message_distortion.htm 

 

Message Distortion: The Effects of Media Content, Perceptions of Media Bias, and Partisan Predispositions on Candidate Evaluations

 

Jeanette Morehouse Mendez

Assistant Professor

Oklahoma State University

Department of Political Science

 519 Mathematical Sciences

Stillwater, OK. 74075

Office: (405) 744-4477

Fax: (405) 744-6534

jeanette.mendez@okstate.edu

 

Online Publication Date: April 21, 2007

Journal of Media Psychology, Volume 12, No. 2, Spring 2007

  

 

Abstract

Candidate evaluations are explored within a motivated reasoning framework, as a product of media content, perceptions of media bias, and partisanship. Based on motivated reasoning, information processing should be biased to arrive at predispositional goals (Pyszczynski and Greenberg, 1987).  Through a computerized experiment, respondents read either balanced or disparate media content concerning two congressional candidates and their views on campaign finance reform.  Respondents then evaluated the candidates and the media content as regards their perceptions of possible media bias and the candidates’ partisanship.  The results show that strong information processing biases exist and vary across partisan groups and types of messages. Both Democrats and Republicans are susceptible to information processing biases when they perceive media bias and the media present disparate viewpoints.  Further, both conditions strongly affected the evaluations of the Democratic candidate, but they affected evaluations of the Republican candidate to a lesser extent.  The findings suggest that media biases might not have the anticipated effect on public opinion that common wisdom assumes.

  

 

Introduction

Individuals on both ends of the political spectrum claim that the media are biased against them.  While political scientists, media outlets, and a variety of pundits have investigated the existence of media bias (e.g., D’Alessio and Allen, 2000), they also have researched how the public comes to form perceptions of media bias (e.g., Baron, 2004), as well as when such perceptions are likely to occur (e.g., Vallone, Ross and Lepper, 1985). The present study investigates a new area based on these previous investigations: the effect of both media content and perceptions of bias on individuals’ political decision-making. 

Drawing on information processing models, specifically motivated reasoning, the study examines the effects of media content and perception of bias on how individuals make candidate evaluations.  With motivated reasoning, individuals’ information processing functions with bias to arrive at predispositional goals (Pyszczynski and Greenberg, 1987).  It is important, both practically and theoretically, to understand whether certain media messages are more believable than others and whether people receive such messages verbatim or alter them, as suggested by the motivated reasoning approach.  Further, given the existing evidence that the public perceives media bias, it is important to examine how these perceptions affect political decision-making.  Understanding the processes by which individuals distort media messages and their effects on decision making will help us to better understand and media influences on public opinion.

To address these issues, the analysis will proceed first by considering the literature concerning perceptions of media bias.  Second, information processing models will be used as the basis for understanding perceptions of bias and the effects of media content and bias on individuals’ candidate evaluations.  Next, an experimental research design is introduced and a model of candidate evaluations as a function of media content, perceptions of media bias, and partisanship is tested. 

Biases: Media Content, Perception, and Information Processing

The concept of media bias has received much attention in recent months, not only in academic publications but also in the mainstream media.  Media bias results when media outlets, editors, or journalists advance a particular viewpoint based on ideology, partisanship, or personal preferences (Baron, 2004).  Therefore, while bias generally refers to a prejudice toward one point of view, media bias exists because of which issues are covered and how they are covered.  Specifically, researchers generally have concurred on three possible sources of media bias: gatekeeping bias, coverage bias, and statement bias (D’Alessio and Allen, 2000). 

In gatekeeping bias, partisan bias results when editors or journalists select stories that favor one party over another.  This means certain issues are covered and certain issues are not covered.  In coverage bias, partisan bias occurs when one party receives more coverage than the other group, in terms of words, lines, paragraphs, and pictures.  This form of bias manifests when one group receives more actual space in the story than another group.  Lastly, in statement bias, partisan bias appears when one party receives coverage that is more favorable relative to the other party.  Here, bias results from actual coverage that is more favorable to one group. 

