what mediates issue framing effects needs to take account of the varying levels of political awareness and differences in strength of values among the receivers of the frame (i.e., among citizens). Even though the whole history of public opinion research has taught us that the great variation in people's conceptualization and awareness of politics is highly consequential for their way of thinking about politics (e.g., Basinger & Lavine, 2005; Converse, 1964; Sniderman, Brody, & Tetlock, 1991; Zaller, 1992), this insight has only had limited impact on the study of mediators of issue framing effects.

Accordingly, in this paper I propose a dual‐process model of issue framing effects to specify through which psychological processes and among whom issue frames can be expected to change opinion. Specifically, I contend that issue framing effects on opinion are mediated not only through a process of changing cognitive importance of a receiver's issue‐relevant considerations, as is argued in most recent work on framing effects (e.g., Druckman & Nelson, 2003; Nelson, Clawson, & Oxley, 1997; Nelson & Oxley, 1999), but also through a process of altering the content of such issue‐relevant considerations (see Zaller, 1992). Furthermore, the relative impact of the importance change and content change mediators, respectively, is expected to be moderated by the receiver's level of political awareness and strength of political values. To test this dual‐process model an experiment was conducted to illuminate the mediators and moderators of issue framing effects. Results are consistent with both the claim of dual mediating processes and the suggested moderators.

Thus, the present study expands our knowledge of the nature of issue framing effects in two ways. First, framing effects on opinion can be mediated through at least two different processes, namely changing importance and changing content of underlying considerations. Furthermore, the impact of individual‐level factors such as political awareness and values is not limited to moderating the magnitude of framing effects; they also moderate the actual processes through which framing effects occur.

How and When Does Issue Framing Affect Opinion?

Issue Framing and Framing Effects Defined

Framing is used in a variety of ways across several disciplines (e.g., Reese, Gandy, & Grant, 2001) and, as noted by Chong & Druckman (2007a), “There is substantial conceptual confusion about types of framing effects, and the relationship between framing and related concepts” (p. 114).

In political science, two different uses of the term have developed, known as “equivalency frames” and “issue frames” (Druckman, 2004, p. 672; see Sniderman and Theriault, 2004, pp. 135–136). “Equivalency frames” present a problem or an issue in alternative ways by using “different, but logically equivalent, words or phrases” (Druckman, 2001a, p. 228). An example of this is how Kahneman and Tversky (1984, p. 343) frame the effectiveness of a program to combat “an unusual Asian disease” in terms of 200 out of 600 people who “will be saved,” or in terms of 400 out of 600 people who “will die.”

Although equivalency framing effects are enlightening to the study of, for example, individual‐level decision making or rationality (for a recent application in the realm of politics, see Druckman, 2004), this kind of framing is clearly not the most widespread in the policy discourse and political news reaching most citizens. As Sniderman and Theriault (2004), among others, point out, “It is difficult to satisfy this requirement of interchangeability of alternatives outside a narrow range of choices” (p. 135). Certainly, in the everyday of politics, competing elites and news reporters would probably not limit themselves to presenting exactly the same information in logically equivalent ways, if such a variation in interpretation without supplying any new bit of information or symbol was possible at all.

Instead, political actors engage in “issue framing,” struggling to give their interpretation of the issue. In contrast to equivalency frames, alternative issue frames are not limited to presenting exactly the same information but focus on a particular aspect of a policy problem. Because most political issues are complex and multifaceted, they are open for, and often dependent on, communicators to “suggest what the controversy is about, the essence of the issue” (Gamson & Modigliani, 1987, p. 143). Issue framing, then, “is the process by which a communication source . . . defines and constructs a political issue or public controversy” (Nelson, Clawson, & Oxley, 1997, p. 567) by “emphasizing a subset of potentially relevant considerations” (Druckman, 2004, p. 672). Thus, issue frames are more than a single argument or a position on an issue, because frames provide meaning to an issue and suggest how to understand and think about it (Nelson & Kinder, 1996, p. 1057).

However, this does not mean that issue frames do not carry evaluative content. On the contrary, “A frame generally implies a policy direction or implicit answer to what should be done about the issue” (Gamson & Modigliani, 1987, p. 143). Entman further specifies that issue frames perform some or all of the following functions: “defining effects or conditions as problematic,”“identifying causes,”“conveying a moral judgment,” and “endorsing remedies or improvements” (2004, p. 5). Hence, issue frames can potentially affect the outcome of an issue. The major political interest in issue frames stems from the observation that issue frames can “lead a double life” (Kinder & Sanders, 1996, p. 164): they exist as such frames in communication just described and as frames in thought, that is, “cognitive structures that help citizens make sense” of the issue (Chong & Druckman, 2007, pp. 105–106; Kinder & Sanders, 1996, p. 175).

Of course, frames in thought can influence one's opinion. Consistent with prior work on framing (Callaghan, 2005, p. 181; Jacoby, 2000, p. 751; Sniderman & Theriault, 2004, p. 136; also see Petty & Wegener, 1998, p. 324), my understanding of an issue framing effect is that it occurs if a frame in communication is able to modify a receiver's opinion from one position to another. Some scholars suggest reserving the term “framing effect” to opinion change generated through a specific psychological process (e.g., Nelson & Oxley, 1999). In contrast, I argue that opinion can be influenced by a frame through distinct psychological mechanisms, and hence I prefer to treat such processes as part of the empirical analysis, not as a definitional matter.

What Mediates Issue Framing Effects?

But how is this phenomenon called issue framing able to affect citizens' political opinions? It is a question about mediators of framing effects. A mediator is a variable that “represents the generative mechanism through which the focal independent variable is able to influence the dependent variable of interest” (Baron & Kenny, 1986, p. 1173). In a recent contribution, two leading framing researchers acknowledge that they “have been in two minds about the likely psychological mechanisms underlying framing effects” (Kinder & Nelson, 2005, p. 117). Instead, they arrive at the idea that “The precise effect doubtless varies from person to person, depending on the amount of attention given to the issue” (p. 118). However, this idea has not yet been subject to further elaboration or empirical evaluation. In this section, I develop a dual‐process model to explain how issue framing effects vary from person to person and which individual‐level characteristics moderate the nature of framing effects.

