International Journal of Public Opinion Research Vol. No. ß The Author . Published by Oxford University Press on behalf of The World Association for Public Opinion Research. All rights reserved. doi:./ijpor/edm Advance Access publication July
THE ROLE OF COMMUNICATION IN PUBLIC OPINION PROCESSES: UNDERSTANDING THE IMPACTS OF INTRAPERSONAL, MEDIA, AND SOCIAL FILTERS Lindsay H. Hoffman, Carroll J. Glynn, Michael E. Huge, Rebecca Border Sietman, and Tiffany Thomson ABSTRACT This study examined multiple factors associated with the process of public opinion including relevant predispositions, media use, interpersonal discussion, and perceptions of community opinion in order to test a theoretical model of public opinion. We conceptualized these factors as intrapersonal, media, and social ‘filters’ within the public opinion process. To test the impact of these filters, we conducted a survey with two independent samples—the first sample was collected during the introduction phase of a community ballot issue and the second just a week before the issue vote. Findings indicate all three filters impacted public opinion regarding the ballot issue. Within these filters, important subprocesses were analyzed to better understand each filter’s contribution to the formation of public opinion. Ordinary least squares (OLS) regression equations used to test the proposed process model revealed that the intrapersonal filter accounted for a substantial amount of the overall variance in public opinion, but that media and social filters were also important predictors. Results highlight the importance of communication variables in the formation of public opinion.
The study of the public opinion process in social science literature often includes psychological (attitudes and beliefs), social (group discussion and norms), and political (elite perspectives presented in the media) components. Communication has received less attention as a central variable in this process, but scholars have long argued that opinions develop through a dynamic This article was first submitted to IJPOR May , . The final version was received April , .
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
discourse between cognitive and social components (e.g. Davison, ; Price & Roberts, ; Van Leuven & Slater, ; Kinder, ; Glynn, ). While communication is often seen as imbedded in the process, it is clear that its role should be specifically investigated and examined across multiple levels, ranging from individual predispositions and interpersonal influence to media content and use. Public opinion researchers often focus only on one level of the process at a single point in time, examining how individuals form opinions (e.g. Price, ; Pan & Kosicki, ), conform to majority pressure (e.g. Asch, ; Moscovici, ; Blanton & Christie, ), or are influenced by mass media presentations of public opinion (e.g. Mutz & Soss, ; Moy, McCoy, Spratt, & McCluskey, ; Tsfati, ). However, when conceptualizing public opinion as a process, these various components should be examined together as they overlap and intertwine to form public opinion. The purpose of this study is to simultaneously examine multiple factors associated with the process of public opinion—including relevant predispositions, media use, interpersonal discussion, and perceptions of community opinion—in order to arrive at a more comprehensive understanding of the overall opinion formation process. The public opinion process can be conceptualized as interactions among various influential factors. We conceptualize these factors as intrapersonal, media, and social ‘filters,’ and propose a series of hypotheses specific to each filter. We also examine the filters together as predictors of favorability toward a local community issue.
PUBLIC OPINIO N AS A PROCESS Although the idea of public opinion as a process has been elucidated by numerous scholars (Bryce, ; Foote & Hart, ; Davison, ; Price & Roberts, ; Price, ; Noelle-Neumann, ; Crespi, ), few have proposed empirical models for examining the process. Noelle-Neumann (, ) has been credited with developing a theory that integrates mass communication effects and public opinion,1 but few scholars have attempted to apply a design that taps into each part of the public opinion process described above.2 Our conceptual model of the public opinion process is based on the theoretical contributions of Davison (), Price and Roberts (), and Crespi (). By examining multiple components of the public opinion process in one model, we are able to observe the impacts of these factors at two different points in the process. We assessed this process in the context 1 Although some have argued that Noelle-Neumann excluded the role of communities, organizations, and reference groups (Glynn & McLeod, ). 2 However, see Davison () for a technique on simulating the public opinion process through roleplaying games.
PUBLIC OPINION FILTERS
of a community school levy-and-bond ballot issue—both at early-stage development and at the issue’s culmination. DEVELOPING
A
THEORETICAL FRAMEWORK
Three models that acknowledge the overlapping and interactive effects of different predictive factors guide the present study. Davison’s () model of communication examines the interplay between public and private discussions and appears to be the basis for several contemporary perspectives about the public opinion process, including work by Price and Roberts (), Noelle-Neumann (), and Crespi () (see Glynn, ). According to Davison, an issue develops momentum when an idea is communicated from one person to another; only widely discussed topics will develop into public issues. In this sense, public opinion does not just appear, but rather takes root in interpersonal discussion, eventually developing into a social force. Price and Roberts () divide the public opinion process into inter-level relations among individuals, groups, and organizations over time. Communication is central to the process, even at the intrapersonal level, where the relationship between cognitions and behavior can be conceptualized as a ‘continuing dialectic’ (p. ). According to Price and Roberts (), information obtained from both media and social sources is integrated with old information as public opinion evolves. Individuals incorporate new opinionrelated information with old cognitions, selecting which new ideas to incorporate and which to dismiss or ignore. Finally, Crespi’s () model of public opinion includes () interactions among predispositions and perceptions of the external world at the intrapersonal level; () the collective opinions that emerge from communicating these individual opinions through discussion and the media; and () the legitimization or enactment of those opinions. This explication across multiple levels enables scholars to view public opinion as a process, preventing fallacies, such as reification or reductionism, that have plagued researchers in the past (for further discussion on these issues, see Allport, , Price, , p. ). Building upon these theoretical models, our model focuses on three sources of influence: () the central role of predispositions and interests (i.e. ‘intrapersonal filters’); () the vital function of the media as disseminators of public opinion and issue specific information (i.e. ‘media filters’); and () the importance of interpersonal political discussion in giving an issue momentum and strengthening opinion, as well as supplying additional relevant information (i.e. ‘social filters’). For most issues, there are almost infinite number of possible perspectives that could be considered as public opinion takes shape, yet it is only a limited number of these perspectives that will ultimately affect the public opinion process. How are the few selected from the many? Each source of influence inherently filters out certain information while
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
allowing other ideas to become integrated with the overall opinion framework. Specifically, these filters sort out what is most important, conferring salience on certain aspects of a given issue. Filters also capture the temporal nature of public opinion by acknowledging that the process of public opinion formation occurs over time and progresses through various stages. Yet as public opinion evolves, the impact of different filters is not entirely lost. In this way, public opinion is iteratively shaped and transformed as it moves through each stage, or filter. From Davison’s and Price and Roberts’ models, we incorporate the essential components of interpersonal and mediated communication. Crespi’s model provides us with a temporal framework from which to analyze the process of public opinion, and also includes predispositions. With these perspectives in mind, our model incorporates the main influence of each of these key filters as well as the interactions between them. In other words, we will attempt to demonstrate that intrapersonal cognitions, media exposure, and interpersonal discussion are often intricately intertwined in their influence on the formation of public opinion. Figure presents our conceptualization of the public opinion process, with communication as a driving force behind the model. This model demonstrates that, following issue introduction, the intrapersonal filter is activated. Individuals think through the issue, assimilating information in line with their own predispositions and interests, often basing these considerations on relevant individual factors. These factors could include demographic characteristics, such as income or the presence of children in one’s household, or on relatively stable attitudes, such as political ideology or attitudes toward tax increases. As an issue progresses and approaches a focusing event, such as an election, media coverage and interpersonal discussion increase, and subsequent awareness of an issue also increases (Downs, ). In addition, opinions are simultaneously shaped and strengthened by perceptions gained through discussion and media use. Each of these components has the potential to influence ‘public opinion,’ operationalized as the expression of opinions around a focusing event.