While these definitions might appear similar, they are distinct.  For example, the coverage of certain issues, gatekeeping bias, even when bias is the intent, does not have to result in coverage that is more favorable to one party.  Further, a party can receive coverage that is more favorable without receiving more space in a story.  This could happen if a story focuses more attention to the negative aspects of one party and mentions a few positive traits of the other party.  The party receiving coverage that is more positive could actually receive less space.  

While researchers have agreed that different sources of bias exist, they have disagreed about how bias should be measured.  This has led to inconsistent and sometimes contradictory findings.  Much of the debate and inconsistent findings have been due to the subjectivity involved in content analysis (e.g. Efron, 1971), which yields differing accounts of media bias from both sides of the political spectrum (Groeling and Kernell, 1998).  For example, some scholars measure statement bias by counting the number of times that one party receives positive coverage and another receives negative coverage (e.g. Stempel and Windhauser, 1991).  Others do not regard this as media bias and instead suggest there could be objective reasons why one party receives the more favorable coverage as, for example, when one candidate runs a better campaign (e.g., Dalton, Beck, and Huckfeldt, 1998). 

Due to such measurement problems, research findings have been extremely mixed.  Some studies have documented a liberal media bias (e.g., Efron, 1971; Keely, 1971; Bozell and Baker, 1990), while others have found a conservative one (e.g., Cooper and Soley, 1990; Lee and Solomon, 1991; Liebling, 1964).  Still others have shown that, on average, media content is devoid of bias.  For example, Hofstetter (1976) analyzed television coverage of the 1972 presidential election and found that negative points, on average, were balanced by positive ones and coverage was neutral.  Similarly, D’Alessio and Allen (2000) conducted a meta-analysis of 59 qualitative studies of media bias and found no significant bias in newspaper coverage.  Further, Niven (2003) compared newspaper coverage of members of Congress who switched political parties and found the stories for each party to be similar and devoid of bias.

Therefore, while the possible sources of media bias have been defined, differences in measurement and interpretation of these sources have led to inconsistent research findings.  A common thread in the research has been that there is a potential for the media to be biased.  Given this potential, and public discourse about media bias, it makes sense to examine how individuals come to interpret media bias and make political decisions.

Information Processing and Perceptions of Media Bias

While research has produced mixed results about the level of media bias, U.S. citizens seem to believe that media bias does exist, prompting researchers to examine individuals’ perceptions of media bias and how they process media information.  One survey revealed that 78% of the public perceives the press to be biased, and many attribute the cause of media bias to the media’s intention to persuade (ASNE, 1999).  Interestingly, this same survey revealed that the public is not very concerned with the media’s attempts to persuade, “perhaps because they believe they’ve built sufficient filtering mechanism to identify and neutralize it [i.e., media bias] when they think they see it” (ASNE, 1999, 2). 

Research has examined individuals’ perceptions of media content and identified specific situations in which individuals do perceive biases.  For example, strong partisans view the media as opposing their viewpoints or as being hostile to them (e.g., Arad and Carnevale, 1994; Beck, 1991; Dalton, Beck, and Huckfeldt, 1998; Giner-Sorolla and Chaiken, 1994; Gunther, 1992; Gunther, Christen, Liebhart, and Chia, 2001; Vallone et al., 1985).  In the first experiment to explore the “hostile media effect,” Vallone, Ross, and Lepper (1985) found that after viewing a videotape of the 1982 Beirut massacre, pro-Arab and pro-Israeli students reported that the tape and the editorial staff were biased in favor of the opposition and that neutral viewers would be influenced toward the opposition if they viewed the tape. 