In the literature, three possible mediators of issue framing effects may be identified (cf. Nelson, Clawson, & Oxley, 1997). The first mediator has been called the accessibility model because it hypothesizes that, “As particular frames rise to prominence, some opinion ingredients [i.e., considerations] are highlighted and made more accessible while others are shunted to the side. In the process, elite frames transform the meaning of public opinion” (Kinder & Sanders, 1996, p. 174, emphasis added; also see Cappella & Jamieson, 1997, Chap. 4; Iyengar, 1991, pp. 130–136; Sniderman, Brody, & Tetlock, 1991, p. 52; Zaller, 1992, pp. 80–85). Thus, according to this model, issue frames change opinion not by changing the content of a person's considerations, but through a process of making a person's pre‐existing considerations more accessible so that these considerations will be more likely to feed into the opinion formation process (Zaller, 1992, pp. 48, 119).¹1 A consideration is simply defined as any reason to favor or oppose an issue (Zaller, 1992, p. 40).
However, the accessibility model has more often been suggested than tested (Nelson, Clawson, & Oxley, 1997, p. 578), and those who have actually tested accessibility as a mediator of framing effects have not been able to support the model (Nelson, Clawson, & Oxley, 1997; Nelson & Willey, 2001, pp. 261–263; de Vreese, 2004).²2 Moreover, researchers of priming have also questioned accessibility as a mediator of priming effects (Miller & Krosnick, 2000; but see Valentino, Hutchings, & White, 2002). However, in a recent contribution, Chong and Druckman (2007b, pp. 639–640) revive the idea that, among some receivers, framing effects are mediated by mere accessibility, though they do not directly test accessibility as a mediator of framing effects.
Hence, by now, the accessibility model empirically considered does not seem to be a very likely mediator of framing effects.

This brings us to the next mediating model: the importance change model. The proponents of this mediator of framing effects maintain that issue frames work “by stressing specific values, facts, or other considerations, endowing them with greater apparent relevance to the issue than they might appear under an alternative frame.” By these means, issue frames “affect opinion simply by making certain considerations seem more important than others; these considerations, in turn, carry greater weight for the final attitude” (Nelson, Clawson, & Oxley, 1997, p. 569; see also Chong & Druckman, 2007; Price & Tewksbury, 1997). Thus, while issue frames in the accessibility model work by making preexisting considerations accessible in memory, issue frames in the importance change model influence opinion by affecting the perceived relative importance of different already accessible considerations.

Nelson et al. and others have found support for the importance change model across a variety of studies and issues such as civil liberties, welfare policy, adoption reform, and doctor‐assisted suicide (Druckman, 2001b; Druckman & Nelson, 2003; Nelson, 2004; Nelson & Oxley, 1999; Nelson, Clawson, & Oxley, 1997; Nelson, Oxley, & Clawson, 1997). Therefore, as the importance change model seems both theoretically plausible and empirically supported, it is reasonable to believe that issue framing indeed does affect opinions through a process of changing importance of considerations.

However, there is also reason to suppose that importance change is not the only process mediating issue framing effects. Instead of changing the importance of a citizen's already existing considerations, an issue frame might be able to put forward some new arguments or information that the citizen had not previously thought about and, thereby, deliver a new consideration—a reason to favor or oppose the issue. In this case, the issue frame is able to change a citizen's opinion by changing the content of his or her underlying considerations that go into the opinion. Thus, this process can be considered a third mediator of issue framing effects: the content change model (see Callaghan & Schnell, 2005, p. 13; Chong & Druckman, 2007a, pp. 115–116).

The mediating process of the content change model is well known from traditional persuasion or belief‐change models and is prominently represented in political science by Zaller's receive‐accept‐sample (RAS) model of opinion formation (Zaller, 1992). In this model, new information from an issue frame can affect citizens' opinions through “a two‐step process involving reception of [the] new ideas and acceptance of some as new considerations, thereby altering the balance of considerations in people's minds” (Zaller, 1992, p. 119, italics in original). Thus, if an issue frame contains some new arguments accepted by the receiver, the issue frame is able to influence opinion not by making some preexisting considerations seem more important but by changing the content of a person's considerations. This content change can occur either by adding new considerations or changing existing considerations.

However, as a mediator of issue framing effects, the content change model partly contradicts the way the proponents of the importance change model perceive the very nature of issue framing (e.g., Nelson, Clawson, & Oxley, 1997). Although these researchers do not rule out content change as an effect of issue framing (e.g., Nelson & Oxley, 1999), they see the mediating process of importance change as distinctive and defining for issue framing effects compared to other communication effects. As Nelson and Willey argue,

Indeed, frames that act this way [by changing content of considerations] stretch the boundaries of what we should consider true ‘framing effects.’ If frames actually supply new information that alters our beliefs and expectations about policy change, then how are they any different from conventional persuasive messages . . . ? Framing as we understand the term is best reserved for messages that don't supply new information, but rather restructure or reinterpret existing information. (2001, p. 256)

They simply say that if an issue framing effect is mediated by the process of content change and not the process of importance change, it is not really an issue framing effect.

However, it is a problem that Nelson and colleagues do not separate issue framing as an independent variable from its possible effects on the dependent variable: opinion change and the mediators of this change. Citizens are different and are probably influenced by issue frames in different ways (cf. Kinder & Nelson, 2005). If a news story interprets, say, a Ku Klux Klan rally as a threat to public order, some receivers of this issue frame would probably think that their consideration about public order threats at rallies in this case is more relevant and, in turn, deserves a greater weight in the formation of their final opinion on the Ku Klux Klan rally. Hence, in this situation the framing effect is mediated by the process of changing the importance of considerations. At the same time, though, other receivers of the very same news story—who have never in any way thought of public rallies as a threat to public order—might accept the issue frame as a reason to consider that threat. Hence, in this situation the issue frame could change opinion through the process of changing the content of considerations. But does the news story not in both situations deliver an issue frame (in fact, the same issue frame) about the rally? It does, because it is the very same news story. But the processes by which the issue frame is able to influence opinion among the receivers are different, simply because the receivers of the issue frame are different. This argument is consistent with Entman (1993, p. 53), who points out that the effect of a frame depends on the receiver's cognitive schemata (see, e.g., Druckman & Nelson, 2003).

To sum up the discussion, the argument is that issue framing must be considered an independent variable and that this independent variable can have different effects, depending on its receivers. To accommodate this argument, I here suggest a dual‐process model of issue framing effects. According to this model, an issue frame is able to influence citizens' opinions through two different processes: either by changing the relative importance of considerations or by changing the content of considerations. In addition, there is no reason to believe that both processes could not be working at the same time; a citizen's opinion change could then be a result of both new considerations and changing importance or weight of the considerations. This basic expectation of the dual‐process model is illustrated in Figure 1 and can be stated as follows:

Figure 1

A Dual‐Process Model of Issue Framing Effects.