INTRAPERSONAL FILTERS Explication of public opinion first requires assessment of the emergence of individual opinions that are activated when an issue is introduced (Crespi, ). Zaller () refers to these activated cognitions as ‘political predispositions . . . which regulate the acceptance or nonacceptance of the political communications the person receives,’ (p. ). In the early stages of opinion development, individuals likely have little issue-specific knowledge on which to base their views, and might rely more heavily on predispositions that are associated with the topic at hand.
PUBLIC OPINION FILTERS
Issue introduction
Increasing
Time Demographics Issue awareness
Intrapersonal filters Predispositions Issue interest
Media Coverage Social filters
Media filters
Issue discussion
Exposure & attention
Perceived community support
Discussion
Focusing event
Opinion strength
Issue culmination
Public opinion
Figure A model of the public opinion process
In the present study, these predispositions are defined as social and economic consequences as well as personal importance of a local ballot issue on increasing funding for local schools. Predispositions, in this sense, involve both issue-specific and loosely associated attitudes about school funding. In our model, predispositions and issue interest compose the intrapersonal filter, which impacts an individual’s selection of what is most important about a given issue. We include issue interest as an important component, as those who are more interested in an issue are more likely to express their own opinions relevant to the issue (Crano, ) and seek out other views via the media (Poindexter & McCombs, ). The intrapersonal filter is a powerful component of our model, as it impacts how individuals think about an issue. At this first filter, we hypothesize: H1a: Issue interest will be associated with respondents’ favorability toward the ballot issue. H1b: Issue interest will be associated with respondents’ opinion strength regarding the ballot issue.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
Demographic variables such as age, income, and political ideology could also help explain early differences in public opinion. Our model operationalizes demographic influence primarily as a means of statistical control, but their influence will also be assessed.
MEDIA FILTERS Public opinion has been defined as a collection of views regarding an issue that affects many (Corbett, ). Issue-relevant information often is delivered via mass media, which simultaneously act as a channel for information dissemination as well as another filter within the public opinion process; media outlets choose between many options when determining precisely which issues to cover. The direct impact of the media filter is that those who are exposed to this selected information have the opportunity to obtain more knowledge about an issue. Although a person may not perceive all issues reported in the media as having great personal importance, he or she may see an issue as being important to other people, creating ‘collective issue salience’ (Mutz & Soss, , p. ). In addition, ‘attitudes may shift when people learn of others’ views because knowing the opinions of others induces people to think of arguments that might explain those others’ positions,’ (Mutz, , p. ). Ultimately, media coverage that reflects the views of mass others may not directly change individuals’ opinions, but it can encourage them to reassess their own views in light of this new information. The more attention people pay to an issue, the more developed their own opinions are likely to become. In other words, issue-specific media attention will be associated with respondent favorability and opinion strength regarding the issue: H2a: The attention respondents pay to media coverage about the issue will be associated with respondents’ issue favorability. H2b: The attention respondents pay to media coverage about the issue will be associated with respondents’ opinion strength. Additionally, the ‘quasi-statistical sense’ of Noelle-Neumann’s () spiral of silence theory suggests that attention to media will be associated with perceptions of public opinion. Although her work is rooted in the ‘dual climate’ of public opinion, which presumes that media misrepresent public opinion (Scheufele & Moy, ), we hypothesize simply that the two will be associated: H2c: The attention respondents pay to media coverage about the issue will be associated with respondents’ perceptions of community support for the issue.
PUBLIC OPINION FILTERS
SOCIAL FILTERS: INTERPERSONAL DISCUSSION Discussion about an issue often occurs throughout the process, and like media use, can be influential in developing opinion strength. Political discussion is characterized by all kinds of political talk, as long as the conversation is voluntarily carried out without any specific agenda (Kim, Wyatt, & Katz, ). Research suggests that frequency of such discussion contributes to a number of politically desirable outcomes, such as greater political knowledge and participation (Scheufele, ; Kwak, Williams, Wang, & Lee, ;) as well as greater preparation for deliberation (Dutwin, ). Additionally, political discussion leads to higher-quality opinions through greater opinion strength (Kim et al., ), long hailed as an important consideration in evaluating public opinion (e.g. Allport, ; Schwarz, Groves, & Schuman, ). Political discussion is also a direct way for individuals to assess the views of those in their community. We expect frequency of discussion to influence one’s opinion and increase opinion strength: H3a: Frequency of discussion about the issue will be associated with respondents’ issue favorability. H3b: Frequency of discussion about the issue will be associated with respondents’ opinion strength. Similar to the logic presented for Hc, we propose that individuals who monitor their environment through interpersonal discussion will also develop perceptions of community support for the issue. As outlined by Scheufele and Moy (), media are one component of the quasi-statistical sense, while interpersonal discussion and observation are the other: H3c: Frequency of discussion about the issue will be associated with respondents’ perceptions of community support for the issue.