In addition to the hostile media effect, three additional psychological theories can help explain perceptions of media bias.   First, according to social judgment theory, individuals process information relative to their own viewpoints and accept information that is congruent and supportive and reject disconfirming information (Sherif, Sherif and Nebergall, 1965).  Beyond rejecting information, information can also be distorted to fit with our categories of judgment (Sherif, Sherif and Nebergall, 1965).  In terms of perceptions of bias, social judgment theory would predict that individuals would project biases on sources that did not support their ideology.  And in fact, this is what D’Alessio (2003) found-- subjects regarded stories as biased if the stories opposed their own viewpoint. 

Second, and related, cognitive dissonance can be used to better understand perceptions of media bias and information processing.  Festinger (1957) first introduced cognitive dissonance to explain how and why people rationalize information to make information consistent with their beliefs, thus reducing dissonance.  When new information is presented and contradicts a person’s prior beliefs, he or she attempts to reduce the conflict, sometimes by projecting a bias onto the media source.  Similar to social judgment theory, under cognitive dissonance, one would expect an individual to project a bias onto a media source that does not support the individual’s viewpoint. Lastly, and again related, attribution theory can be used to understand why individuals might perceive media bias.  Attribution theory addresses how people explain things (Heider 1958).  The assumption is that people will interpret their environment in a way to maintain a positive self-image.  In doing so, individuals could perceive biases to maintain this positivity.

What has received less research attention is what effect these perceptions of bias have on individuals’ opinions and actions.  In an early examination of perceptions of media bias, Stevenson and Greene (1980) asked students to react to both favorable and unfavorable news stories concerning the 1976 presidential candidates.  The results showed that when students perceived the media reports as biased, they responded more strongly to them.  While the content of these reports could be considered biased by some definitions and unbiased by others, the results showed that media content elicited perceptions of bias and these perceptions then affected the number of responses students made about the stories.  The question remains as to what effect these perceptions have on individuals’ political decision-making.

 

Information Processing and Candidate Evaluations

            The same models used to understand perceptions of bias could also be used to examine the effects of these perceptions.  In particular, motivated reasoning is employed to model individuals’ decision-making, namely their candidate evaluations, as a function of media content and perceptions of media bias.  Individual decision-making refers in this context to evaluation made for a like-minded candidate or an opposition candidate.

The theory of motivated reasoning explains how information processing is biased in a manner consistent with individuals’ predispositional goals (e.g. Baumeister and Newman, 1994; Lodge and Taber, 2000; Pyszczynski and Greenberg, 1987).  Motivated reasoning is most prevalent when preexisting attitudes are either challenged or strongly held, when information is low, when attention is short, or when a person is distracted or not motivated to be accurate (Lodge and Taber, 2000).  Therefore, partisans evaluate information in line with their partisan goals when forming candidate evaluations and they will be unlikely to regard consistent information as biased.

            When reaching decisions consistent with partisan goals, individuals seek out information that supports these goals.  The information can be either congruent or incongruent with one’s preferences, and both types of information can encourage motivated reasoning.  Congruent information readily supports prior predispositions, while incongruent information causes motivated reasoning because individuals must counter-argue the incongruent information so that it supports prior predispositions (Lodge and Taber, 2000).

            The tone of messages helps individuals identify congruent and incongruent information.  Tone refers to the positive or negative mood conveyed by the author.  Kahn and Kenney (2002) show that new stories can vary in tone, and not all stories are neutral.  For example, in their analysis of campaign stories, the authors classify tone as favorable toward one candidate or unflattering toward another candidate.  For the purposes of this analysis, tone will refer to the positive or negative mood established by the journalist.  For example, is the content positive or negative in describing the candidate?

Tone is considered in the analysis because tone can affect evaluations.  Previous research has shown that partisans use negative information to maintain consistent, unfavorable evaluations of the opposition (e.g., Fishle, 2000; Goren, 2002; Stoker, 1993).  Specifically, Sweeney and Gruber (1984) and Goren (2002) found that negative trait descriptions were more important than corresponding positive trait descriptions when partisans formed candidate evaluations.  According to Lau (1982), negative information appears more prominent and stands out relative to positive information (i.e., the figure ground hypothesis).  Therefore, individuals will be more likely to use negative information rather than positive information to support partisan goals.   