H1: Issue framing effects on opinion can be mediated both through the process of importance change and through the process of content change.

But what regulates whether a mediator is present or not? To answer this question I now turn to the role of moderators.

What Moderates the Mediators?

If the issue frames contained in news stories can change opinions among citizens through two different psychological processes, what exactly is it that determines which process is at stake among whom? This is a question about moderators of the mediators. Here, two different moderators are suggested: the individual's level of political awareness and strength of political values.³3 Also, framing effects can be moderated by such contextual factors as source (Druckman, 2001b), competing frames (Chong & Druckman, 2007b; Druckman, 2004; Sniderman & Theriault, 2004), deliberation (Druckman & Nelson, 2003), and individual‐level “need to evaluate” (Druckman & Nelson, 2003) or “need for cognition” (Hartman & Weber, 2006).

Political awareness can be defined as a person's “intellectual or cognitive engagement with public affairs” and expresses “the extent to which an individual pays attention to politics and understands what he or she has encountered” (Zaller, 1992, p. 21, italics in original). Almost by definition, highly politically aware people have thought more about political issues in general and, thus, have more issue relevant considerations in their possession (Zaller, 1992, pp. 121, 167–169; Zaller & Feldman, 1992, p. 591). In addition, the more politically aware citizens are more likely both to be exposed to issue frames and to be able to comprehend such frames and integrate them in their formation of opinions (Zaller, 1992, p. 42). But what are the implications of these two observations for the impact of the importance change process and the content change process, respectively, as mediators of framing effects?

To mediate issue framing effects, the importance change process requires both that the receiver has at least some considerations about the issue and that the receiver is able to comprehend the issue frame and to integrate it in opinion formation. Therefore, the more politically aware, or sophisticated, citizens should be just as susceptible to framing effects mediated by importance change as the less politically aware citizens, “if not more so” (Nelson, Oxley, & Clawson, 1997, pp. 227–228; see Druckman & Nelson, 2003, p. 732). “This is precisely because more sophisticated individuals are by definition more likely to have such information [i.e., issue relevant considerations] already stockpiled in long‐term memory. Indeed, one could argue that holding such beliefs in memory is necessary for framing to have any impact at all” (Nelson, Oxley, & Clawson, 1997, p. 228; also see Entman, 1993, p. 53). Taking this argument seriously, it must be predicted that the mediating process of importance change will be more salient among the moderately and highly politically aware recipients of an issue frame than among the less politically aware recipients.⁴4 Furthermore, some studies show more ambivalence among sophisticates (Alvarez & Brehm, 2002), and McClosky and Zaller (1984) reports higher “don't know” response rates among elite respondents on questions involving competing values. This could indicate that the more politically aware are in more need of issue frames to help decide the relative importance of competing considerations.
This expectation can be stated as follows:

H2: The importance change process is more important as a mediator of issue framing effects among the moderately and most politically aware citizens than among the least politically aware citizens.

The logic of the other process mediating framing effects—content change—is similar to the logic of traditional persuasion models. As Zaller (1992, p. 126) points out, in general, political awareness is correlated positively with reception of elite and media messages such as issue frames and negatively correlated with uncritical acceptance of such messages. Thus, as Zaller (1992) argues, political messages (including issue frames) more often change the content of considerations, and in turn opinions, among the moderately aware citizens because they can both comprehend media messages—what the least aware citizens can only do to a lesser degree—and more often end up accepting these messages than do highly aware citizens who are generally more resistant to political messages. As a result, in many cases a nonmonotonic relationship between political awareness and the impact of the content change process as a mediator of framing effects can be expected. This is captured in the following hypothesis:

H3: The content change process is more important as a mediator of issue framing effects among the moderately politically aware citizens than among the least or most politically aware citizens.

Thus, I expect that importance change will mediate framing effects among the most politically aware, both importance change and content change will mediate framing effects among the moderately aware, whereas there will only be minimal framing effects among the least politically aware.

Furthermore, the strength of a citizen's values can be expected to moderate through which processes issue framing effects are mediated (see Druckman & Nelson, 2003; Petty & Krosnick 1995; Shen & Edwards, 2005). Values are general or abstract beliefs that “pertain to desirable end states or behaviors” and “guide selection or evaluation of behavior and events” (Schwartz, 1992, p. 4). As such, political values are not only one primary source of specific attitudes and opinions (Feldman, 1988), they also help citizens to interpret and judge political messages like issue frames (e.g., Shah, Domke, & Wackman, 1996; Shen & Edwards, 2005). Accordingly, stronger values are more helpful than weaker values in guiding how to process issue frames.

In relation to the importance change model, strength of values implies that the weaker a person's values, the stronger the importance change model will be as a mediator of framing effects. Weak values do not provide clear guidance about which considerations are more important, for example, free speech or public order. In addition, sometimes weak values are also a source of internal inconsistency of underlying considerations about an issue (Zaller & Feldman, 1992, p. 612), and here an issue frame helps the person to make up his or her mind. Conversely, an individual with strong values does not need an external cue to decide which consideration to emphasize in forming an opinion, because stronger values provide the individual with more clear‐cut information on which position to take on an issue (e.g., Lavine, Huff, Wagner, & Sweeney, 1998). This expectation can be stated as follows:

H4: The importance change process is more important as a mediator of issue framing effects among citizens with weak values than among citizens with strong values.

In relation to the content change model, a similar relationship can be expected: Stronger values dampen the strength of the content change process as mediator of framing effects. As Zaller (1992) argues, values provide important standards used by citizens to filter the stream of political messages they encounter. Thus, the stronger the values, the better equipped a person will be to decide whether to reject or accept an issue frame as a new consideration, because strong values send strong signals on whether to accept or to reject an incoming message. Hence, citizens with weak values will end up accepting issue frames more often than those with strong values and, in turn, more often change the content of their considerations.⁵5 Here, I limit my analysis to investigating this more simple expectation that stronger values will, in general, increase rejection of issue frames and therefore, overall, minimize framing effects. A more detailed account would expect only value‐inconsistent issue frames to have limited impact (see Zaller, 1992, p. 44).
This expectation can be expressed as:

H5: The content change process is more important as a mediator of issue framing effects among citizens with weak values than among citizens with strong values.

