Hoffman

Communication Research http://crx.sagepub.com/

When the World Outside Gets Inside Your Head: The Effects of Media Context on Perceptions of Public Opinion
Lindsay H. Hoffman Communication Research published online 2 February 2012 DOI: 10.1177/0093650211435938 The online version of this article can be found at: http://crx.sagepub.com/content/early/2012/05/07/0093650211435938

Published by:
http://www.sagepublications.com

Additional services and information for Communication Research can be found at: Email Alerts: http://crx.sagepub.com/cgi/alerts Subscriptions: http://crx.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav

>> OnlineFirst Version of Record - May 13, 2012 OnlineFirst Version of Record - Feb 2, 2012 What is This?

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

435938
435938HoffmanCommunication Research © The Author(s) 2011 Reprints and permission: http://www. sagepub.com/journalsPermissions.nav

CRXXXX10.1177/0093650211

When the World Outside Gets Inside Your Head:The Effects of Media Context on Perceptions of Public Opinion
Lindsay H. Hoffman1

Communication Research XX(X) 1­–23 © The Author(s) 2012 Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/0093650211435938 http://crx.sagepub.com

Abstract Citizens are variably influenced by information flow depending on their location within a social structure. One method of understanding this influence is through an assessment of multiple levels of analysis. Although many scholars have called for such analyses, few have heeded that call. This research addresses the relevance of “context” to the study of media effects on perceptions of public opinion. Survey data from the American National Election Studies are combined with a content analysis of campaign news in 24 regional newspapers, as well as advertising data, in order to parse out contextual media effects. Results show that perceived public opinion varies significantly across media markets. Newspaper use and personal candidate preference had a significant effect on the likelihood of perceiving Kerry to be the state-winning candidate.There was also a significant cross-level interaction between media context and political discussion on perceptions of public opinion. Keywords multilevel modeling, media effects, public opinion, social reality

In the early 20th century, Walter Lippmann’s treatise on the “world outside and the pictures in our heads” encouraged generations of communication scholars to examine the effects of the environment on perceived social reality. More recently, Price (1988; Price & Roberts, 1987) called for an assessment of media as the central mechanism linking members of mass society, and McLeod, Pan, and Rucinski (1995) argued that “it is essential to develop cross-level theories,” if we are to understand public opinion processes (p. 76; see also Pan & McLeod, 1991). This study examines contextual effects on perceptions of public opinion in a highly charged
1

University of Delaware, Newark, DE, USA

Corresponding Author: Lindsay H. Hoffman, Department of Communication, Department of Political Science & International Relations, University of Delaware, 250 Pearson Hall, Newark, DE 19716, USA Email: [email protected]

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

2

Communication Research XX(X)

political campaign, adding to existing theoretical work on the impersonal influence (Mutz, 1998) and social projection (Mullen et al., 1985). By examining the influence of perceptions of public opinion in a multilevel context, the research demonstrates more nuanced effects of traditional indicators, such as political discussion, on perceptions of public opinion.

The Influence of Individual Media Use on Public Opinion Perceptions
Media operate as “epistemological devices” (McLeod et al., 1995, p. 73) by which individuals understand the social and political world, providing the raw material for individuals to understand the opinion climate around them, ultimately influencing what we know and how we know it (McLeod et al., 1995; Rhodebeck, 1995). Much research has demonstrated that media use and content have effects on perceived public opinion and social reality (Christen & Gunther, 2003; Daschmann, 2000; Gunther, 1998; Gunther & Storey, 2003; Mutz, 1989, 1998; Mutz & Soss, 1997). What has not been explored is whether these perceptions might differ by contexts like media markets, where both news coverage and actual climate of opinion may vary. For many political issues, media serve as the primary source of information (Books & Prysby, 1991, 1995; Christen & Gunther, 2003; Huckfeldt & Sprague, 1995; Mutz, 1989). But, how do these effects manifest themselves differently depending on where an individual lives? Residing in different regions may encourage people to obtain information about issues depending, at least in part, on their location. Depending on the sampling of public opinion polls and how close the race is, media consumers will receive different information about which candidate is ahead and which is behind. For instance, in the 2004 U.S. presidential election, some polls showed Bush and Kerry in a dead heat, others revealed Bush as the winner, and still others predicted Kerry would win (Patterson, 2005). Representations of public opinion have been found to influence individuals’ perceptions of public opinion, which can be defined as an individual’s “awareness, assessment or sense of relevant others’ opinions” (Glynn, Ostman, & McDonald, 1995, p. 253). As much of the work on perceptions originated in the fields of psychology and social psychology, the primary impetus for research has been discovering cognitive mechanisms that produce perceptions, and misperceptions, of social reality. However, individual perceptions are influenced by other mechanisms as well, namely media contexts. For example, research has demonstrated that media have an “impersonal influence,” impacting social perceptions of an issue (Mutz, 1998). Eveland (2002) acknowledged that many of the theories and hypotheses on perceptions of social reality focus primarily on the misperceptions of public opinion and media effects. Although the accuracy of perceptions—whether a person has perceived the media or others correctly or not—is important and a central component of research discussed here, we are at risk of perhaps missing the bigger picture. “Whether they are accurate or not,” Gunther and Storey (2003) concluded, “perceptions can have a self-fulfilling effect on the realization of communication goals” (p. 213). As such, the perceptions individuals have of social reality can influence their behavior, regardless of whether they are accurate. The accuracy of these perceptions, then, is arguably not as relevant as answering the question: Do these

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

Hoffman

3

perceptions vary by information context? If these perceptions vary when the actual content to which individuals are exposed varies, we might also be able to assess whether those perceptions influence political behavior.

The Influence of Media Context on Public Opinion Perceptions
Many of the conceptualizations of “context” have their origins in political science. But what is a media context? Context itself has been defined in numerous ways across multiple disciplines. Although definitions vary, Huckfeldt and Sprague (1995) differentiated “networks,” which are individually structured, from “contexts,” which are structurally imposed. Books and Prysby (1991, 1995) offer perhaps the most exhaustive and mutually exclusive account of the dimensions of context, in addition to providing a succinct definition. Their definition is rooted in the primitive meaning of the term: “a geographically bounded unit” (Books & Prysby, 1991, p. 2; see also Weatherford, 1983). Books and Prysby were specifically interested in “local context,” which they conceptualized as “areally defined groups of people” (p. 3). The boundary of this unit—often specified as neighborhoods, communities, or states—is what differentiates it from nongeographic “networks” such as families, clubs, and interest groups. A media context, then, is a geographically bounded unit that is defined by the media that are specific to that unit. We also know that at the individual level, media influences cognitions like political efficacy and trust (Miller, Goldenberg, & Erbring, 1979) and candidate name recognition (Goldenberg & Traugott, 1980). Also, research has connected survey data with content analyses to demonstrate that beyond individual characteristics, media have the potential to set the public agenda (e.g., Iyengar & Kinder, 1987; McCombs & Reynolds, 2002). Yet what is missing from many of these studies is attention to cross-level influences; instead of examining individual and contextual variables in concert, many scholars simply assume some sort of flow from media content to the public without clearly examining the mechanisms of that information flow. This flow of information serves as a primary mechanism for contextual effects, and some seminal work has demonstrated such effects (e.g., Dalton, Beck, & Huckfeldt, 1998; Huckfeldt & Sprague, 1995; Orbell, 1970; Shaw, 2006; Tichenor, Donohue, & Olien, 1980). When it comes to election campaigns, in particular, research has demonstrated that campaign strategies vary significantly by location and, at the aggregate level, this influences how well a candidate does in a particular location (e.g., Campbell, Alford, & Henry, 1984; Ridout, Shah, Goldstein, & Franz, 2004; Shaw, 2006; Stewart & Reynolds, 1990). These campaigns, and the information therein, transform over time as a result of political communication between elites, the media, and individuals. In this way, campaign effects can be traced to changes in political communication—or the flow of information between actors—over time.