Combining congruency with tone suggests that negative information should invoke motivated reasoning both when the negative information is congruent (e.g., negative information about the opposition) and incongruent (e.g. negative information about the like-minded candidate) because motivated reasoners may increase their support of a positively evaluated candidate upon learning new negative information about the opposition candidate (Lodge and Taber, 2000).  Based on this view, the first hypothesis is as follows:

Hypothesis 1:  Negative, incongruent information will lead to more favorable likeminded candidate evaluations, while negative, congruent information will lead to more unfavorable opposition candidate evaluations.

            In addition to a message’s congruency and tone, its content is important. While previous research of the hostile media effect examined balanced media coverage (Vallone, Ross, and Lepper, 1985), not all coverage is balanced.  In terms of campaigns, generally candidate stake out opposing viewpoints on key issues.  How the media covers these situations is not always balanced.  Information transmitted via the media about two candidates can be disparate, where one candidate is presented positively and the other candidate is presented negatively.  Given this, disparate information should cause motivated reasoning because the conflicting viewpoints would enable individuals to quickly seek out information to support prior predispositions.  Balanced media coverage should be less likely to cause motivated reasoning because individuals will have to actively search through the content to ascertain the meaning and find supportive information.  That said, the second hypothesis is as follows:

Hypothesis 2: Favorable like-minded candidate evaluations and unfavorable opposition candidate evaluations will be strongest when the media content is disparate.

Lastly, and the topic that has been given less previous research attention, is the impact of perceptions of media bias on individuals’ candidate evaluations.  Despite the lack of previous research, motivated reasoning provides a broad framework indicating that individuals will use information to support partisan goals.  Therefore, an individual’s perception of bias for the opposition should be viewed as incongruent and negative information, leading to unfavorable opposition candidate evaluations.  Similarly, a perception of media bias for the like-minded candidate is treated as positive, congruent information, where individuals would favor media content that supports their candidate.  This situation should lead to favorable, like-minded evaluations.  More concisely:

Hypothesis 3: Perceptions of media bias for the opposing candidate will lead to more unfavorable opposition candidate evaluations, while perceptions of bias for the like-minded candidate will lead to more favorable like-minded candidate evaluations.

 

Methods

The present analysis examines the effects of media content and perceptions of media bias on individuals’ evaluations of two candidates who show support for the same issue or different viewpoints that result in disparate coverage.  The various options of balanced or disparate coverage are presented in four scenarios:

·        Democrat supports side A and Republican supports side B

·        Democrat supports side B and Republican supports side A

·        Both candidates support side A

·        Both candidates support side B

A controlled, computerized experiment was designed to examine individuals’ evaluations of hypothetical candidates, given certain expectations and content scenarios.  The experiment was conducted in the fall of 2001, with a sample of 325 undergraduate students from a large, Midwestern university who participated in the study for extra credit.  The survey contained 6 knowledge questions, 4 party identification questions, a fabricated newspaper article, 5 questions designed to measure the respondents’ evaluations and knowledge of the newspaper article, 14 candidate trait evaluation questions, and 10 questions designed to assess the favorability of the candidates, each with follow-ups regarding the intensity of the respondents’ evaluations.

The students were randomly assigned to read one of the four articles.  All of the articles had the same, simple premise: Two candidates were running for Congress (a Democrat and a Republican), and each had a particular viewpoint concerning campaign finance reform as it concerns political action committees.  The content of each article was different, reflecting each of the four scenarios described above.  One of the disparate articles presented one candidate supporting side A with the other supporting side B; the balanced articles presented both candidates supporting side A and side B, respectively.

The positive content and the negative content, which established tone, were the same.  For the positive content, the candidate was identified as a moral citizen who: a) supported campaign finance reform; b) did not accept donations from political action committees; and c) was not “bought” by corporate interests.  For the negative content, the candidate was portrayed as: a) being beholden to corporate interests; b) having amassed a significantly large war chest; and c) opposing campaign finance reform. 