Experimental Design and Measures

To test these hypotheses, I conducted an experiment to see through which processes issue framing effects were mediated and if participants' political awareness and strength of political values moderated which mediator was present. The framing manipulation consisted of different versions of a newspaper story. Compared to most previous studies of framing effects, the present design has two distinctive features allowing a more detailed test of possible framing effects. First, the design included a control group making it possible to test the effectiveness of each frame (Chong & Druckman, 2007b). Without the control group as a baseline of comparison, it is impossible to determine if difference in opinion between two treatment groups results from one or the other or both frames having an effect on the measured opinion (Gaines, Kuklinski, & Quirk, 2007). Second, the study included both a pretest and a posttest with a repeated measure of the dependent variable. While randomization makes a between‐subjects analysis of the posttest sufficient to gauge framing effects, the pretest makes it possible to assess if the framing manipulation resulted in actual opinion change.

Participants

A diverse group of Danish students (N = 408) were recruited as participants on a voluntary and nonpayment basis.⁶6 The participants were students at seven folk high schools, a traditional Danish kind of nondegree education, usually obtained between high school and college, in Eastern Jutland. The experiment was conducted from April 30 through June 2, 2004.
This student sample was chosen for convenience, and although not representative of the Danish population, the group is more diverse than would be the case had participants been recruited among university students, and there is a reasonable variation in demographic and political variables.⁷7 Of the participants, 66% were female; median age was 21 years. Participants' political orientations were more left wing than the population as a whole (62% left wing; 19% centrist; 20% right wing), as were their party preferences (31% voted for left‐wing parties; 36% for center‐left parties; 10% for center‐right; 22% for “others” or “don't know”). However, variation in political interest and factual political knowledge approximated the pattern in the entire population (see Goul Andersen & Clement, 2003, p. 413; and Togeby, 2004, pp. 192–195, respectively). No significant differences between experimental groups could be observed on these variables (or on any of 14 other policy opinion items included in the pretest). This suggests that the randomization was successful, and hence, differences between conditions can be attributed to the experimental stimuli.
Due to the young age of the participants, traditional scepticism would suggest the observed framing effects to be perhaps a little stronger than what applies to the population as a whole (Sears, 1986), but recent studies have found no differences in framing effects or related processes between student and nonstudent samples (Kühberger, 1998, p. 36; Miller & Krosnick, 2000, p. 313). Thus, effects of the mediating and moderating variables are presumably generalizable beyond the group of participants. To cover the purpose of the study, participants were informed that they were participating in a study about “how clearly newspapers are able to communicate with their readers” (for similar cover stories, see, e.g., Iyengar, 1991, p. 20; Iyengar & Kinder, 1987, p. 11; de Vreese, 2003, p. 129).

Procedure

In the first sessions, participants completed a paper‐and‐pencil pretest questionnaire booklet containing various demographics questions, the pretest measure of the dependent variable (issue opinion), and a measure of political values (see below). Standard practice in experiments on priming and framing effects is to measure political values and other predispositions in the posttest right after the experimental treatment (see, e.g., Druckman & Nelson, 2003; Nelson, Clawson, & Oxley, 1997; Valentino, Hutchings, & White, 2002). However, there is a risk that the expected framing effects affect not only the dependent variable measure but also the specific items tapping a moderator variable like political values. Therefore, to eliminate any possible impact of the experimental treatment on the measures of political values, these were measured in the pretest questionnaire.

In the second sessions—held between five and 14 days later to prevent pretest sensitization and context effects—participants were randomly assigned to one of three conditions. In the two treatment groups, participants read one of two different manipulated newspaper stories covering a debate about a recent social welfare bill, each representing a different framing of the issue. The control group read a story about a nonrelated issue (international terrorism). The newspaper articles were inserted in questionnaire booklets, which also contained a posttest measure of issue opinion and measures of the two mediators and political awareness. After completion of posttest questionnaires, participants were debriefed. Between the two sessions 90 participants dropped out, reducing the number of cases in the within‐subjects analyses (and analyses involving strength of political values) to 258. On the other hand, 70 individuals who did not take part in the first session participated in the second session, hence increasing the number of cases in the posttest between‐subjects analyses to 328.⁸8 In order not to treat posttest‐only participants any different from other participants, their political values were not measured. Therefore, these participants cannot be included in the analyses involving strength of values as a moderator.

Measures

Issue frames. In the beginning of the posttest booklet, participants were asked to read “a newspaper article from Politiken,” one of the largest, nation‐wide newspapers in Denmark. However, whereas the article assigned to the control group actually was taken from that newspaper,⁹9 “Globalt indgreb mod flytning af terrorpenge”[Global initiative against transactions of terror money]. Politiken, April 26, 2004, Section 1, p. 14.
the two treatment articles had been constructed for the purpose of this experiment and were edited in cooperation with Politiken's editorial staff to make the articles appear as if they were real. Both treatment articles covered a debate about the same social welfare bill. The actual draft bill was proposed by the government in October 2002. The government emphasized that the aim of the bill was to give unemployed citizens receiving welfare benefits a greater economic incentive to find a job. To achieve this end, the government proposed that welfare benefit rates should be decreased after six months on welfare. However, political parties in opposition as well as interest groups framed the proposal by emphasizing that it would only create more poor people, not increase employment, as many of the citizens receiving welfare benefits would not be able to find a job. These two issue frames—the “job frame” and the “poor frame,” respectively—were expressed in the two newspaper articles presented to participants in the experiment.¹⁰10 The two issue frames were identified based on a review of actual press coverage of the social welfare proposal in the leading newspapers and the Danish news agency, Ritzaus Bureau. The identification process followed guidelines suggested by Cappella and Jamieson (1997, p. 47): the frames had their own conceptual and linguistic characteristics; they were commonly observed in a variety of media outlets over a couple of years or more; and they both represent a simplistic and stereotypical interpretation that would enable most people to distinguish them from each other.