Linking Individual- and Contextual-Level Influences
As individuals obtain information about presidential campaigns primarily through media or interpersonal discussions (Books & Prysby, 1991, 1995; Feldman & Price, 2008; Huckfeldt

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

4

Communication Research XX(X)

& Sprague, 1995; Mutz, 1989), it is important to assess the nature of that media content, which varies by context. Moreover, the effects of media might also vary by context. Yet much of the research has tended to aggregate individual-level information to create contextual data. For example, Gimpel, Dyck, and Shaw (2004) aggregated voter turnout in neighborhoods to assess homogeneity of party identification, concluding that Republicans vote more, but vote significantly less when more Democrats are in their local surroundings. This study examines effects on perceptions of public opinion using both individual- and contextual-level data. But, why might we expect to see different effects for individuals in varying contexts? Theoretically, two methods for information processing may provide some idea as to why individuals in different contexts might interpret information differently: accessibility and applicability. The former suggests that the information that comes to mind most readily will be the information most likely to be used in making a judgment (Higgins, 1996; Shrum, 2002). The process by which media can affect audience evaluations is sometimes referred to as knowledge activation (Price & Tewksbury, 1997). As such, what one “knows” about his or her reality can vary depending on what is perceived in the media. The knowledge activated by the media that is judged to be applicable to the current situation (e.g., when answering a survey question about who will win an election) influences how that information is interpreted (Roskos-Ewoldsen, Roskos-Ewoldsen, & Carpentier, 2002). As Shrum (2002) noted, “media consumption enhances accessibility, which influences the information that becomes a part of that small subset of available information” (p. 74). These determinants lead to judgments of set size and probability, suggesting that one might interpret, for example, how many people in one’s state support a candidate based on the information they obtain through the media. To further theorize how these individual-level processes might be influenced by context, we might look to the associative network model of memory (Roskos-Ewoldsen, RoskosEwoldsen, & Carpentier, 2002; Srull & Wyer, 1979), which purports that information is stored in nodes that are connected by associative pathways. An underlying assumption of this model is that people are cognitive misers, and heuristics serve as a less cognitively straining form of retrieving information (Price & Tewksbury, 1997). In this sense, if one’s environment—that is, the media context—emphasizes one candidate over another, an individual might presume that this represents the opinion in his or her community, and that this is reality. These various theoretical perspectives can be combined to propose that (a) people who are exposed to information (b) tend to process information in ways that are not cognitively straining, and (c) make interpretations based on the way they have processed this information when called upon to make decisions about related issues or topics.

Thinking Multilevel
Research has tended to see media as having fixed and unvarying relationships with individuals across space (McLeod & Blumler, 1987). Although communication research has long acknowledged the effects of community-level influences on personal behaviors and attitudes (e.g., Tichenor, Donohue, & Olien, 1980), it is only recently that the methods to

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

Hoffman
MACRO (contextual level)

5
Media content Frequency of candidate mentions

Moderating Variables Candidate preference Political Discussion MICRO (individual level)

Perceptions Perceived statewinning candidate

Figure 1. Proposed cross-level interactions between media content and perceptions with moderating variables at the individual level

assess these effects have become more widespread in the discipline (Slater, Snyder, & Hayes, 2006). As many media effects, such as perceptions of public opinion, are perceived to be indirect (e.g., Gunther, 1998; Mutz, 1989, 1998), it becomes even more essential to include multiple levels of influence. Also, by including contextual data in the analysis, it is possible to account for greater variation in this outcome variable, ultimately increasing explanatory power (Jerit, Barabas, & Bolsen, 2006). Acknowledging the inherently multilevel nature of the present phenomenon, I first propose a baseline prediction based on the assumption that the frequency of candidate mentions will differ across media markets, and therefore its effects on perceived public opinion will vary. In other words, if contextual differences are to be found, then we must first establish that the perceptions of public opinion vary significantly across different media contexts. Hypothesis 1 (H1): There will be significant group-level heterogeneity among media contexts in perceived public opinion. Contextual effects do indeed vary by individual characteristics and predispositions, and such effects can only be asserted when individual behavior depends on an external factor after individual-level determinants are taken into account (Huckfeldt, 1980; Huckfeldt & Sprague, 1995). But, does media content influence perceptions of public opinion as these perceptions interact with individual-level predictors? As communication is a “multilevel phenomenon” (Slater et al., 2006), multilevel modeling is an appropriate method because it is more flexible than other methods in handing unbalanced groups, missing data, and analyzing models with multiple variables at each level (Slater et al., 2006). Multilevel models also allow researchers to examine cross-level interactions to determine whether the effects vary across levels of analysis (Steenbergen & Jones, 2002). See Figure 1 for a graphical representation of the proposed cross-level relationships.

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

6

Communication Research XX(X)

Media content serves as the contextual (or Level 2) measure here, and is operationalized by a measure of the ratio of Kerry-to-Bush candidate references in regional newspapers. The measure of Kerry-to-Bush mentions provides a snapshot of the media environment during the presidential campaign. Those individuals who are more attentive to the campaign by reading the newspaper—an individual-level variable—might arguably perceive public opinion to correspond to what they read in the news. As variation in media attention affects political knowledge (Jerit et al., 2006), it is hypothesized to have a variable effect on perceptions of public opinion. In other words, the more an individual uses the newspaper, and the more that newspaper mentions Candidate A over Candidate B, the more likely that respondent might be to say that Candidate A is likely to win his or her state. This is rooted in the theories of information processing outlined above, which suggest that the more frequently and recently a person receives information, the more readily accessible that information will be when called upon. Yet the media environment is also likely to influence these perceptions. Hypothesis 2 (H2): Media context will moderate the effect of newspaper reading on perceived public opinion, such that individuals with high levels of newspaper reading who reside in media contexts that mention one candidate more frequently will perceive that candidate to win the state. However, when asked to describe others, individuals often simply refer to their own self-reactions (Robbins & Krueger, 2005). This social projection hypothesis suggests that perceptions of public opinion will rely mostly on personal opinions, such that people presume others to have the same opinions as themselves. Research has demonstrated that projection and media effects are likely to interact in predicting perceptions of public opinion (Shamir & Shamir, 2000). The question driving the next hypothesis is, simply, what role do personal opinions play in influencing perceptions when we look at the actual media context? When taking media context into account, if a person prefers Candidate A, but there is more content in his media market about Candidate B, will projection effects persist? It is possible that the “reality” of one’s media context might diminish the effects of social projection, because people tend to see media coverage as mirroring public opinion (Gunther, Christen, Liebhart, & Chia, 2001; Gunther & Christen, 2002). This research suggests a convergence of individual perceptions of those who perceive Candidate A and Candidate B once the actual media environment is accounted for. However, it is uncertain which direction this effect will follow, so I propose a nondirectional hypothesis: Hypothesis 3 (H3): Media context will significantly moderate the effect of individual candidate preference on perceived public opinion. Interpersonal discussion about politics can also influence perceptions of public opinion (e.g., Huckfeldt & Sprague, 1995) and media context moderates this effect (e.g., Freedman, Franz, & Goldstein, 2004). Huckfeldt (1986) and Huckfeldt and Sprague (1995) suggest that social interaction is a key mechanism for contextual effects, because such interaction is more likely among those individuals nearer to each other in location. Social interaction