The positive information was classified as that related to support for campaign finance reform and the negative information as that related to opposition to reform for three reasons.  First, the experiment occurred in fall 2001, following the terrorist attacks, when many believed that the political mood was one of bipartisanship.  Also, at this time, the McCain-Feingold Act, a bipartisan campaign reform act, was rapidly gaining in popularity. Second, the articles were clearly written in language intended to convey positive and negative cues.  For example, the phrase, “taking back power from special interests” would have a negative connotation for those opposed to reform.   Lastly, and most importantly, a separate sample of 119 undergraduate students read sections of the articles, with the candidate names and party identifications omitted.  For each, they identified the content as positive or negative.  The results show that 87% of the respondents identified content where the candidate supports campaign finance reform as positive and 97% reported content where the candidate opposes campaign finance reform as negative. 

For these three reasons, information is considered positive when the candidate supports reform and negative when the candidate opposes it.  This creates two balanced situations, in which each candidate has the same viewpoint, and two disparate situations, with one favoring the Democratic candidate (the Democratic candidate receives positive coverage and the Republican candidate receives negative coverage), and one favoring the Republican candidate (the Republican candidate receives positive coverage and the Democratic candidate receives negative coverage). 

The remaining sections of the survey provide the dependent and independent variables for the analysis. 

 

Dependent Variables

The respondents were asked to report whether they favored or opposed each candidate, a common approach in candidate evaluation research (e.g., Lavine, 2001).  The results of this question provided a separate Democratic candidate evaluation (coded -1 if the respondent opposed the Democrat, a 0 if the respondent pressed the space bar and recorded no evaluation, and a 1 if the respondent favored the Democrat).  It also provided separate Republican candidate evaluation (coded -1 if the respondent opposed the Republican, a 0 if the respondent pressed the space bar and recorded no evaluation, and a 1 if the respondent favored the Republican).  In the sample, 43% favored the Democratic candidate, 7% had no evaluation, and 50% opposed him.  For the Republican candidate, while 47% favored him, 7% had no evaluation, and 46% were opposed.  Overall, using the two variables as the dependent variables, provided a simple and direct assessment of each candidate. 

Independent Variables

            The three main independent variables for this analysis were message content, partisanship, and perception of bias.  Three dummy variables were used and scored as 1 if the respondent read the article and 0 if the respondent did not, where the balanced, negative article was the excluded category.  Perception of bias was assessed by asking the respondents, “Which candidate do you think the newspaper favored in the article you read, or didn’t it favor any candidate?”  The variable was coded as 0 if the Democratic candidate was favored, 1 if either both candidates or neither candidate were favored, or 2 if the Republican candidate was favored.

            Respondents also reported their partisan attachment by answering the National Election Studies questions, creating a seven-point party identification scale.  From here, two variables were created, one with Democrats as the in-party and the other with Republicans as the in-party (coded 0 for Independents and out-party identifiers, 1 for weak in-party identifiers, and 2 for strong in-party identifiers).

To ensure an equal distribution across categories, for the sample and for each article content type, the distribution of the seven-point party identification variable was examined.  Partisanship was even across the sample and the treatment groups, with the exception of the small portion of true Independents.  Beyond this simple cross-tabulation, a series of t-tests between the treatment groups tested whether the mean of party identification within each group was different from zero compared to the other groups.  Across all groups, the mean difference was zero, and all alternative hypotheses failed to reach significance at the 0.05 level.  It was concluded that party identification did not vary across treatment groups. 