BRINGING
THE
PROCESS TOGETHER
In many mass communication and public opinion studies, individual factors, media content and use, and social contexts often are studied independently rather than as a process. However, these components should be seen as ongoing subprocesses that collectively lead to the formation of a public’s opinion about various issues (Van Leuven & Slater, ). Public opinion is inherently a social process in which individuals are continually adjusting their opinions through discussion and attention to media (Price & Roberts, ). One of the most prominent themes to emerge in the vast literature on public opinion is the role communication plays through discussion, debate, and collective decision-making (Price, ). Because of the central role of communication in this process, we incorporate both interpersonal and mass
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
communication—along with individual considerations—as key elements of the public opinion process. Moreover, we acknowledge that perceptions of public opinion can influence individuals’ opinions, thus contributing to the overall process of public opinion (e.g. Noelle-Neumann, ; Mutz & Soss, ; Moy, Domke, & Stamm, ). Thus, perceptions of community support for the issue—which we conceptualize as existing at the intersection of each filter— will be entered into the final model as a predictor of favorability toward the ballot issue and opinion strength: H4: Each filter (i.e. intrapersonal, media, and social) will contribute to the formation and strength of public opinion. Finally, we aim to examine the relative influence of each filter at both the introduction and culmination of the local issue: RQ1: Will the impacts of the three filters on opinion formation differ between Time 1 and Time 2?
METHODS We examined this process in the context of a proposed bond/levy issue for a suburban Midwestern school district in Columbus, Ohio. The district, which was comprised of a square-mile area at the time of the study, had recently experienced a substantial population influx, with the number of students growing from in to in .3 The proposed ballot issue sought financial support to build new schools, purchase land, and update buses, technology, and textbooks. Although some research has attempted to investigate public opinion longitudinally (cf., Shamir & Shamir, ), most rely on data collected at one point in time, reducing the study of public opinion to a one-shot, static view of individual opinions (e.g. Major, ; Moy, Domke, & Stamm, ). An additional problem is the rarity of having adequate lead-time to gather baseline measures of public opinion (Mutz & Soss, ). We were afforded this atypical opportunity through close attention to the proposed ballot issue before it came to public attention. In order to obtain a process-oriented understanding of public opinion, we measured public opinion at two different stages. Time (T) coincided with the introduction of the issue to the general public and occurred before a great deal of media coverage or interpersonal discussion took place. Data from this introductory period served as a baseline measure with which 3
Information obtained from data on the school district’s web site.
PUBLIC OPINION FILTERS
we could compare information obtained from our second time of data collection. Time (T) collection occurred—as we will demonstrate—after a significant increase in interpersonal discussion, issue awareness and interest, as well as shifts in perceptions of community support. In other words, we had a unique opportunity to examine early opinion development as well as the status of public opinion near its definitive focusing event—in this case, the vote on the ballot issue. By capturing two collections of survey data, we present antecedents, covariates, and outcomes of a public opinion process. PROCEDURE The main instrument was an independent-samples telephone survey collected at the beginning and end of the issue attention cycle surrounding a local ballot issue. A survey research center at The Ohio State University conducted these surveys using random digit dialing (RDD). The first set of data was collected during the last week of January , while the second set of data was collected between February and March , , just before the election.4 INDEPENDENT VARIABLES Respondents were asked about components of the proposed filters, including demographic variables, predispositions, media use, frequency of discussion, as well as perceptions of the opinions of others in the community.5 In addition to standard demographic questions regarding gender and level of education, we were interested in: () age, as older residents who are on a fixed income and no longer have school-aged children may be less likely to support tax increases; () number of school-aged children in the household, as those with children attending school have a vested interested in improved educational facilities and resources; () income, as those with higher incomes—and, most likely, higher property values—have more to lose with the introduction of new or increased taxes; and () political ideology, as those who are more fiscally conservative tend to favor less taxation, while those who are more liberal tend to support increased education spending. With these considerations in mind, it is necessary that we statistically control for these demographic characteristics. 4 The response rate was calculated using the American Association for Public Opinion Research Response Rate formula. For T, , randomly generated telephone numbers were dialed as many as times. Of these numbers, were found to be valid numbers, and interviews were completed in . percent of the cases. For T, valid numbers were found. From these households, interviews were completed in . percent of the cases. The margin of sampling error for each sample was . percent. Though the calculated response rate for our sample is somewhat low, recent work by Keeter, Miller, Kohut, Groves, & Presser () demonstrates that such response rates typically do not result in a significant loss of data quality. 5 All question wording can be found in the Appendix.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
TABLE Descriptive statistics of major variables Min–Max
Independent variables Relevant predispositions Issue awareness Issue interest Perceived community support Discussion frequency Issue-specific media use (Attention Exposure) Political ideology (Liberal—Conservative) Dependent variables Favorability toward issue Opinion strength
Time (T)
Time (T)
Mean
SD
Mean
SD
– – – – – –
. . . . . .
(.) (.) (.) (.) (.) (.)
. . . . . .
(.) (.) (.) (.) (.) (.)
–
.
(.)
.
(.)
– –
. .
(.) (.)
. .
(.) (.)
At the outset of the public opinion process, it is important to acknowledge that people hold certain predispositions toward most issues. These inclinations provide individuals with a starting point or anchor from which to begin forming their own opinions (Helson, ). We assessed relevant predispositions by asking respondents about the perceived benefits of the issue for their families and their community if the ballot issue were to pass. We also inquired about the anticipated benefits of the school district updating buildings and technology, as well as anticipated financial stress resulting from passage of the issue. In other words, we asked respondents to evaluate possible advantages and disadvantages of making improvements to schools in the local school district. After appropriate score reversing, a higher score on this index indicates a more positive predisposition toward school levies and related ballot issues. Seven items were averaged to create an index of relevant predispositions toward the ballot issue (Cronbach’s ¼ . at T and . at T). To gauge salience of the issue at the intrapersonal level, we asked respondents to estimate their own level of issue awareness and interest (see Table for all descriptive statistics). Respondents were also asked to indicate how many of the last seven days they were exposed to various media outlets—three local newspapers, one metro newspaper, and local television news (i.e. – scale for each outlet). Respondents were then asked to indicate how much attention they paid to stories regarding the ballot issue (from ¼ ‘none’ to ¼ ‘a lot’). The exposure scores were multiplied by the level of issue attention respondents reported to arrive at a measure of issue-specific media use. These scores were then averaged to create an index of issue specific media use.