The treatment articles were similar in structure, including length, headline size, byline, and number of sources. Both stories presented the same, objective information about the proposal. Thus, both groups read that the proposal was “aimed at setting an upper limit on how much money people on welfare can receive from the government,” that “welfare benefits will automatically be reduced after six months on welfare,” and were provided with estimates of how much money families and singles could lose. In addition, each article contained both issue frames. The job frame was established, in both stories, by explaining that the argument behind the bill was “to give welfare receivers an incentive to find a job” supplemented by information that “Figures from the Ministry of Employment show that today people on welfare do not gain much by taking a job.” The poor frame was established, in the job frame condition, by noting in the second paragraph that “Experts and social workers criticize the proposal for creating more poor people,” while it was in the framed part in the poor frame condition. By presenting the same basic information on the proposal along with a summary of each side's position, it was insured that the main variation between the articles was the selective emphasis on one aspect of the issue, not that they presented totally different information or positions on the issue (Nelson, 2004, p. 586). Furthermore, the inclusion of these identical features made the articles appear consistent with journalistic norms about presenting conflicting positions, as well as served the objective not to make the manipulations unrealistically strong (Kinder & Palfrey, 1993, p. 27).

Following previous studies, the frames were established by varying headlines, story leads, and quotations (e.g., Nelson & Oxley 1999, 1052; see Tankard, 2001). Consistent with definitions of issue framing, headlines emphasized the job aspect, “More people to be forced into jobs,” or the poverty aspect, “New bill creates more poverty.” Quoted sources, the Minister of Employment and the leader of the Danish Association of Heads of Social Services, respectively, constructed the issue by identifying causes of why people are on welfare and conveyed moral judgments (cf. Entman, 2004, p. 5; Gamson & Modigliani, 1987, p. 143). See Appendix A for the articles in full‐length.

Issue opinion. People tend to answer general policy opinion questions in a different way than they answer specific ones (Jacoby, 2000). In this study, posttest opinion measures in the treatment groups related to a specific policy proposal. Therefore, to allow reliable pretest versus posttest and treatment groups versus control group comparisons, the opinion questions in the pretest as well as in the control group posttest were also formulated in a way relating to a specific policy proposal. Thus, the dependent opinion variable was measured in the following way: In the pretest, all participants were asked to what extent they on a 7‐point scale agreed or disagreed with the statement “Social welfare benefits should be cut down in order to give people on welfare a greater incentive to find a job.” In the posttest, after reading the news story, participants in the two treatment groups were asked, “In the article above, a proposal from the government is presented to cut down social benefits in order to give people on welfare a greater incentive to find a job. We would like to hear your opinion on this. Do you agree or disagree on this proposal?” (again, on a 7‐point scale). Question wording in the control group was slightly different: “Recently, the government has proposed to cut down social benefits in order to give people on welfare a greater incentive to find a job. We would like to hear your opinion on this. Do you agree or disagree on this proposal?”¹¹11 I acknowledge that the question wordings might overexpose participants to the “job frame.” However, as this formulation applies to all conditions, it should not affect group comparisons. Furthermore, both within‐subjects and between‐subjects analyses reveal the expected changes and differences in opinion (see below), indicating that the possible bias in the question wordings had only negligible effects.

Importance of considerations. Following previous studies (e.g., Druckman & Nelson, 2003; Nelson, Clawson, & Oxley, 1997), importance of issue‐relevant considerations was measured by participants' rating of five different considerations directly related to the two issue frames and/or relevant to welfare policy opinions (Goul Andersen, 2003). To avoid pretest sensitization and suspicion of the real purpose of the study, measures of importance and content (see below) of considerations were included in the posttest only. Hence, it is not possible in this design to say if issue frames changed the importance (or content) of considerations, but following previous work, by comparing across experimental groups it can be assessed whether the issue frames have an effect on importance ratings. (See Appendix B for question wordings.)

Content of considerations. Beliefs about why people receive welfare benefits, or who is responsible for those people's situation, appear to be important ingredients underlying welfare policy opinions (e.g., Goul Andersen, 2003; Iyengar, 1991; Kluegel & Smith, 1986). Beliefs can be defined as “the cognitive components that make up our understanding of the way things are, that is, the information that individuals have about objects or actions” (Glynn et al., 2004, p. 120). The content of such beliefs or considerations with relevance to the policy of welfare benefit rates was measured with items from Kluegel and Smith's (1986) Explanations for Poverty Scales, translated and adjusted to the Danish context (see Appendix B). Four items formed an Internal attribution scale (α = .78) measuring considerations relating to explanations found at the individual recipient level (e.g., “. . . because they lack proper moral standards and ability to pull themselves together”), while five items formed an External attribution scale (α = .69) measuring considerations relating to explanations found at a sociostructural level or otherwise outside the direct responsibility of the individual recipient (e.g., “. . . because today business and industry only employ high‐efficiency labor”). Thus, these two scales are used as measures of the content of issue‐relevant considerations, but again, the design only allows us to assess if the frames affected the content of considerations, not whether they changed.

Political awareness. In line with Zaller's work (1992), political awareness was measured by six general political knowledge questions (see Appendix B). Based on the number of correct answers participants were divided into three groups: little politically aware (zero to two correct answers; 27% of participants), moderately politically aware (three to four correct answers; 30%), and highly politically aware (five to six correct answers; 43%).

Strength of political values. Values about economic equality (or egalitarianism) can be expected to be relevant to people's opinion formation on the social welfare issue, at least in Denmark (Goul Andersen, 2003; also see Feldman & Steenbergen, 2001). A measure of Equality value was created based on a 5‐item scale (α = .66; see Appendix B). To obtain a measure of strength of value, participants were divided into two groups: Those scoring in the extreme ends of the Equality scale (19% of participants in both ends) were categorized as having strong values and those scoring a moderate (or neutral) position on the Equality value scale were categorized as having weak values (62% of participants).