In addition to the distribution of partisanship, the distribution of political knowledge was tested to ensure that the sample was not overrepresented with either people who knew a lot about politics or those who did not know much about politics.  Either outcome could have affected individuals’ evaluations of candidates.  To measure political knowledge, at the beginning of the survey, respondents answered the six questions developed by Delli Carpini and Keeter (1989).  These questions included naming the office held by Dick Cheney, identifying which branch of government is responsible for declaring a law unconstitutional, identifying the vote necessary to override a veto, identifying which party controls the House of Representatives, stating whether one political party is more conservative at the national level, and identifying which party is more conservative at the national level.  Respondents received 1 point for each correct answer given, creating a scale of 0 to 6.  A set of t-tests to assess the mean difference between treatment groups based on political knowledge confirmed that there was no difference in the means for political knowledge across treatment groups.

Procedure

            Given the ordered nature of the dependent variables an ordered logit analysis will be used.  Ordered logit is the appropriate statistical technique since the response categories of the dependent variable are ranked from low to high.  Using ordinary least squares and assuming interval categories would produce misleading results (Long, 1997).  Ordered logit uses maximum likelihood estimation.  The fit of the model is determined by the likelihood ratio Chi-squared.  The coefficients of an ordered logit tell how the log-odds of each of the produced cut-points decreases with a one unit increase in the odds of being a higher category.  The significance of the effect of the independent variable on the dependent variable are determined by the z-scores.  The interpretation of the ordered logit cannot be done by the raw coefficients.  To show the magnitudes of effects of the independent variable, predicted probabilities are calculated.  These probabilities show the effect of a single variable, holding all other variables constant (Long and Freese, 2006).

Results

Message Content and Candidate Evaluations

First, it was hypothesized that negative message content would affect candidate evaluations.  To address this, two ordered logit regression models were used—one for each dependent variable for candidate evaluation, including dummy variables for message content, party identification variables, interactions between party identification and content, and perception of bias as independent variables.  Table 1 displays the results of both analyses.  The likelihood ratio Chi-squared for both models are significant, showing a good fit for both models (χ2=85.12, p=0.00 for Democratic evaluation model, χ2=49.70, p=0.00 for Republican evaluation model).  The results for individual interactions show that party identification and message content influenced Democratic and Republican evaluations.  Specifically, for Democratic evaluations, being a Democrat leads to a more favorable evaluation (z=2.36, p=0.02).  Further, having read both the article favoring the Republican candidate and the article favoring both candidates also leads to more favorable Democratic evaluations (z=1.68, p=0.09 and z=1.82, p=0.07, respectively).  This affect is also contingent on partisanship, where the interaction of the Republican party identification variable and having read the article favoring the Republican candidate, as well as the interaction of the Republican party identification variable and having read the article favoring both candidates produce significant effects (z=2.11, p=0.03 and z=1.84, p=0.06, respectively).  In terms of the Republican evaluation model, being a Republican or having read the article favoring the Republican candidate leads to a more favorable Republican candidate evaluation (z=1.85, p=0.06, and z=1.85, p=0.06, respectively).  Further, the interaction of the Democratic party identification variable and having read the article favoring the Democratic candidate produces significant effects (z=1.96, p=0.05).

 

TABLE 1.  The Effects of Partisanship and Media Content on Candidate Evaluations

 

 

 

 

Evaluation of the Democratic

Candidate

Evaluation of the  Republican

Candidate

Party Identification- In-party Democrats

  0.83**

            0.01

 

(0.35)

 (034)

Party Identification- In-party Republicans

0.44 

  0.67+

 

(0.38)

 (0.36)

Positive Democratic Article

          0.59

0.83

 

(0.69)

 (0.66)

Positive Republican Article

          1.01+

  1.07+

 

(0.60)

 (0.58)

Positive Both Candidates Article

 1.09+

 0.12

 

(0.60)

 (0.58)

Perception of Newspaper Bias

 -0.47**

 0.36*

 

(0.17)

(0.17)

Positive Democratic Article* In-party Democrats

          0.73

-1.18*

 

(0.63)

 (0.60)

Positive Republican Article* In-party Democrats

          0.21

 -0.40

       