PUBLIC OPINION FILTERS
We measured respondents’ perceptions of community support6 of the issue by asking, ‘If a poll were taken today, in your estimation, what would be the percentage of members of this community who would be in favor of the school levy?’7 The frequency with which respondents discussed the issue was assessed with a question that asked how often they talked to others about the issue. Specifically, we asked respondents, ‘How frequently do you talk to others about the proposed school levy?’ DEPENDENT VARIABLES The outcome variable measured how favorable respondents were toward the ballot issue. Response options ranged on a five-point scale from ‘very unfavorable’ to ‘very favorable,’ with higher values representing more favorability (Scale: –; MT ¼ ., SD ¼ .; MT ¼ ., SD ¼ .). We also asked respondents about their voting intentions if they were at least somewhat likely to vote. Because this issue reached its culmination at the ballot box, it might at first seem more appropriate to use the voting measure as the criterion variable. We are, however, interested in the process of public opinion, which includes those individuals who do not intend to vote. In order to measure opinion strength, we obtained each respondent’s deviation from the midpoint answer of support for the ballot issue (i.e. halfway between the most and least supportive options) and used this absolute value as a quantification of opinion strength. In other words, those who said they were ‘very supportive’ or ‘not at all supportive’ held strong opinions, while those who were ‘moderately supportive’ held relatively weaker opinions. With this measure, we are able to observe the strength of personal opinions (Scale: –; MT ¼ ., SDT ¼ .; MT ¼ ., SDT ¼ .). RESULTS The independent-samples telephone survey consisted of residents of the school district (NT ¼ , NT ¼ ). Fifty-eight percent of respondents were female, with an average reported age of . years (SD ¼ .). The mean household income was approximately $,. About percent reported having one or more children under in their household, with an average of . children (SD ¼ .). The sample also leaned toward 6 Perceived community support is used as a dependent variable in analyzing the subprocesses specific to media and social filters, as discussion and media use can impact these filters. Perceived community support is used as an independent variable in the overall model, which incorporates all three filters in predicting public opinion. 7 Research has suggested that asking for percentage estimates is more useful than asking whether ‘most’ people agree or disagree, because this method leads to artificial calculations of perceptions (Tipton, Prichard, & Prichard, ).
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
Time 1 Not at all Somewhat Very Time 2
0%
20%
40%
60%
80%
100%
Question wording: ‘How aware are you of the proposed ballot issue?’
Figure Levels of issue awareness at Times and d
a conservative political ideology. With a more conservative ideology coded as , moderate as , and liberal as , we calculated a mean ideology of . (SD ¼ .) across both survey points. As might be expected, issue awareness [t(.) ¼ ., p < .] and interest [t(.) ¼ ., p < .] significantly increased as the election approached (see Figures and ). Overall opinion strength was also significantly greater at T when compared with T, t(.) ¼ ., p < .. Ultimately, the ballot issue was passed by a slim margin in the actual vote. INTRAPERSONAL FILTER At the intrapersonal level, key demographic variables were associated with overall favorability toward the ballot issue at both times of the survey. As seen in Table , both age and length of residence were negatively associated with issue favorability. On the other hand, the number of children in a respondent’s household and a respondent’s interest in the issue were positively correlated with favorability toward the ballot issue. Though the observed correlations are not remarkably strong, it is worth noting that the relationship between several demographic variables and issue favorability differed between T and T of the survey. The correlation between income and issue favorability weakened and became nonsignificant between T and T. The relationship also weakened for level of education. Additionally, political ideology was less strongly correlated with issue favorability at T for both liberals and conservatives. This finding suggests 8 According to official Election Board results, , residents ( percent of registered voters) cast a vote on the levy/bond issue with (. percent) voting for and (. percent) voting against.
PUBLIC OPINION FILTERS
Time 1 Not at all Not Very Slightly Moderately Very Time 2
0%
20%
40%
60%
80%
100%
Question wording: ‘How interested are you in the proposed ballot issue?’
Figure Levels of issue interest at Times and
TABLE Bivariate correlations between issue favorability and intrapersonal filter variables
Age Gender Income Number of children Length of residence Education Liberal Conservative Issue Interest
Time
Time
.** . .** .** .** .** .* .* .**
.** .** . .** .** .* . . .**
Note: Except for issue interest, independent samples t-tests revealed no significant differences between listed variables at T and T. **p < ., *p < ..
that while some demographic attributes were connected to issue favorability at both times, there was temporal variation with other demographic traits. For income, education, and political ideology, the correlation with issue favorability existed at the outset of the issue, but weakened as the culmination of the issue approached. In other words, respondents seemed to rely more on certain personal constants when issue attention and interest were lower. By dummy coding for time (T ¼ ; T ¼ ) and restructuring our data to include both survey time points in one file, we will be able to further test the temporal variation of these and other key independent variables.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
TABLE Effect of intrapersonal filter (Ordinary least squares regressions) Respondent favorability b Relevant predispositions Issue interest Number of children in home Age Gender Education Income Identified as liberal Identified as conservative Time vs. Time Predispositions Time Issue interest Time Model fit
S.E.
.*** . .# . . . .* . . . . .# . . . . . . . . . . . . Adjusted R ¼ .
Opinion strength b
S.E.
. . .*** . . . .* . . . . . . . . . . . .* . . . . .# Adjusted R ¼ .
Note: All coefficients are unstandardized. ***p < ., **p < ., *p < ., #p < ..