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

Hoffman

7

is only one of several mechanisms through which context affects individuals (Books & Prysby, 1991; Erbring & Young, 1979), but it is the dominant framework in the literature on contextual effects. Individuals obtain politically relevant information from their surroundings and this information is naturally locationally biased (Burbank, 1997). In other words, media can provide information that is in many ways dependent on and reflective of geographic location, thus serving as the fodder for political discussion. It is hypothesized that media context will moderate the effect of social interaction on perceptions in much the same way as H2. Hypothesis 4 (H4): Media context will significantly moderate the effect of political discussion on perceived public opinion, such that individuals with high levels of political discussion who reside in media contexts that mention one candidate more frequently will perceive that candidate to win the state. The analyses will also control for political advertising because, in addition to media content, the proliferation of advertising in a media market is likely to influence perceptions of who will win an election. What will be the media effects, above and beyond advertising? Research has demonstrated that exposure to campaign advertising can increase interest, discussion, knowledge, voting likelihood, and vote intentions (e.g., Brians & Wattenberg, 1996; Freedman, Franz, & Goldstein, 2004; Johnston, Hagen, & Jamieson, 2004; Ridout et al., 2004; Zhao & Chaffee, 1995). However, examining these effects in a multilevel model is an intuitive way to add depth to this line of research. Therefore, a ratio of Democratic-to-Republican favored ads will also be included in these models.

Method
This study used three sources of data: (a) newspaper content related to the 2004 presidential campaign from August 24 to November 1, 2004, (b) a survey of Americans during this campaign (American National Election Studies [ANES], 2004), and (c) advertising data from the University of Wisconsin Advertising Project (2004).

Newspaper Content Analysis
A matching process was undertaken to match individual cases from the ANES data file with newspapers in their geographic area, during which number of cases were systematically removed from analysis because they did not match the geographic data, or the geographic locale did not include electronically accessible newspapers. The process of matching newspapers to individual cases from the ANES data file was as follows: 1. Match respondent’s PSU to media context: First, upon obtaining information about the primary sampling unit (PSU) for each individual respondent in the ANES 2004 survey file, I cross-referenced that metro area with information

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

8

Communication Research XX(X)

obtained from the Standard Rate and Data Service (SRDS; 2003). If that area was not included in the list of metro areas, I looked up the county in which the area was located. Then, for each individual census tract, I used the U.S. Census 2000 Factfinder (U.S. Census, 2000) and Social Explorer (2006) to map whether that census tract was indeed included in the metro area specified by the PSU. First, the U.S. Census Factfinder “geo within geo” selection category was used in order to select states, then specific census tracts within each state. The decision factor was whether the census tract was in either the “urban area” or “place” as specified by the U.S. Census that matched the PSU given by ANES. The urban area classification is a collective term that cuts across other hierarchies and can be in metropolitan or nonmetropolitan areas. The “place” is a concentration of population either legally bounded as an incorporated place—which have legal descriptions of borough, city, town, or village—or identified as a “Census Designated Place” (U.S. Census, 2000). If the census tract included pieces of either the urban area or place, it was included. Only if the census tract did not include either urban area or place associated with the PSU was it excluded. If it was unclear from Factfinder whether the census tract should be included, I used Social Explorer (2006) to confirm, because Social Explorer’s maps permit zooming and also have cloropleth functionality, which can indicate areas of greater population density such as metro areas. 2. Select newspaper(s) in the media market: I then selected the newspaper(s) for each metro area that had the highest circulation figures from the SRDS (2003). If the area was rural, ANES data were collected by county, so I used the SRDS (2003) county-area analysis and cross-referenced with either the American Factfinder or Social Explorer to confirm that the census tract was indeed in that county and not near other metro areas. I also checked the SRDS (2003) for adjacent counties if census tracts bordered other counties; if a newspaper included in those adjacent counties had greater circulation than the one(s) reported for the county of interest and were in the same state, it was also included. This followed the same procedure as if a media market had competing newspapers, such that content from both newspapers was included in the media context matching that respondent. 3. Assess communities with more than one newspaper: There were two cities with two newspapers operated by the same company: Detroit Free Press and Detroit News, and the Seattle Post-Intelligencer and Seattle Times. In order to maintain a reasonable amount of content while also representing the content in those areas, one newspaper from each company was randomly selected to include in the sample. These newspapers were the Detroit Free Press and the Seattle Times. Four other communities included more than one newspaper that were not run by the same company. These were Denver (Denver Post and Rocky Mountain News), New York (New York Times and New York Daily News), Salt Lake City, Utah (Salt Lake Tribune and Deseret News), and Boston/Worcester (Boston Globe and Worcester Telegram & Gazette). For both Denver and Salt Lake City, both newspapers were listed as having relatively equal rates of circulation, and thus

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

Hoffman
the content in both newspapers were included for cases in those communities. In New York and Boston, the distribution of newspapers throughout the city varied by location, so cases were assigned newspapers based upon their location in the area and the highest corresponding circulation for that area. 4. Deal with missing data and assess deletion of cases: A number of ANES cases were excluded before content data collection began for various reasons. One reason was that individual cases in certain census tracts did not fit in the metro or urban area. These census tracts (and the corresponding number of excluded cases) were census tracts 7,101 and 131 (Flint, Michigan) in the Detroit Metropolitan Statistical Area, or MSA (17 cases); census tracts 125.03 and 153.02 (Florida) in the Lakeland-Winter Haven MSA (9 cases); census tract 230 in the Manchester-Nashua, NH MSA (3 cases); and census tract 4027.22 (Missouri) in the St. Louis MSA (6 cases). Two other cases were deleted simply because they were the only cases with a given newspaper, and it was deemed inefficient to include them. These were both in New Haven, Connecticut in areas where the Connecticut Post and the Waterbury Republican American were listed as having the highest circulation, respectively.