Results

I expect issue framing effects to be mediated through two different processes, importance change and content change. Further, the relative impact of these processes is expected to be moderated by individuals' political awareness and strength of political values. As a first step in order to test these hypotheses, though, it is necessary to determine whether there is a framing effect to mediate and moderate at all. Support for the social welfare proposal was expected to be higher among participants receiving the “job frame” and lower among participants receiving the “poor frame” relative to the control group. A comparison of opinion across the experimental groups reveals the expected framing effects. Support for the social welfare proposal is significantly higher in the job frame condition than in the poor frame condition (M = .55 vs. M = .36, respectively; t(215) = 4.337, p < .000).¹²12 As all hypotheses are directional, significance tests are one‐sided.
Furthermore, opinion in both treatment groups differs from opinion in the control group (M = .44) in the expected directions (job frame condition: t(216) = 2.554, p < .01; poor frame condition: t(216) = −1.876, p < .05). The within‐subjects analysis further demonstrates that the issue frames did actually move opinions. Paired samples t‐tests show significant changes of opinion in the expected directions in both the job frame condition (M_pretest = .44 vs. M_posttest = .54; t(80) = 3.373, p < .001) and the poor frame condition (M_pretest = .43 vs. M_posttest = .38; t(82) = −1.682, p < .05), while there is no change in the control group (M_pretest = .42 vs. M_posttest = .44; t(92) = 1.007, n.s.).¹³13 The within‐subjects opinion change is further supported in an OLS regression analysis where posttest opinion was regressed on pretest opinion along with dummies for experimental conditions, thus assessing the effectiveness of experimental stimuli controlling for initial opinion (Markus, 1979). Furthermore, including Equality values and a measure of perception of fraud in the social welfare system as controls did not change the magnitude of framing effects. Hence, the observed framing effects on opinion change seem to be robust.
Together these results are consistent with the conclusion that both test issue frames were effective in moving opinion on the social welfare proposal. At the same time, opinion in the control group did not change. For simplicity of presentation, the following sections only report results for comparisons between the two treatment groups.

Mediators of Issue Framing Effects

I analyze mediators of framing effects using the same pathanalytic approach as previous studies (e.g., Druckman & Nelson, 2003; Nelson & Oxley, 1999). A first illumination of which processes mediate issue framing effects can be obtained by looking at the effects the two issue frames have on the indicators of importance of considerations and content of considerations. As can be seen from the results in Table 1, the different framings affected indicators of both processes. First, significant differences could be observed in participants' importance ratings on “Incentive to work,”“Nobody should be poor,” and “Not too expensive,” thus supporting the importance change process as a mediator of the framing effects. Second, the issue frames also influenced the content of participants' considerations, i.e., beliefs about whether recipients of welfare benefits can blame themselves for their lack of job (“Internal attribution”), which supports the content change process as a mediator of framing effects. Nevertheless, it is important to stress that while these variables are indicators of the suggested importance change and content change processes, like in previous studies involving these mediators (e.g., Druckman & Nelson, 2003; Nelson, Clawson, & Oxley, 1997), the variables only measure posttest differences in importance and content of considerations, not change.

Table 1. Framing Effects on Issue Opinion and Indicators of Mediators

	Experimental condition
	Job frame	Poor frame	Difference
Issue opinion	.55	.36	.19***
Importance of considerations:
Incentive to work	.78	.70	.08**
Nobody should be poor	.86	.91	−.05*
Not too expensive	.41	.35	.06*
No fraud	.75	.70	.05
Decent living conditions	.74	.75	−.01
Content of considerations:
Internal attribution	.46	.39	.07*
External attribution	.51	.53	−.02
N (minimum)	105	107

• Note. Entries are mean values; all variables are recoded to range from 0 through 1. Test of significance calculated with independent samples t‐test. *p < .05; **p = .01; ***p < .001 (one‐sided).

Thus, the results in Table 1 support the argument that both mediating processes may be relevant. However, to establish empirically that the processes of importance change and content change really are mediators of the framing effects, the importance and content measures must be affected by the framing and, in turn, influence issue opinion. In addition, the direct effect of the issue framing variable on the opinion variable must decrease after controlling for the mediators. If these conditions are met in the expected direction, the importance of considerations and content of considerations variables can be said to mediate issue framing effects (Baron & Kenny, 1986, pp. 1176–1177). To test whether these conditions are met, a series of OLS regression models were estimated and results are presented in the path model in Figure 2.

Figure 2

Path Model of Mediators of Issue Framing Effects.
The figure displays unstandardized OLS regression coefficients. All variables are recoded to range from 0 through 1. Issue frame is coded such that job frame = 1 and poor frame = 0. Mediator variables are coded such that higher numbers indicate greater perceived importance or greater approval of internal and external attribution beliefs, respectively. Issue opinion is coded such that higher numbers indicate more support for the legislative proposal. Statistically significant mediator indicators are in bold face. N = 203. *p < .05; **p ≤ .01; ***p < .001 (one‐sided).

The results support the proposed dual process model as indicators of both the importance change process and the content change process mediate the framing effects. Except that “Nobody should be poor” did not have an effect on issue opinion, the same indicators as in Table 1 are statistically significant as mediators. In addition, the partial effect of “Issue frame” on “Opinion” is nearly reduced to half the size after controlling for mediators (from unstandardized coefficient = .19, p < .000; to unstandardized coefficient = .11). In sum, results support that issue‐framing effects can be mediated by dual processes consistent with Hypothesis 1.

Political Awareness as a Moderator

To test whether the two mediating processes have different impacts among citizens with different levels of political awareness, the analyses reported in Figure 2 were repeated separately for three groups of participants with different levels of political awareness.¹⁴14 The models reported in Figures 3 and 4 were estimated without including Importance of Decent living conditions and Content of External attributions, because initial analyses revealed they had no mediating effects in any of the models.
I expected importance change to mediate framing effects among the most politically aware, both importance change and content change to mediate framing effects among the moderately aware, whereas only minimal framing effects were expected among the least politically aware. The results as shown in Figure 3 support these expectations from the dual‐process model of framing effects. First, importance of considerations mediates framing effects among moderately and highly aware participants, but not among the least aware. This is consistent with the expectation in Hypothesis 2. Second, consistent with Hypothesis 3, the content change process mediates issue framing effects among the moderately politically aware, but not among the least or most politically aware. These findings clearly suggest a moderated mediation of framing effects, where the actual psychological mechanisms underlying framing effects vary by level of political awareness. The differential presence of importance change and content change further supports that they are two different psychological mediators.

Figure 3

Path Model of Political Awareness as Moderator of Issue Framing Effects.
The figure displays unstandardized OLS regression coefficients. For coding, see Figure 2. Statistically significant mediator indicators are in bold face. Only variables that have at least one statistically significant mediating effect in one of the models are shown. Effect of Issue frame on Opinion before control for mediators: Least aware: unstandardized regression coefficient = .11, n.s.; moderately aware: .31, p < .000; most aware: .16, p < .01. N, least aware = 49; N, moderately aware = 69; N, most aware = 88. ⁺p < .06; *p < .05; **p ≤ .01; ***p < .001 (one‐sided).