Our first hypothesis stated that relevant predispositions would be associated with favorability and opinion strength toward the ballot issue. While controlling for key demographic variables listed above, relevant predispositions were found to be a strong predictor of favorability toward the ballot issue and a marginally important predictor of opinion strength. The overall demographic and predisposition model accounted for percent of the variance in predicting issue favorability, though this relationship changed little over time (i.e. between T and T; see Table ). The dummy coded time variable was not a significant predictor of issue favorability, nor did this variable contribute anything of significance to the model when interaction terms were created with time, issue interest and the index of relevant predispositions. While the influence of predispositions and issue interest is apparent, this relationship did not seem to change as the issue progressed, suggesting a temporally stable relationship between intrapersonal variables and issue favorability. The same predictors were used in a model estimating the opinion strength of respondents. Demographics and relevant predispositions accounted for about percent of the overall variance in respondents’ opinion strength, with issue interest and a measure of the effect of the passage of time significantly contributing to the model. There was also a marginally significant interaction between respondents’ interest in the ballot issue and the passage of time, indicating that opinions became stronger over time among those who were relatively more interested in the ballot issue. Demographics and relevant
PUBLIC OPINION FILTERS
TABLE Effect of media filter (ordinary least squares regressions) Respondent favorability
Issue specific media attention Time vs. Time Media attention Time Model fit
Opinion strength
Perceived community support
b
S.E.
b
S.E.
b
S.E.
.*
.
.**
.
.**
.
.
.
.
.
.**
.
.
.
.
.
.
.
Adjusted R ¼ .
Adjusted R ¼ .
Adjusted R ¼ .
Note: All coefficients are unstandardized. ***p < ., **p<., *p < ., #p < ..
predispositions seemed to have a stable effect on favorability toward the ballot issue as well. While it appears that there was a temporal moderation of the impact of issue interest on opinion strength, the overall influence of time at the intrapersonal filter appears to be minimal. MEDIA FILTER Although it was not surprising that overall media use was similar at T and T [t(.) ¼ ., p ¼ .], we found that issue-specific media use also did not increase between samples, t(.) ¼ ., p ¼ .. We anticipated an increase in issue-specific media use to help explain the previously discussed temporal increase in issue awareness and interest. Our second hypothesis predicted that respondents’ attention to media coverage of the issue would be linked with the strength of their favorability and opinion strength toward the ballot issue. In line with this prediction, those who reported that they paid relatively more attention to stories about the ballot issue had a significantly less favorable opinion of the ballot issue and significantly stronger opinions. Similar to the intrapersonal filter, there was no interaction between time and media use, indicating that the impact of the media filter did not significantly change between T and T (Table ). SOCIAL FILTERS The characteristics of interpersonal discussion also differed between T and T. There was a significant increase in the overall amount of conversation involving the ballot issue between periods of data collection, as borne out by an independent-samples comparison, t(.) ¼ ., p < .. Our third set
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
TABLE Effect of social filter (Ordinary least squares regressions) Respondent favorability
Discussion frequency Time vs. Time Discussion freq. Time Model fit
Opinion strength
Perceived community support
b
S.E.
b
S.E.
.
.
.**
.
.
.
.#
.
.**
.
.
.
.
.
.
.
Adjusted R ¼ .
Adjusted R ¼ .
b .
S.E. #
.
Adjusted R ¼ .
Note: All coefficients are unstandardized. ***p < ., **p < ., *p < ., #p < ..
of hypotheses predicted a relationship between interpersonal discussion of the issue and issue favorability, opinion strength, and perceptions of community support, as information obtained through the social filter is the most direct reflection of community opinion. Analysis revealed that frequency of discussion was not a significant predictor of issue favorability, even when controlling for the influence of time. Discussion frequency was significantly associated with opinion strength, as those who talked more about the issue held stronger opinions. There was however, no temporal interaction between discussion frequency and opinion strength. The association between discussion frequency and perceptions of community support was marginal, and there was no significant interaction between discussion and time in terms of impacting perceptions of community support. It appears that discussing the issue led to stronger opinions and the perception that the community was slightly less supportive of the issue. Because no significant interactions were found between variables within the social filter and time, we can discern that the influence of discussion for this issue was relatively stable over time. TESTING
THE
PROCESS MODEL
The final hypothesis predicted that each filter would contribute to the overall opinion formation process. This hypothesis tests our central thesis that each of these components will significantly contribute to the overall model. We also examined what changes in variance took place between T and T. To address this issue, we included key variables from each filter as predictors, as well as the dummy coded time variable (T ¼ , T ¼ ). We included three cross-filter interaction terms (e.g. discussion x media attention) to assess possible synergies between filters. While controlling for the influence of time,
PUBLIC OPINION FILTERS
TABLE Simultaneous effect of all filters (ordinary least squares regressions) Respondent favorability b Intrapersonal filter Time vs. Time Relevant predispositions Issue interest R (F value) Media filter Media attention R (F value) Social filter Discussion frequency Perceived community support R (F value) Interactions Predispositions Media attention Predispositions Discussion Media attention Discussion R (F value)
S.E.
Opinion strength b
S.E.
. . .*** . .* . . (.)***
.# . . . .*** . . (.)***
. .# . (.)#
.** . . (.)**
. .*
. .
. (.)#
.* .*
. .
. (.)**
.
.
.
.
.#
.
.
.
.
.
.
.
. (.)
. (.)
Note: All coefficients are unstandardized. ***p < ., **p < ., *p < ., #p < ..
the intrapersonal filter, media filter and social filter each contributed to an improvement in the fit of the model, though the significance of the media and social filters’ contributions were marginal. As seen in Table , elements of the intrapersonal filter by far accounted for the most variance. Within the media filter, an increase in issue-specific media use led to a marginal decrease in favorability toward the issue. Within the social filter, greater perceived community support led to a significant increase in favorability toward the issue. Of the three interaction terms, relevant predispositions and frequency of discussion combined to increase favorability toward the issue. Each of the three filters also impacted overall opinion strength. Greater issue interest led to increased opinion strength, as did increased media attention and discussion frequency. Perceived community support, on the other hand, was negatively associated with opinion strength, indicating that those who thought their community was more supportive of the issue held weaker opinions than those who thought the community was less supportive of the issue.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
Research question addressed the possibility of differential impacts of the three filters on opinion formation between Time and Time . We assessed this by building interaction terms consisting of each of the key predictor variables and the dummy coded time variable. Analysis revealed that none of the three filters significantly interacted with the time variable in explaining additional variance in respondents’ favorability toward the ballot issue. Possible reasons for this lack of a temporal effect are discussed below.