9

Finally, a number of cases were deleted because their matching newspapers were not obtainable in available online databases. These newspapers were the (New Jersey) Press of Atlantic City (29 cases), Columbus (Georgia) Ledger-Enquirer (19 cases), Des Moines (Iowa) Register (22 cases), Shreveport (Louisiana) Times (46 cases), Austin (Minnesota) Daily Herald (47 cases), Meridan-Wallingford (Connecticut) Record Journal (3 cases), White Plains (New York) Journal News (4 cases), Camden/Cherry Hill (New Jersey) Courier Post (3 cases), Saginaw (Michigan) News (29 cases), Greensburg (Indiana) Daily News (47 cases), and the Wheeling (West Virginia) Intelligencer/News Register (12 cases). Three additional newspapers only provided content from the previous six months, so they and their corresponding cases were deleted: the Boston Globe (18 cases), Los Angeles Times (5 cases), and New York Newsday (15 cases). In this study, the media data are at Level 2 of the multilevel model, and the individual survey data are at Level 1. Designated market areas (DMAs) served as the contextual unit of analysis for the newspaper data. DMAs define markets at the local level and allow researchers to examine news coverage by selecting those newspapers with the greatest circulation by area (Long, Slater, Boiarsky, Stapel, & Keefe, 2005). Long et al. (2005) suggested that newspapers with the largest circulation have the largest reach and, in turn, the largest resource base at their disposal. The total daily circulation, which includes weekday and Sunday rates, was used as the reference for largest circulation (SRDS, 2003). Keywords were entered into the search functions of online newspaper databases to assess which terms retrieved the most inclusive sample of content (i.e., content related to the 2004 presidential campaign). The inclusiveness of each set of search terms was assessed by examining the precision of that set of search terms. Precision is an estimate of the “conditional probability that a particular text is relevant, given that is retrieved”

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

10

Communication Research XX(X)

(Hornik, Stryker, Yanovitzky, & Wray, 2006, p. 415). A minimum precision rate was set a priori at 80% (as exemplified in Hornik et al., 2006). This stage essentially serves as the filter stage, through which articles applicable to the campaign are sampled.1 With the article as the unit of analysis, I also tested for recall (Hornik et al., 2006) by using a more open set of search terms (e.g., “election”) on the same sample. Recall is an estimate of the “conditional probability that a particular text will be retrieved, given that it is relevant” (Stryker et al., 2006, p. 414). The most accurate and precise search term was “Kerry or Bush and campaign.” The average precision estimates for the term were, across databases and newspapers, 84.6%. The average recall estimate, too, was above the a priori standard at 88.2%, averaged across databases and newspapers. Wordstat computerized content analysis software was used to analyze content (Provalis Research, 2005). The dictionary creation process included three stages: (a) purposefully selecting terms (such as “Bush” or “Kerry”); (b) analyzing with Key Word in Context (KWIC) to assess whether any of these terms were included in inappropriate contexts, and (c) using the “feature extraction” tool to extract words and phrases in the actual sample of content that were unidentifiable by any dictionary. In order to test that the dictionary was reliably capturing content, a random subset of the articles (10%, or 1,031 articles) was analyzed, and this subset was then manually coded. Krippendorf’s alphas were .94 for the two “Bush” and “Kerry” dictionaries. A ratio was then computed for Kerry-to-Bush content per each media context. This ratio provides—in one number—a snapshot of how frequently each candidate appeared in a given newspaper.

2004 ANES Face-to-Face Survey
Employing multistage sampling, the 2004 ANES served as an ideal survey for the analysis because of its grouped nature. The face-to-face preelection survey was conducted September 7 to November 1, 2004. The area probability sample consisted of a cross section of respondents that yielded 1,212 face-to-face interviews in the preelection study. Postelection interviews were administered November 3 through December 20. The preelection response rate was 66.1% and the postelection response rate was 88%. Newspaper readership was measured by one item that asked how many days in the past week the respondent read a print newspaper (M = 2.87, SD = 2.81).2 Individual candidate preference was measured with, “Who do you think you will vote for in the election for President?” or if the respondent indicated he or she was not likely to vote, “If you were going to vote, who do you think you would vote for in the election for President?” Kerry was preferred among 46% of respondents, while 45% preferred Bush (8.8% provided some other response). Perceived public opinion was measured by individual responses to the ANES 2004 question, “Which presidential candidate will carry this state?” Of all respondents, 46.5% responded with Kerry; 45.6% responded with Bush (8% provided some other response). The political discussion item was measured by whether or not respondents “ever talk about politics with family and friends.” More than 70% of respondents said they did, and 20.2% said they did not.

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

Hoffman

11

Advertising Data
The University of Wisconsin Advertising Project (2004) data were used to assess the proportion of Democratic-to-Republican advertisements aired by state.3 A variable was created that compared the number of ads supportive of the Democratic candidate (Kerry) to the number of ads supportive of the Republican candidate (Bush) in each state. These data were collected by the Campaign Media Analysis Group (CMAG) using a system that monitors the transmissions of national broadcast and cable networks, as well as local adverting in the top 100 markets (University of Wisconsin Advertising Project, 2004). All ad airings prior to September 7, 2004 were deleted from the dataset to concur with the NES data, and ads airing only on cable were also deleted, since information on which viewers in which states actually viewed the advertisements was not available. Five media markets were removed from the data analysis because they were not available in the Wisconsin advertising data (Chicago, Lakeland, Richmond, Roanoke, and Salt Lake City). The result of the matching process of the NES data, newspaper data, and advertising data was a final sample of 554 people and 22 newspapers.

Analysis
As some NES cases were excluded in this process, the pared-down sample was compared to the original data, and there were no significant differences in either demographics or the variables of interest in the models. The mean age of respondents in the final trimmed sample (n = 554) was 46.72 (SD = 16.83) and the median response to level of education was “more than 12 years of schooling, no higher degree.” White respondents comprised 70.2% of the final sample, while 17.9% were Black and 5.1% were Hispanic, with the remainder including categories of combined race, unknown, or refused responses. Median household income was US$45,000 to US$49,999. The Level 1 (NES) and Level 2 (newspaper content and advertising) data were analyzed with HLM software (Raudenbush, Bryk, & Congdon, 2007), using a logistic link function.

Results Media Context Analysis
A total of 10,878 articles were coded in the original 27 contextual units. The New York Times featured the most articles (n = 1536, 14.1%), followed by the Washington Post (n = 1076, 9.9%). The Worcester Telegram & Gazette had the least amount of articles (n = 81, 0.7%). In order to have a logical measure of candidate-mention frequency, a ratio of Kerryto-Bush mentions was computed. This ratio can be interpreted as how many Kerry mentions there are for every one Bush mention per thousand words.4 Table 1 shows the Kerry-toBush ratios for each media market. The number of ads was also calculated from September 7 to Election Day. The greatest number of ads aired in Florida (63,990 ads) and the fewest in California (1 ad). The descriptive statistics for both Level 1 and Level 2 data, after missing values were deleted, are provided in Table 2.

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

12

Communication Research XX(X)

Table 1. Ratio of Kerry-to-Bush Mentions by Newspaper and Ratio of Democratic-toRepublican Ads Per State
Kerry-toBush Ratio in Newspaper Contenta 0.99 0.61 0.94 0.85 0.91 0.89 0.84 0.78 0.62 0.89 0.81 0.95 0.92 0.85 0.78 0.94 0.77 0.75 0.89 0.81 0.81 1.24 Democrat-toRepublican Ratio in Adsb 1.34 149.00 19.78 475.00 2.19 1.59 2.19 2.00 7.00 989.00 3.44 1.79 2.04 90.00 90.00 1.74 2.00 2.00 2.97 2.74 11.33 19.78 % Respondents Accuratec 56.0 76.5 88.2 66.7 89.5 84.0 50.0 63.6 92.6 82.6 60.0 100.0 64.4 90.9 100.0 55 86.7 86.7 91.9 60.0 78.6 89.2

Newspaper Arkansas Democrat Gazette Birmingham News Boston Globe Buffalo News Cleveland Plain Dealer Denver Post/Rocky Mountain News Detroit Free Press Fresno Bee Houston Chronicle Knoxville News Sentinel Manchester Union Leader Miami Herald Milwaukee Journal Sentinel New York Daily News New York Times Philadelphia Inquirer Sacramento Bee San Francisco Chronicle Seattle Times St. Louis PostDispatch The Washington Post Worcester Telegram & Gazette
a