Strength of Political Values as a Moderator

Finally, it is tested whether strength of political values moderates the mediating process at stake in translating issue framing into opinion. The same analytical strategy as above is followed, thus repeating the analyses in Figure 2 separately for participants with strong and weak Equality values, respectively. Results of these analyses are presented in Figure 4. As expected, the opinions of those participants with weak values are affected through the process of importance change as well as the process of content change (consistent with Hypothesis 4), whereas there are no framing effects among those with strong values (consistent with Hypothesis 5). Thus, these results suggest that among receivers with weak values, issue frames are able to influence recipients not only through a process of changing importance but also changing content of considerations.

Figure 4

Path Model of Strength of Equality Values as Moderator of Issue Framing Effects.
The figure displays unstandardized OLS regression coefficients. For coding, see Figure 2. Statistically significant mediator indicators are in bold face. Only variables that have at least one statistically significant mediating effect in one of the models are shown. Effect of Issue frame on Opinion before control for mediators: Weak values: unstandardized regression coefficient = .20, p < .001; strong values: .10, n.s. N, strong values = 67; N, weak values = 92.
*p < .05; **p < .01; ***p < .001 (one‐sided).

Discussion

Most previous work on the nature of issue framing effects has investigated either possible mediators or possible moderators of framing effects, but few have attempted to integrate mediators and moderators of framing. In this article, I have extended these approaches by suggesting that a more adequate understanding of the mechanisms of issue framing effects needs to take into account the varying levels of political awareness and differences in strength of values among citizens. My results lend support to the proposed dual‐process model of issue framing effects. First, results showed that issue frames can exert their influence on opinion through two different psychological processes, by changing importance and by changing content of underlying issue‐relevant considerations. Importantly, these two mechanisms appear to act differently depending on receivers' political awareness. The most politically aware were framed through importance change alone, while the moderately politically aware were framed through importance change as well as content change. The frames affected the least politically aware only minimally, and hence no mediators were employed. This pattern clearly suggests moderated mediation of framing effects. Finally, strength of values moderated framing effects, and among receivers with weak values both mediators were present.

These findings replicate the importance change model suggested by Nelson and colleagues (Nelson, Clawson, & Oxley, 1997; Nelson, Oxley, & Clawson, 1997) but at the same time extend their work by shedding some initial light on when this process is supplemented by another mediator: the content change model. My results are also consistent with previous studies finding political awareness to enhance framing effects and strong predispositions to decrease framing effects (e.g., Druckman & Nelson, 2003; Nelson, Oxley, & Clawson, 1997; Shen & Edwards, 2005). However, the argument advanced here goes beyond simply understanding political awareness and predispositions as moderating the magnitude of framing effects; indeed, they appear also to moderate the actual processes through which framing effects occur. Thus, issue framing effects may not act uniformly across individuals.

This study should be seen as only a first step toward investigating more systematically moderated mediation relationships in framing research. Consistent with definitions of issue framing (e.g., Entman, 2004; Gamson & Modigliani, 1987), this study varied how a welfare policy problem was interpreted by focusing on causes of the problem and possible solutions. The present design, though, leaves to future studies to test if the observed moderated mediation of framing effects is different on other issues, or in other types of framing, for example the kind of issue framing Nelson (2004) denotes “goal ranking.”

Relatedly, by investigating effects of only two framed articles, this study cannot probe further into the role of a frame's intensity. Issue frames may vary in intensity in terms both of how often they are repeated and how easy they are to comprehend (see Zaller, 1992). If an issue frame is very easy to understand for everyone, or “loud” enough, the least politically aware also should be influenced by the frame. As shown in the present study, exposure to a frame is not sufficient to be framed. Sniderman & Theriault (2004) make a similar point: “Exposure to a message is one thing; reception, another. This difference between exposure and reception applies with special force the less well‐informed or attentive citizens are to politics” (p. 157). In relation to the dual‐process model proposed here, issue frames “loud” enough to influence the least politically aware would probably extend the content change mediator also to this group, because they should be less resistant than the moderately aware (Zaller, 1992), while it is more of an open question whether or not they would also possess sufficient contextual information to be influenced through importance change (cf. Nelson, Oxley, & Clawson, 1997). Thus, the moderating role of political awareness on magnitude and mediators of framing effects should be studied in contexts with varying intensity of an issue frame. In a promising line of work, Chong & Druckman (2007b) investigate how frames of different strength and repetitiveness influence the least and most politically knowledgeable. They find that the least knowledgeable might not understand the frame, and hence not be influenced in the first place, consistent with the findings in the present study. However, when repeated, even the least knowledgeable were able to get the frame.

Finally, the argument about dual mediators of framing effects made in this paper also relates to how we should conceive a “framing effect.” Some framing researchers suggest reserving the term to include only those instances where a frame in communication influences opinion through the changing importance of preexisting considerations (e.g., Nelson & Oxley, 1999; Nelson & Willey, 2001). In contrast, I have considered a framing effect to be any effect of a frame in communication on a receiver's opinion and thus have treated it as an empirical question through what mediators the frame is able to alter opinion. If we are aiming at understanding through what mechanism(s) an issue frame can influence opinion, the latter definition seems most straightforward. The former definition, however, raises the possibility that a frame in communication may have multiple effects, of which only some are framing effects (i.e., effects mediated through importance change). As Chong & Druckman (2007a) note, “Frames in communication . . . often introduce new considerations about a subject in addition to highlight existing beliefs” (pp. 115–116). This observation is clearly consistent with the results of the present study. But does that mean we should understand the same message as simultaneously a persuasive message and a framing message? Does this not blur the distinction between a frame in communication and an argument? On the other hand, what if persuasive messages or arguments also influence opinion, in part, by changing importance of conflicting considerations underlying opinion? Are persuasive messages able, then, to frame opinion? Chong & Druckman (2007a) observe, “To date there has been little attention given to whether the moderators and mediators of framing and persuasion are similar” (p. 25). Thus, a promising path toward more clarity on the framing concept appears to be investigating mediators and moderators of framing and persuasion effects.

APPENDIX A: Experimental Stimulus

Job frame

More people to be forced into jobs

Government will effectively stop the possibility of people receiving welfare benefits without trying to find a job.

By Christian Hüttemeier

A new government proposal is to motivate people on welfare benefit rates to get out and work instead of passively receiving benefits from the government.

Experts and social workers criticize the proposal for creating more poor people, but the government refuses.

“We have to arrange our social system in a way that always provides an incentive to take a job. We do welfare recipients a disservice if we don't try to find them a job,” says Minister of Employment Claus Hjort Frederiksen.