DISCUSSION This study examined public opinion at two key points during the progression of a community issue. Using OLS regression for our main analyses, we examined a variety of influences that contributed to public opinion. We did this by examining three filters—intrapersonal, media, and social—that affect the public opinion process. Moreover, we incorporated communication variables throughout our process model in response to calls for examining communication as a central component of this process (e.g. Davison, ; Price & Roberts, ; Glynn, ). The implications of each of the three filters are detailed below. At the intrapersonal level, we found that relevant predispositions and issue interest were the best predictors of issue favorability. In examining subprocesses of the intrapersonal filter, we found that demographic characteristics were correlated with issue favorability, though there was variability between periods. Future research should further test this finding, which suggests that individual considerations could be more salient at the introduction of an issue, but become less important as the issue becomes more public. It is possible that when issue culmination approaches, predominant information and influence processes shift away from the internal (i.e. intrapersonal) toward the external (i.e. media use, discussion). Though our regression analyses do not bear this conclusion out for this particular study, it is possible that this lack of temporal influence is attributable to the issue at hand. In many communities, school levy and bond issues are a recurrent phenomenon, and opinions may be relatively stable from the outset. Though the intrapersonal filter held strongest sway over respondents’ issue favorability, there were still significant impacts to be found within the media and social filters. Interestingly, the more attention respondents paid to media reports regarding the ballot issue, the less likely they were to favor the passage of the issue. At the same time, increasing media attention also led to stronger opinions, and these relationships persisted even when matched against other key variables in a multiple regression context. Our study also reveals interesting results regarding the influence of interpersonal discussion
PUBLIC OPINION FILTERS
on perceived community support, such that respondents who talked more frequently to others about the issue perceived less community support for the issue (although this was marginally significant). These results suggest that, at least in the present example, media use and discussion frequency led people to perceive less community support for the ballot issue, even though the majority of voters agreed to pass the initiative. Future research should more closely examine how communication influences such perceptions. While we must acknowledge that the intrapersonal filter explains a large portion of the variance in the present study, this is not an unexpected or unprecedented result. Much political science research, for example, has claimed that – percent of voting behavior variance is explained by party identification alone (e.g. Finkel, ). What many researchers aim to uncover is the elusive – percent of unexplained variance, even if it is smaller in comparison with what is explained by intrapersonal factors, such as party identification. This research intended to explain the additional variance accounted for by other factors—namely communication—that have an effect on the variance in opinion. Because our work is rooted in that of Davison () and Price and Roberts (), among others, it was essential to parse out the effects of communication, which drives the public opinion process. Indeed, we did find that communication had a significant effect on this process. Though the variance explained by the intrapersonal filter was indeed largest, the contribution of communication was not insignificant. In fact, our analyses could be considered a conservative approach; future research might explore whether the variance explained by the intrapersonal filter might be ‘activated’ by information and communication (Finkel, ; Gelman & King, ), and thus explained by media and interpersonal filters after all. We should acknowledge limitations of our survey. Media attention to the issue, as revealed by respondents’ self-reports, was relatively low. Had the survey taken place during a statewide or presidential election, we might have seen more of a difference between respondents who were attending or not attending to the campaign through media use and discussion. Findings also indicate that opinions regarding the ballot issue may have been relatively stable between the two survey points (T and T). It was our attempt to survey a group of people who had little familiarity with an issue. While there were obvious impacts of media and social filters over time, it is possible that a more novel or rare issue would generate more temporal variance in terms of predicting issue favorability and opinion strength. It is in such circumstances that change over time is best assessed. This balance between obtaining a widely varying sample and catching respondents before they have been exposed to a stimulus is a paradox faced by many survey researchers, particularly those interested in the media. We feel that our approach provides
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
insight into the public opinion process, as well as a framework from which other researchers can draw in developing a more holistic model of the public opinion process. Although our sample was representative of the community we were studying, it was certainly not representative of all communities, as respondents in the community tended to be wealthy, white, and conservative. For this reason, the community could have been somewhat homogeneous in beliefs, predispositions, and opinions, making it difficult to generalize the results to a larger population. Furthermore, we examined opinions and perceptions about only one issue. Results may have varied if the issue were, for example, a presidential election, which generates more media coverage and interpersonal discussion. And finally, although the present cross-sectional independentsamples survey serves as an improvement over previous cross-sectional studies, it should be noted that a panel study—examining the same group of people from T to T—could more directly assess causality.
CONCLUSION By surveying the public at two points in time, and including time as a key independent and moderating variable, we believe our results serve as robust support for our model of the public opinion process. Our regression model, which incorporated three operationalized filters of public opinion, also included interaction terms that captured the important synergistic effects of the combinations of those filters. Myriad factors influence the public opinion process, and these factors can play a different role at different times the process. In-depth analyses of each component are invaluable contributions to the literature and should continue. In order to obtain a more holistic view of the public opinion process, however, we combined those subprocesses into a more comprehensive model. In this sense, we were able to make conclusions about the relative effects of each filter or component of the public opinion process while adequately controlling for other factors. Results indicate that individual characteristics are dominant during the formative phase of the public opinion process. Yet as time goes by, increasing media coverage and more frequent discussion may exert a stronger influence on public opinion. In furthering the filter analogy, we might suggest that as media and social influence are introduced, the intrapersonal filter becomes more ‘porous,’ such that predispositions and demographics have less of an impact on the public opinion process when media and social filters are taken into account. Although there are undoubtedly multiple sources of influence, at its core, public opinion is rooted in communication processes. As Price () succinctly argued, ‘Whether viewed in philosophical, political, sociological,
PUBLIC OPINION FILTERS
or psychological terms, [public opinion] remains fundamentally a communication concept’ (p. ). The present study not only viewed public opinion as a communication concept, but also tested the unique aspects of communication in an advanced model to parse out competing effects. Future research should apply this model to broader events and more heterogeneous communities to test whether these findings hold true.