State Arkansas Alabama Massachusetts New York Ohio Colorado Michigan California Texas Tennessee New Hampshire Florida Wisconsin New York New York Pennsylvania California California Washington Missouri District of Columbia Massachusetts

State winner Bush Bush Kerry Kerry Bush Bush Kerry Kerry Bush Bush Kerry Bush Kerry Kerry Kerry Kerry Kerry Kerry Kerry Bush Kerry Kerry

This ratio provides a snapshot of how frequently each candidate appeared in a given newspaper, and represents Kerry mentions to Bush mentions. The dictionary creation process included three stages: (a) purposefully selecting terms (such as “Bush” or “Kerry”), (b) analyzing with Key Word in Context (KWIC) to assess whether any of these terms were included in inappropriate contexts, and (c) using the “Feature Extraction” tool to extract words and phrases in the actual sample of content that were unidentifiable by any dictionary. Krippendorf’s alphas were .94 for the two “Bush” and “Kerry” dictionaries. b As California is considered a safe state, only one ad was aired during the period measured, resulting in a ratio of 2 to 1 (it was a Kerry ad) because 0 cannot be included in a ratio. c The average percentage of respondents across all states who were correct about their state-winning candidate was 77.9%.

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

Hoffman
Table 2. Descriptive Statistics Level 1 Descriptive Statistics Variable Newspaper use (days) Candidate preference (1 = Kerry) Perceived state-winning candidate (1 = Kerry) Political discussion (1 = yes) Level 2 descriptive statistics Variable Ratio of Kerry-to-Bush mentions Ratio of Dem-to-Rep ads
Note. J refers to the number of Level-2 units.

13

N 429 429 429 429 J 22 22

M 3.05 0.51 0.56 0.82 M

SD 2.83 0.50 0.50 0.38 SD

Min Value 0 0 0 0 Min Value 0.61 1.34

Max Value 7 1 1 1 Max Value 1.24 989.00

0.86 0.13 86.27 226.77

Multilevel Model
Park, Eveland, and Cudeck (2008) have suggested a step-by-step approach to multilevel modeling by first assessing a baseline model to see whether there is even variability across contextual units—in this case, the model tested whether media content varied across contexts. A baseline model was estimated in order to assess the variability of the means across contextual units. As the outcome is binary (0 = perceived Bush win, 1 = perceived Kerry win), a multilevel logistic model was employed.5 Hypothesis 1 proposed that there would be significant group-level heterogeneity in perceptions of who would win a respondent’s state, and received initial support from the chi-square test, χ2 (21) = 146.00, p < .0001. A commonly used descriptive statistic in multilevel modeling is the intraclass correlation coefficient (ICC), which illustrates group-level heterogeneity. This measure reveals how similar individuals nested within one group (here, media market) different from individuals in other groups. The ICC indicated that about 42% of the variance in perceived statewinning candidate is between media markets, which means there is a clustering effect in these data. If the ICC had been zero, there would be no clustering effect, or no differences in the outcome based upon being located in a group.6 Following the steps proposed by Park et al. (2008), Hypotheses 2, 3, and 4 were tested with an Intercepts and Slopes as Outcomes (ISO) model, which includes predictors at both Levels 1 and 2 (Raudenbush & Bryk, 2002) and the dependent variable at Level 1. This model, shown in Table 3, assesses how powerful one’s own preferences, newspaper reading, and discussion habits are in the presence of contextual newspaper and advertising information on the likelihood of perceiving Kerry to win the election in one’s state, controlling for the other predictors. The ISO model is essentially an explanatory model that accounts for variability across media contexts, using the full potential of multilevel

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

14

Communication Research XX(X)

Table 3. Results From Intercepts-and-Slopes-as-Outcomes Multilevel Logistic Model Predicting Perceived State-Winning Candidate Fixed Effects Intercept model  Intercept, γ00   Kerry-to-Bush ratio, γ01   Dem-to-Rep ad ratio γ02 Newspaper use model  Intercept, γ10   Kerry-to-Bush ratio, γ11   Dem-to-Rep ad ratio γ12 Candidate preference model  Intercept, γ20   Kerry-to-Bush ratio, γ21   Dem-to-Rep ad ratio γ22 Political discussion model  Intercept, γ30   Kerry-to-Bush ratio, γ31   Dem-to-Rep ad ratio γ32 Random effects Intercept, u0j Newspaper use, u1j Variance component 1.73 0.19 Coefficient 0.45 3.57 −0.00 0.13 0.23 −0.00 1.91* −1.78 0.00 0.41 4.42* 0.00 df 19 19 OR 1.57 35.57 1.00 1.14 1.26 1.00 6.77 0.17 1.00 1.51 83.17 1.00 χ2 124.77 24.97 p .28 .28 .27 .08 .69 .14 .00 .40 .91 .24 .048 .26 p .000 .161

Note: Restricted Maximum Likelihood Estimation (REML) was used for estimates of variance components. Unit-specific results were used to interpret the multilevel logistic model. All predictors entered into the models were grand-mean centered. *p ≤ .05. **p ≤ .001.

modeling in order to examine whether there is a significant cross-level interaction between media context and individual variables in predicting perceptions of public opinion. In this way, we can model the variability in the regression coefficients—both intercepts (expected score of a person who is zero on all Xi) and slopes (the expected change in score with a one-unit increase in Xi)—across Level 2 units (Raudenbush & Bryk, 2002). Results show that, not surprisingly, individual candidate preference was a positive and significant predictor of perceived state-winning candidate in this model (see Table 3). The odds ratio was 6.77 for candidate preference, which means that in looking at Bush versus Kerry supporters, the odds of a Kerry supporter perceiving a Kerry win were significantly higher than for a Bush supporter, contingent on other variables in the model at their means. However, to truly tap into the multilevel nature of these data, the important coefficients in this model are the interaction terms between media context and the individual-level predictors. Hypothesis 2 proposed that higher newspaper reading would interact with

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

Hoffman

15

Figure 2. A model-based graph of the cross-level interaction between media context and political discussion on perceived public opinionaThe y-axis represents the Level 1 variable measuring individuals’ likelihood of saying that Kerry would win their state. The x-axis represents the Level 2 variable measuring the ratio of Kerry newspaper mentions in one’s media context to Bush mentions in one’s media context.

media context to predict perceived state-winning candidate. The results show that there was no significant interaction effect on perceived state-winning candidate when newspaper use increased by one unit (this was also the case for advertising data). Hypothesis 2 was not supported. Hypothesis 3 predicted that media context would moderate the effect of individual candidate preference to predict perceived state-winning candidate. There was no significant cross-level interaction, so Hypothesis 3 was not supported. Hypothesis 4 predicted that media context would moderate the effect of political discussion on perceived public opinion. Table 3 shows that the Kerry-to-Bush ratio in media context did significantly moderate the effect of political discussion on perceived state-winning candidate. The differentiating effect of political discussion on perceptions of public opinion increased when this contextual variable was entered into the model. The odds ratio is 83.17 (p < .05), suggesting that, when controlling for Democrat-to-Republican advertising ratio, newspaper use, and candidate preference at their average, the odds of a person who discussed politics saying Kerry would win was higher than people who didn’t discuss politics, with a one-unit increase in Kerry-to-Bush content. This was, in fact, the only significant cross-level interaction in the ISO model. As Figure 2 depicts, talking about politics interacted significantly with one’s media context to influence a perceived Kerry win, such that

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

16

Communication Research XX(X)

more Kerry content and political discussion increased the likelihood that a respondent, who is average on the other predictors in the model, would perceive Kerry as the winner.