The controversy is about a new governmental legislative proposal aimed at setting an upper limit on how much money people on welfare can receive from the government. Furthermore, welfare benefit rates will automatically be reduced after six months on welfare.

This means a family can lose up to 2,580 DKK and singles up to 1,290 DKK each month, after taxes.

The argument behind the proposed cut down is to give welfare benefit receivers an incentive to find a job. Figures from the Ministry of Employment show that today people on welfare do not gain much by taking a job.

Away from the dole

“We can neither morally, politically nor in human terms maintain a system that keeps people on the dole because there is no pay off in finding a job. If this proposal is approved, we offer an incentive so that even a little extra effort will give a pay off,” says Minister of Employment Claus Hjort Frederiksen.

“Some people categorically refuse to take a job or to get some help finding a job. Of course they shall not be able to just lean back and passively receive welfare benefits on the expense of tax payers without any consequences what so ever,” Claus Hjort Frederiksen underlines.

Parliament is expected to vote on the proposal before the summer holidays.

Poor frame

New bill creates more poverty

More poor people and greater inequality in society will be the consequences if government cuts welfare benefits to the poor.

By Christian Hüttemeier

A wide range of experts and social workers attack the government's proposal on reducing welfare benefit rates. The bill will hurt the weakest people in society and help to create new generations of poor and socially marginalized citizens, critics warn.

“These cuts hurt the poor. The standard of living of most weak and vulnerable will no doubt be reduced. If you push people this way, you definitely end up making people really weak,” says the leader of the Danish Association of Heads of Social Services, Ole Pass.

The controversy is about a new governmental legislative proposal aimed at setting an upper limit on how much money people on welfare can receive from the government. Furthermore, welfare benefit rates will automatically be reduced after six months on welfare.

This means a family can lose up to 2,580 DKK and singles up to 1,290 DKK each month, after taxes.

The argument behind the proposed cut down is to give welfare benefit receivers an incentive to find a job. Figures from the Ministry of Employment show that today people on welfare do not gain much by taking a job.

Large motivation

“The motivation is large among most welfare recipients. However, their qualifications are not in demand by employers and therefore they cannot just go out and find a job. This proposal is a completely wrong way of dealing with this problem,” Ole Pass says.

“We are talking about a group of people cut off from the job market; we require a more individual effort instead of a standardized welfare cut. All this will do is to make these people even poorer,” Ole Pass underlines.

Parliament is expected to vote on the proposal before the summer holidays.

APPENDIX B: Question Wording

Importance of considerations: “Thinking about the welfare benefit rates, which considerations are in your opinion the most important? Below is a list of different considerations. Please indicate for each consideration how important you think it is, debating the welfare benefit rates.” The responses were measured on a 5‐point scale (1 = “Not at all important consideration”; 5 = “Very important consideration”). The five considerations were: “There should always be an incentive for people to take a job instead of receiving welfare benefits”; “Nobody is to live in poverty”; “Government expenditures on welfare benefits should not be too expensive”; “No defrauder should receive welfare benefits”; and “Unemployed should have benefit rates making it possible to maintain a decent standard of living conditions.”

Content of considerations: “Next, we would like to hear, in your opinion, what might be the reason for some people to receive welfare benefits. Below is a list of different explanations you can agree or disagree with. Please indicate for each explanation how important you think it is as a reason for people receiving welfare benefits. A lot of people receive welfare benefits . . .” Subsequently a list of different explanations was to be rated on 5‐point scales (1 = “Not at all important explanation”; 5 = “Very important explanation”). Internal attribution items: “. . . because welfare benefit rates are so generous that people have no incentive to find a job,”“. . . because they lack proper moral standards and ability to pull themselves together,”“. . . because they need to learn not to be a burden to society,”“. . . because they don't do enough in order to find a job and get on with their lives.” External attribution items: “. . . because even if they really try to get on with their lives, some people simply fail to succeed on today's labor market,”“. . . because the school system did not ensure that they finished an educational degree,”“. . . because today business and industry only employ high‐efficiency labor,”“. . . because some people just have bad luck,” and “. . . because the country's economy does not provide the right conditions for people to find a job.”

Political awareness: “Finally, here are some questions about politics. For each question, please write what you think is the right answer” (open‐ended); “Which political parties are in the governing coalition?”; “Who is responsible for governing the public hospitals?”; “Among the Danish political parties, some more than others support immigrants and refugees. Do you think The Social Liberals are more or less supportive?”; “Some political parties attach greater importance than others to immediate tax cuts. Do you think The Conservative Party is among those attaching greater or lesser importance to tax cuts?”; “Is Austria a member of the European Union?”; and “Which international organization does Kofi Annan lead?”

Political value: The Equality value scale contained five items. Three of them were measured with response categories on a 7‐point Likert scale: “High incomes ought to be taxed more strongly than they are today”; “In politics one should strive to give everybody the same economic conditions, no matter their education and employment”; and “Many public sector activities could be made both better and cheaper if left to private firms.” Two items were measured in the format of opposing statements: “First a question about social expenditures. A says: Social reforms have gone too far in this country. More than now, people should manage without social security and support from the public sector.—B says: Those social reforms that have been made in our country should be maintained at least to the same extent as now. Do you mostly agree with A or B?” and “It is also discussed whether the users of public sector services should pay more. A says: There should be more users' payment in the public sector.—B says: Apart from small and insignificant areas, there should not be more users' payment.” Response categories were “Agree with A,”“Agree with B,”“Neither,” and “Don't Know.”

ACKNOWLEDGMENT

I thank Carsten Daugbjerg, Jamie Druckman, Stanley Feldman, Christoffer Green‐Pedersen, Peter Nannestad, Thomas Nelson, Michael Bang Petersen, Rune Stubager, Lise Togeby, Claes de Vreese, the anonymous reviewers, and participants in the Research Section on Danish Politics, Department of Political Science, University of Aarhus, for helpful advice. I also benefited from discussions in the framing research group at the Amsterdam School of Communications Research, University of Amsterdam. Earlier versions of this paper were presented at the 2005 Annual Meeting of the Midwest Political Science Association in Chicago, IL, and the 2005 Conference of the Nordic Political Science Association in Reykjavik, Iceland. I acknowledge support from the Research Foundation of University of Aarhus. Correspondence concerning this article should be sent to Rune Slothuus, Department of Political Science, University of Aarhus, Universitetsparken, 8000 Aarhus C, Denmark. E‐mail: [e-mail].