REFERENCES Allport, G. W. (). Personality: A psychological interpretation. New York: Social Science Research Council. Asch, S. E. (). Studies of independence and conformity: A minority of one against a unanimous majority. Psychological Monographs, , . Blanton, H., & Christie, C. (). Deviance regulation: A theory of action and identity. Review of General Psychology, , –. Bryce, J. (). The American commonwealth (Vol. III). London: Macmillan & Co. Corbett, M. (). American public opinion: Trends, processes, and patterns. New York: Longman. Crano, W. (). Vested interest, symbolic politics, and attitude-behavior consistency. Journal of Personality and Social Psychology, , –. Crespi, I. (). The public opinion process: How the people speak. Mahwah, NJ: Erlbaum. Davison, W. P. (). The public opinion process. Public Opinion Quarterly, , –. Davison, W. P. (). A public opinion game. Public Opinion Quarterly, , –. Downs, A. (). Up and down with ecology: The issue attention cycle. The Public Interest, , –. Dutwin, D. (). The character of deliberation: Equality, argument, and the formation of public opinion. International Journal of Public Opinion Research, , –. Finkel, S. E. (). Reexamining the ‘minimal effects’ model in recent presidential campaigns. Journal of Politics, , –. Foote, N. N., & Hart, C. W. (). Public opinion and collective behavior. In M. Sherif & M. O. Wilson (Eds.), Group relations at the crossroads (pp. –). New York: Harper & Bros. Gelman, A., & King, G. (). Why are American presidential election campaign polls so variable when votes are so predictable? British Journal of Political Science, , –. Glynn, C. J. (). Public opinion as a social process. In S. Dunwoody, D. McLeod, L. B. Becker & G. M. Kosicki (Eds.), The Evolution of key mass communication concepts (pp. –). Cresskill, NJ: Hampton Press. Glynn, C. J., & McLeod, J. M. (). Implications of the spiral of silence theory for communication and public opinion research. In K. R. Sanders, L. L. Kaid & D. Nimmo (Eds.), Political Communication Yearbook (pp. –). Carbondale, IL: Southern Illinois University Press.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
Helson, H. (). Adaptation level theory: An experimental and systematic approach to behavior. New York: Harper & Row. Keeter, S., Miller, C., Kohut, A., Groves, R. M., & Presser, S. (). Consequences of reducing nonresponse in a national telephone survey. Public Opinion Quarterly, , –. Kim, J., Wyatt, R. O., & Katz, E. (). News, talk, opinion, participation: The part played by conversation in deliberative democracy. Political Communication, , –. Kinder, D. R. (). Communication and opinion. Annual Review of Political Science, , –. Kwak, N., Williams, A. E., Wang, X. R., & Lee, H. (). Talking politics and engaging politics: An examination of the interactive relationships between structural features of political talk and discussion engagement. Communication Research, , –. Major, A. M. (). Correlates of accuracy and inaccuracy in the perception of the climate of opinion for four environmental issues. Journalism and Mass Communication Quarterly, , –. Moscovici, S. (). Social influence and conformity. In G. Lindzey & E. Aronson (Eds.), The handbook of social psychology (Vol. , p. –). New York: Random House. Moy, P., McCoy, K., Spratt, M., & McCluskey, M. R. (). Media effects on public opinion about a newspaper strike. Journalism and Mass Communication Quarterly, , –. Moy, P., Domke, D., & Stamm, K. (). The spiral of silence and public opinion on affirmative action. Journalism & Mass Communication Quarterly, , –. Mutz, D. C. (). Mechanisms of momentum: Does thinking make it so? The Journal of Politics, , –. Mutz, D. C., & Soss, J. (). Reading public opinion: The influence of news coverage on perceptions of public sentiment. Public Opinion Quarterly, , –. Noelle-Neumann, E. (). Return to the concept of the powerful mass media. Studies of Broadcasting, , –. Noelle-Neumann, E. (). The spiral of silence: Public opinion–Our social skin. Chicago: University of Chicago Press. Pan, Z., & Kosicki, G. M. (). Assessing news media influences on the formation of whites’ racial policy preferences. Communication Research, , –. Poindexter, P. M., & McCombs, M. E. (). Revisiting the civic duty to keep informed in the new media environment. Journalism & Mass Communication Quarterly, , –. Price, V. (). Social identification and public opinion: Effects of communicating group conflict. Public Opinion Quarterly, , –. Price, V. (). Public Opinion. Thousand Oaks: Sage. Price, V., & Roberts, D. F. (). Public opinion processes. In C. Berger & S. Chaffee (Eds.), Handbook of communication science (pp. –). Newbury Park, CA: Sage.
PUBLIC OPINION FILTERS
Scheufele, D. A. (). Talk or conversation? Dimensions of interpersonal discussion and their implications for participatory democracy. Journalism and Mass Communication Quarterly, , –. Scheufele, D. A., & Moy, P. (). Twenty-five years of the spiral of silence: A conceptual review and empirical outlook. International Journal of Public Opinion Research, , –. Schwarz, N., Groves, R. M., & Schuman, H. (). Survey methods. In D. T. Gilbert, S. T. Fiske & G. Lindzey (Eds.), The handbook of social psychology (th ed., Vol. , p. –). Boston: McGraw-Hill. Shamir, J., & Shamir, M. (). The anatomy of public opinion. Ann Arbor: The University of Michigan Press. Tipton, L., Prichard, P., & Prichard, N. (1987, August). Public opinion as collective coorientation. Paper presented at the meeting of the Association for Education in Journalism and Mass Communication, San Antonio, TX. Tsfati, Y. (). Media skepticism and climate of opinion perception. International Journal of Public Opinion Research, , –. Van Leuven, J. K., & Slater, M. D. (). How publics, public relations, and the media shape the public opinion. Public Relations Research Annual, , –. Zaller, J. (). The nature and origins of mass opinion. Cambridge, England: Cambridge University Press.