Discussion
The purpose of this study was to assess the effects of media content on perceptions of public opinion accounting for the context of the media market in which individuals live. This study also aimed to advance theoretical work on multiple levels of analysis, as called for by several scholars. Overall, the results suggest that there are significant differences between individuals in various media markets in whom they perceived to be the state-winning candidate in the 2004 presidential election. At the individual level, a personal preference for Kerry influenced the likelihood that respondents would say Kerry would win the state. The significant direct effect of candidate preference on perceived public opinion is not surprising, if one adopts the idea that individuals project their own opinion onto others. Yet as the ratio of Kerry-to-Bush content increased, those people who talked about politics with family and friends were more likely to perceive Kerry as winning one’s state. At the very least, these results suggest that there are indeed cross-level interactions between media context and individual-level predictors—in this case, political discussion—in predicting perceptions of public opinion. Yet there is more to the puzzle than simply demonstrating that cross-level interactions exist. Importantly, for respondents who discussed politics, the likelihood of saying Kerry would win one’s state was increased when the Kerry-to-Bush ratio increased by one unit. However, this interpretation is actually somewhat unrealistic, because a one-unit increase in Kerry-to-Bush content translates into twice as many Kerry mentions than Bush mentions, which did not occur in these data. One way of obtaining a more realistic measure is to raise the odds to a fraction, such as one tenth. In this way, 4.42 becomes 4.421/10, or 1.16. This can be interpreted as for every one-tenth unit increase in the ratio of Kerry-to-Bush content, the likelihood of saying Kerry would win increased by 54% when a person discussed politics and was average on other predictors in the model. In other words, above and beyond the direct effects of candidate preference and political discussion alone, there is a synergy between political discussion and one’s media context to increase the likelihood of perceiving a candidate to win one’s state. Essentially, these results suggest that media effects were amplified for people who discussed politics and lived in a relatively Kerry-heavy media context. This “amplification” effect suggests that if interpersonal political discussion mirrors (at least in its relative emphasis on one candidate over another) one’s media context, a person becomes potentially more likely to perceive that the public will share that opinion. These results support the work of Huckfeldt and Sprague (1995), who conjectured that, while people may choose their discussion partners and topics, these choices are bounded by a social and geographic structure. In the public opinion literature, we see that interpersonal communication has long played an important role. We see it in the anecdotes, such as Lippmann’s (1922) story of the islanders who “had acted as if they were friends, when in fact they were enemies” (p. 1) because they were relying on old news. We see it in early theoretical models of the

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

Hoffman

17

public opinion process, such as Davison’s (1983) assertion that public opinion is rooted in interpersonal discussion. Also, we see the evidence that interpersonal communication drives public opinion in myriad studies, such as Hoffman, Glynn, Huge, Thomson, & Seitman’s (2007) finding that interpersonal discussion was one of the “filters” that influenced perceived public opinion in the case of a community issue. The present study provides further evidence for the role of interpersonal discussion in the development of public opinion, suggesting that discussion at the individual level interacts with media content at the contextual level to fuel perceptions of social reality. The implications here are that communication—both interpersonal and mediated—can play a synergistic role in creating the “pictures in our heads” (Lippmann, 1922, p. 1) that drive so many important phenomena like voting behavior, political activity, and opinion incongruity (Hayes, Scheufele, & Huge, 2006; Scheufele & Eveland, 2001). These results also support Mutz’s (1998) theory of impersonal influence, where people assume the media are powerful in influencing others, resulting in real consequences for cognitions and behavior. Applied to the present study, we might conclude that political discussion—which was likely based at least in part on what was available in the media context—influenced citizens’ perceptions of who would win their state in the 2004 election. Mutz (1998) proposed that the mechanism driving impersonal influence is likely to be different for “citizens with differing levels of information” (p. 216). The findings herein support that notion, such that citizens who talked more and had more Kerry content in their media context were more likely to be influenced in their perception of a Kerry win. Yet many readers will recall that Bush won the 2004 election. These results point to a truly perceptual effect, where social reality appears to have driven perceptions, such that people who talked about politics and lived in markets where there was relatively more Kerry coverage perceived a Kerry win. This has important implications for the study of both communication and public opinion, suggesting that social reality does not always reflect objective reality. Future research should examine these patterns in other election campaigns as well as in off-election times, to examine the roles that objective and social reality play in driving individuals’ perceptions. It is important to note that there were no significant effects of the ratio of Democraticto-Republican advertisements on perceived state-winning candidate. Previous research has identified advertising content as having an influence on interest, behavior, and attitudes, but these results suggest that advertising content has little to no effect on perceived public opinion. In the present study, at least, advertising was more “noise” than effect on who respondents thought would win the election in their state. The measure of media content was admittedly conservative: a simple frequency of terms associated with either candidate. Yet the use of computerized content analysis allowed for the sample size to be quite large and reduced the amount of human error that is inherently involved in coding media content. Moreover, the exclusion of television, radio, or other local media inherently limits our understanding of these media contexts. Although some research has demonstrated that television news generally does not contain as much campaign coverage as newspapers (e.g., Bernstein & Lacy, 1992), there is certainly reason to believe that including more media sources would provide a more complete picture of a

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

18

Communication Research XX(X)

given media context. Including a measure of valence would also shed light on these results; Perhaps, the tone in which candidates were covered played an important role in the perceived likelihood of winning. Future research should examine valence to see if it plays a significant moderating role in these relationships.7 The measure for political discussion is also a simple one. NES simply asked respondents whether they talked to family or friends about politics. Much research has found that the level of agreement or disagreement in those conversations can have tremendous effects on perceptions. Future research could examine the role of agreement or disagreement in conversations and the subsequent effects on perception. In addition, the variance component for the intercept in the bottom of Table 3 is significant, suggesting that there is still some variance among media markets in the perceived state-winning candidate that has not been accounted for by these Level 1 and Level 2 predictors. Finally, some scholars (e.g., Orbell, 1970) argue that there are several mechanisms by which people obtain political information beyond social interaction and the media. There are clearly other sources of information and types of media content that might add both power and depth to the present analyses. When dealing with a phenomenon such as public opinion, that is transitory and bound up with place and time (Noelle-Neumann, 1993), it is essential to examine potential combinations of processes of influence. Although we know that the media do influence perceptions of public opinion, it is necessary to account for multiple levels of influence. These results suggest that locationally biased information can indeed lead to differential effects among voters. The methods used in the present study, although they have limitations, provide some framework for examining these effects within and across levels of analysis, and scholars are encouraged to use such methods to unravel individual- and contextual-level effects on both cognitions and behaviors.

Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: These materials are based on work supported by, in alphabetical order, the National Science Foundation under grant SES-0118451, the University of Michigan, and the University of Wisconsin. Any opinions, findings and conclusions or recommendations expressed in these materials are those of the author and do not necessarily reflect the views of the funding organizations.

Notes
1. Of articles retrieved, 10% were sampled. Inclusion and exclusion terms available from the author. 2. These means and percentages reflect the entire sample and the descriptive for the reduced sample is given in Table 2. 3. Only 12 media markets included in the newspaper data were available in the top 100 markets selected by the Wisconsin Advertising Project, so state data were used instead of market data.