APPENDIX: SURVEY QUES TION AND RESPONSE WORDING ISSUE AWARENESS, INTEREST,
AND
FAVORABILITY
How aware are you of the school levy on the March ballot that was proposed by the Olentangy Board of Education in order to build three elementary schools and a middle school, as well as purchase buses, textbooks and technology? Would you say you are you very aware, slightly aware, or not at all aware?—Very aware; Slightly aware, Not at all aware; Refused; Don’t know. Regardless of your awareness of the issue, how interested would you say you are in the proposed school levy? Would you say you are very interested, moderately interested, slightly interested, not very interested, or not at all interested?—Very interested; Moderately interested; Slightly interested; Not very interested; Not at all interested; Refused; Don’t know. And when you think about others in your community, how interested do you think they are in this levy issue? Would you say they are very interested, moderately interested, slightly interested, not very interested, or not at all interested?—Same options as above. How favorable are you toward the school levy? Would you say you are very favorable, favorable, neutral, unfavorable, or very unfavorable?—Very favorable; Favorable; Neutral; Unfavorable; Very unfavorable; Refused; Don’t know.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
When you think about others in your community, how favorable do you think they are toward this levy? Would you say they are very favorable, favorable, neutral, unfavorable, or very unfavorable?—Same options as above. How likely is it that you will vote on this issue when it appears on the ballot March nd? Is it very likely, somewhat likely, somewhat unlikely, or very unlikely?—Very likely; Somewhat likely; Somewhat unlikely; Very unlikely; Refused; Don’t know.
PREDISPOSITIONS How beneficial do you think the passing of this levy would be to you and your family? Would it be very beneficial, moderately beneficial, slightly beneficial, hardly beneficial, or not at all beneficial?—Very beneficial; Moderately beneficial; Slightly beneficial; Hardly beneficial; Not at all beneficial; Refused; Don’t know. How beneficial do you think the passing of this levy would be for your community? Would it be very beneficial, moderately beneficial, slightly beneficial, hardly beneficial, or not at all beneficial?—Same options as above. How likely is it that the proposed levy would cause financial stress for you and your family? Is it very likely, somewhat likely, somewhat unlikely, or very unlikely?—Very likely; Somewhat likely; Somewhat unlikely; Very unlikely; Refused; Don’t know. The School Board has proposed the levy to build three elementary schools and a middle school, as well as purchase buses, textbooks and technology. In your opinion, how likely is it that the designated money will be used for the proposed school construction, updated materials and new buses? Would you say it is very likely, somewhat likely, somewhat unlikely, or very unlikely?— Same options as above. How important do you think it is to improve school facilities by updating existing structures and building new schools? Would you say it is very important, somewhat important, somewhat unimportant, or very unimportant?—Very important; Somewhat important; Somewhat unimportant; Very unimportant; Refused; Don’t know. How important do you think it is to update technology and classroom materials in schools? Would you say it is very important, somewhat important, somewhat unimportant, or very unimportant?—Same options as above. How important do you think it is to have more buses to transport students to school? Would you say it is very important, somewhat important, somewhat unimportant, or very unimportant?—Same options as above.
PUBLIC OPINION FILTERS
ISSUE SUPPORT If a poll were taken today, in your estimation what would be the percentage of members of this community who would be [in favor of/opposed to] the school levy?—Open, no options. Do you think that, overall, people in this community have recently changed their opinion to be [more/less] favorable on this issue?—Yes; No; Refused; Don’t know. How likely are you to vote [against/in favor of] the passing of the Olentangy school levy on March nd? Would you say it is very likely, somewhat likely, somewhat unlikely, or very unlikely?—Very likely; Somewhat likely; Somewhat unlikely; Very unlikely; Refused; Don’t know.
MEDIA USE The next few questions will be about how much attention you pay to local newspapers and television. I will read a list of different media. For each one, please tell me how many days out of the last seven days you have read the newspaper or watched the news. Now I will ask you how much attention you pay to these news sources in general. Please respond after I read this list again and tell me if you pay a lot, some, a little, or no attention to each source.— A lot; Some; A little; None; Refused; Don’t know. Regarding stories about the Olentangy school levy, tell me if you pay a lot, some, a little or no attention to stories about the levy in these news sources. [List of local media sources.]
ISSUE DISCUSSION How frequently do you talk to others about the proposed school levy? Would you say you talk about it very frequently, somewhat frequently, somewhat infrequently, very infrequently, or never?—Very frequently; Somewhat frequently; Somewhat infrequently; Very infrequently; Never; Refused; Don’t know.
DEMOGRAPHIC INFORMATION In what year were you born? Including yourself, how many adults, years or older, live in your household most of the year? How many children, years of age or younger, live in your household? How many months have you lived in your current residence? Do you own or rent your residence?—Own; Rent; Refused; Don’t know
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
Now we’d like to ask you about your education. What is the highest grade or year of school you have completed?—Elementary school; High school; Some college; Associates certificate; Bachelor’s degree; Some graduate school; Master’s degree; Doctorate/advanced degree; Refused; Don’t know. These days, many people are so busy they can’t find time to register to vote, or move around so often they don’t get a chance to re-register. Are you currently registered to vote in your precinct or election district?—Yes; No; Refused; Don’t know. When it comes to politics, some people think of themselves as liberal, and others think of themselves as conservative. How would you describe yourself, are you . . . —Liberal; Conservative; Moderate/middle of road/neither; Refused; Don’t know. And, approximately what was your total household income from all sources, before taxes for ? BIOGRAPHICAL NOTE Lindsay Hoffman is a doctoral candidate in the School of Communication at the Ohio State University and will be assistant professor at the University of Delaware beginning Fall . Carroll Glynn is a professor and director of the School of Communication at Ohio State. Michael Huge is a research associate at Ohio State. Tiffany Thomson is a doctoral candidate at Ohio State. Rebecca Border Sietman is an assistant professor of Communication Arts and Director of Debate at Cedarville University. Address correspondence to Carroll Glynn, School of Communication, B Derby Hall, N. Oval Mall, Columbus, OH, USA, E-mail: glynn.@osu.edu