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

Hoffman

19

4. Rate per thousand words was chosen over story as the unit of analysis because of its standardized nature. 5. Restricted Maximum Likelihood Estimation (REML) was used for estimates of variance and covariance components. Unit-specific results were used to interpret the multilevel logistic model. All predictors entered into the models were grand-mean centered. The analytical approach follows that proposed in Park et al., 2008. In general, centering removes high correlations between the random intercept and slopes, as well as high correlations between Level 1 and Level 2 variables and cross-level interactions (Kreft & de Leeuw, 1998). Centering also makes interpretation easier; with grand-mean centering, the intercepts are interpreted as the expected outcome for a subject whose value on Xij is equal to the grand mean. The variance is interpreted as the variance among Level 2 units in the adjusted means (Raudenbush & Bryk, 2002). 6. The ICC for a logistic model can be computed as, ρ = τ00 / (τ00 + π2/3) (Raudenbush & Bryk, 2002, p. 334), which in this case was 2.39 / (2.39 + 3.29) = 0.42. 7. Another interesting future research question might examine the differences between national (The Washington Post, The New York Times) and local newspapers. Although these data demonstrate that the content measured here didn’t differ dramatically across state and national papers, (e.g., The Washington Post’s DMA and The St. Louis Post-Dispatch’s DMA both had a Kerry-to-Bush ratio of 0.81), there could be differences in other types of content.

References
American National Election Studies. (2004). The 2004 National Election Study (data set). Ann Arbor, MI: University of Michigan, Center for Political Studies. Retrieved from www.electionstudies.org Bernstein, J. M., & Lacy, S. (1992). Contextual coverage of government by local television news. Journalism Quarterly, 69, 329-340. Books, J., & Prysby, C. (1991). Political behavior and the local context. New York, NY: Praeger. Books, J., & Prysby, C. (1995). On the future of contextual models of political behavior. In M. Eagles (Ed.), Spatial and contextual models in political research (pp. 255-273). London, UK: Taylor & Francis. Brians, C. L., & Wattenberg, M. P. (1996). Campaign issue knowledge and salience: Comparing reception from TV commercials, TV news, and newspapers. American Journal of Political Science, 40, 172-193. Burbank, M. J. (1997). Explaining contextual effects on vote choice. Political Behavior, 19(2), 113-132. Campbell, J. E., Alford, J. R., & Henry, K. (1984). Television markets and Congressional elections. Legislative Studies Quarterly, 9, 665-678. Christen, C. T., & Gunther, A. C. (2003). The influence of mass media and other culprits on the projection of personal opinion. Communication Research, 30, 414-431. Dalton, R. J., Beck, P. A., & Huckfeldt, R. (1998). Partisan cues and the media: Information flows in the 1992 presidential election. American Political Science Review, 92, 111-126. Daschmann, G. (2000). Vox pop and polls: The impact of poll results and voter statements in the media on the perception of a climate of opinion. International Journal of Public Opinion Research, 12, 160-181. Davison, W. P. (1983). The third-person effect in communication. Public Opinion Quarterly, 47, 1-15. Erbring, L., & Young, A. A. (1979). Individuals and social structure: Contextual effects as endogenous feedback. Sociological Methods and Research, 7, 396-430. Eveland, W. P., Jr. (2002). The impact of news and entertainment media on perceptions of social reality. In J. P. Dillard & M. Pfau (Ed.), The persuasion handbook: Developments in theory and practice (pp. 691-727). Thousand Oaks, CA: SAGE.

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

20

Communication Research XX(X)

Feldman, L., & Price, V. (2008). Confusion or enlightenment? How exposure to disagreement moderates the effects of political discussion and media use on candidate knowledge. Communication Research, 35(1), 61-87. Freedman, P., Franz, M., & Goldstein, K. (2004). Campaign advertising and Democratic citizenship. American Journal of Political Science, 48, 723-741. Gimpel, J. G., Dyck, J. J., & Shaw, D. R. (2004). Registrants, voters, and turnout variability across neighborhoods. Political Behavior, 26, 343-375. Glynn, C. J., Ostman, R. E., & McDonald, D. G. (1995). Opinions, perception, and social reality. In T. L. Glasser & C. T. Salmon (Ed.), Public opinion and the communication of consent (pp. 249-277). New York, NY: Guilford. Goldenberg, E. N., & Traugott, M. W. (1980). Congressional campaign effects on candidate recognition and evaluation. Political Behavior, 2, 61-90. Gunther, A. C. (1998). The persuasive press inference: Effects of mass media on perceived public opinion. Communication Research, 25, 486-504. Gunther, A. C., & Christen, C. T. (2002). Projection or persuasive press? Contrary effects of personal opinion and perceived news coverage on estimates of public opinion. Journal of Communication, 52, 177-195. Gunther, A. C., Christen, C. T., Liebhart, J. L., & Chia, S. C. (2001). Congenial public, contrary press, and biased estimates of the climate of opinion. Public Opinion Quarterly, 65, 295-320. Gunther, A. C., & Storey, J. D. (2003). The influence of presumed influence. Journal of Communication, 53, 199-215. Hayes, A. F., Scheufele, D. A., & Huge, M. E. (2006). Nonparticipation as self-censorship: Publicly observable political activity in a polarized opinion climate. Political Behavior, 28, 259-283. Higgins, E. T. (1996). Knowledge activation: Accessibility, applicability, and salience. In E. T. Higgins & A. W. Kruglanski (Eds.), Social psychology: Handbook of basic principles (pp. 133-168). New York, NY: Guilford. Hoffman, L. H., Glynn, C. J., Huge, M. E., Thomson, T., & Seitman, R. B. (2007). The role of communication in public opinion processes: Understanding the impacts of individual, media, and social filters. International Journal of Public Opinion Research, 19, 1-26. Hornik, R., Stryker, J., Yanovitzky, I., & Wray, R. (2006). Validation of database search terms for content analysis: The case of cancer news coverage. Journalism & Mass Communication Quarterly, 83, 413-430. Higgins, E. T. (1996). Knowledge activation: Accessibility, applicability, and salience. In E. Tory Higgins & Arie W. Kruglanski (Eds.), Social psychology: Handbook of basic principles (pp. 133-168). New York, NY: Guilford. Huckfeldt, R. (1980). Variable responses to neighborhood social contexts: Assimilation, conflict, and tipping points. Political Behavior, 2, 231-257. Huckfeldt, R. (1986). Politics in context: Assimilation and conflict in urban neighborhoods. New York, NY: Agathon. Huckfeldt, R., & Sprague, J. (1995). Citizens, politics, and social communication. Cambridge, UK: Cambridge University Press. Iyengar, S., & Kinder, D. (1987). News that matters: Television and American opinion. Chicago, IL: University of Chicago Press. Jerit, J., Barabas, J., & Bolsen, T. (2006). Citizens, knowledge, and the information environment. American Journal of Political Science, 50, 266-282.

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

Hoffman

21

Johnston, R., Hagen, M. G., & Jamieson, K. H. (2004). The 2000 Presidential Election and the Foundations of Party Politics. Cambridge, UK: Cambridge University Press. Kreft, I. G. G., & de Leeuw, J. (1998). Introducing multilevel modeling. Thousand Oaks, CA: SAGE. Lippmann, W. (1922). Public opinion. New York, NY: Free Press. Long, M., Slater, M. D., Boiarsky, G., Stapel, L., & Keefe, T. (2005). Obtaining nationally representative samples of local news media outlets. Mass Communication & Society, 8, 299-322. McCombs, M. E., & Reynolds, A. (2002). News influence on our pictures of the world. In J. Bryant & D. Zillmann (Eds.), Media effects: Advances in theory and research (pp. 1-18). Mahwah, NJ: Lawrence Erlbaum. McLeod, J. M., & Blumler, J. G. (1987). The macrosocial level of communication science. In S. H. Chaffee & C. R. Berger (Ed.), Handbook of communication science (pp. 271-322). Newbury Park, CA: SAGE McLeod, J. M., Pan, Z., & Rucinski, D. (1995). Levels of analysis in public opinion research. In Theodore L. Glasser & Charles T. Salmon (Eds.), Public opinion and the communication of consent (pp. 55-85). New York, NY: Guilford. Miller, A. H., Goldenberg, E. N., & Erbring, L. (1979). Type-set politics: Impact of newspapers on public confidence. American Political Science Review, 73, 67-84. Mullen, B., Atkins, J. L., Champion, D. S., Edwards, C., Hardy, D., Story, J. E., & Vanderklok, M. (1985). The false consensus effect: A meta-analysis of 115 hypothesis tests. Journal of Experimental Social Psychology, 21, 262-283. Mutz, D. C. (1989). The influence of perceptions of media influence: Third person effects and the public expression of opinions. International Journal of Public Opinion Research, 1, 3-23. Mutz, D. C. (1998). Impersonal influence: How perceptions of mass collectives affect political attitudes. Cambridge, UK: Cambridge University Press. Mutz, D. C., & Soss, J. (1997). Reading public opinion: The influence of news coverage on perceptions of public sentiment. Public Opinion Quarterly, 61, 431-451. Noelle-Neumann, E. (1993). The spiral of silence: Public opinion, our social skin. Chicago, IL: University of Chicago Press. Orbell, J. M. (1970). An information-flow theory of community influence. Journal of Politics, 32, 322-338. Pan, Z., & McLeod, J. M. (1991). Multilevel analysis in mass communication research. Communication Research, 18, 140-173. Park, H., Eveland, W. P., Jr., & Cudeck, R. (2008). Multilevel modeling: People within groups and contexts. In Andrew F. Hayes, Michael D. Slater, & Leslie B. Snyder (Eds.), The Sage handbook of advanced data analysis methods for communication research (pp. 219-246). Thousand Oaks, CA: SAGE. Patterson, T. E. (2005). Of polls, mountains: U.S. journalists and their use of election surveys. Public Opinion Quarterly, 69, 716-724. Price, V. (1988). On the public aspects of opinion: Linking levels of analysis in public opinion research. Communication Research, 15, 659-679. Price, V., & Roberts, D. F. (1987). Public opinion processes. In Steven H. Chaffee & Charles R. Berger (Eds.), Handbook of communication science (pp. 781-816). Newbury Park, CA: SAGE. Price, V., & Tewksbury, D. (1997). News values and public opinion: A theoretical account of media priming and framing. In George A. Barnett & Franklin J. Boster (Eds.), Progress in the communication sciences (pp. 173-212). New York, NY: Ablex. Provalis Research (Version 5.1) [Computer software]. (2005). Wordstat. Montreal, QC: Provalis Research.

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

22

Communication Research XX(X)

Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data analysis methods (2nd ed.). Thousand Oaks, CA: SAGE. Raudenbush, S. W., Bryk, A. S., & Congdon, R. (2007). HLM: Hierarchical Linear and Nonlinear Modeling (Version 6.04, Computer software). Lincolnwood, IL: Scientific Software International. Rhodebeck, L. A. (1995). Contextual determinants of reference group behavior. In M. Eagles (Ed.), Spatial and contextual models in political research (pp. 235-254). London, UK: Taylor & Francis. Ridout, T. N., Shah, D. V., Goldstein, K. M., & Franz, M. M. (2004). Evaluating measures of campaign advertising exposure on political learning. Political Behavior, 26, 201-225. Robbins, J. M., & Krueger, J. I. (2005). Social projection to in-groups and out-groups: A review and metaanalysis. Personality and Social Psychology Review, 9, 32-47. Roskos-Ewoldsen, D. R., Roskos-Ewoldsen, B., & Carpentier, F. R. D. (2002). Media priming: A synthesis. In Jennings Bryant & Dolf Zillmann (Eds.), Media effects: Advances in theory and research (2nd ed., pp. 97-120). Mahwah, NJ: Lawrence Erlbaum. Scheufele, D. A., & Eveland, W. P., Jr. (2001). Perceptions of “public opinion” and “public” opinion expression. International Journal of Public Opinion Research, 13, 25-44. Shamir, J., & Shamir, M. (2000). The anatomy of public opinion. Ann Arbor, MI: The University of Michigan Press. Shaw, D. R. (2006). The race to 270: The electoral college and the campaign strategies of 2000 and 2004. Chicago, IL: University of Chicago Press. Shrum, L. J. (2002). Media consumption and perceptions of social reality: Effects and underlying processes. In J. Bryant & D. Zillmann (Eds.), Media effects: Advances in theory and research (2nd ed., pp. 69-95). Mahwah, NJ: Lawrence Erlbaum. Slater, M. D., Snyder, L., & Hayes, A. F. (2006). Thinking and modeling and multiple levels: The potential contribution of multilevel modeling to communication theory and research. Human Communication Research, 32, 375-384. Srull, T. K., & Wyer, R. S. (1979). Category accessibility and social perception: Some implications for the study of person memory and interpersonal judgment. Journal of Personality and Social Psychology, 37, 1660-1672. Standard Rate and Data Service. (2003). Newspaper circulation. Retrieved from http://www.srds.com/ frontMatter/ips/circulation/index.html Steenbergen, M. R., & Jones, B. S. (2002). Modeling multilevel data structures. American Journal of Political Science, 46, 218-237. Stewart, C., & Reynolds, M. (1990). Television markets and U.S. Senate elections. Legislative Studies Quarterly, 15, 495-523. Stryker, J. E., Wray, R. J., Hornik, R. C., & Yanovitzky, I. (2006). Validation of database search terms for content analysis: The case of cancer news coverage. Journalism & Mass Communication Quarterly, 83(2), 413-430. Tichenor, P. J., Donohue, G. A., & Olien, C. N. (1980). Community conflict and the press. Beverly Hills, CA: SAGE. University of Wisconsin Advertising Project. (2004). Retrieved from http://wiscadproject.wisc.edu/project. php U.S. Census Bureau. (2000). Factfinder. Retrieved from http://factfinder.census.gov/home/saff/main. html?_lang=en

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012

Hoffman

23

Weatherford, M. S. (1983). Evaluating economic policy: A contextual model of the opinion formation process. Journal of Politics, 45, 866-888. Zhao, X. S., & Chaffee, S. H. (1995). Campaign advertisements versus television news as a source of political issue information. Public Opinion Quarterly, 59, 41-65.

Bio
Lindsay H. Hoffman (PhD, The Ohio State University, 2007) is an assistant professor in the Departments of Communication and Political Science & International Relations. She also serves as research coordinator in politics and technology for the Center for Political Communication at the University of Delaware, Newark, Delaware, United States.

Downloaded from crx.sagepub.com at National School of Political on August 28, 2012