(0.53)

 (0.49)

Positive both candidates Article* In-party Democrats

         -0.02

0.38

 

(0.52)

 (0.48)

Positive Democratic Article* In-party Republicans

         -0.94

           -0.05

 

(0.60)

 (0.57)

Positive Republican Article* In-party Republicans

         -1.14*

           -0.40

       

(0.54)

 (0.50)

Positive both candidates Article* In-party Republicans

         -0.95+

            0.06

       

(0.51)

 (0.50)

Threshold 1

          0.69

            0.83

 

         (0.47)

 (0.44)

Threshold 2

          1.07

           1.15

 

         (0.47)

 (0.44)

LR Chi2

   85.12***

   49.70***

N-size

          323

           323

Note: Results based on ordered logit regression.

Two dependent variables were used: Evaluation of the Democratic Candidate and Evaluation of the Republican Candidate- both coded -1 if the respondent reported opposing the candidate, 0 if the respondent pressed the space bar and recorded no evaluation, and 1 if the respondent reported favoring the candidate. 

Two variables are used for party identification, one where Democrat is the in-party and one where Republican is the in-party.  These are coded 0 for Independents and the out-party, 1 for weak identifiers of the in-party and 2 for strong identifiers of the in-party. 

Dummy variables were used for the article content (1 if the respondent read the article, 0 all others). 

Perception of Newspaper Bias is coded as a -1 if the respondent reported the Democratic candidate as favored, a 0 if the respondent reported neither candidate or both candidates were favored, and a 1 if the respondent reported the Republican candidate as favored.

Interactions are used between each dummy variable of article content and each of the party identification variables. 

Standard errors in parentheses.

***p<0.001, ** p<0.01, * p<0.05, +p<0.10.

 

While the significance of the variables can be obtained from Table 1, the magnitude of the effects can be determined by calculating predicted probabilities.  Table 2 presents the predicted probabilities for each candidate evaluation for strong partisans.

TABLE 2. Predicted Probabilities of Candidate Evaluations For Strong Partisans Based on 

 

Newspaper Article Content

 

 

 

 

 

 

 

 

 

 

Newspaper Article Content

 

 

Candidate

+ Democrat

- Democrat

+ Democrat

- Democrat

Party ID

Evaluation Made

- Republican

+ Republican

+ Republican

- Republican

 

 

 

 

 

 

Strong

Favorable

0.90

0.83 

0.77

0.54

Democrats

Democratic

(0.07)

(0.08)

(0.06)

(0.11)

 

 

 

 

 

 

 

Unfavorable

0.88

0.56

0.41

0.62

 

Republican

(0.07)

(0.12)

(0.11)

(0.11)

 

 

 

 

 

 

Strong

Favorable

0.78

0.68

0.67

0.62

 

Republicans

Republican

(0.11)

(0.11)

(0.11)

(0.12)

 

 

 

 

 

 

 

Unfavorable

0.82

0.81

0.73

0.55

 

 

Democratic

(0.10)

(0.09)

(0.10)

(0.16)

 

Note: Cell entries are predicted probabilities obtained from Table 1 estimates. 

Standard errors are in parentheses.  

Newspaper article content refers to the article randomly assigned to the respondent (“+” means positive coverage,  “-” means negative coverage). 

Candidate Evaluation Made is the probability that this candidate evaluation was made.

Perceived Newspaper bias held at its mean.

 

Hypothesis 1 predicted 2 situations.  First, negative, incongruent information about the like-minded candidate will lead to favorable like-minded candidate evaluations.  The results support this.  For example, as column 2 of Table 2 shows, strong Democrats had a 0.83 probability of reporting a favorable Democratic evaluation when exposed to negative, incongruent information about the Democratic candidate.  Strong Republicans also used negative information to support partisan goals.  They had a 0.78 probability of reporting a favorable Republican evaluation when the content was negative and incongruent to the Republican candidate.

 Second, negative, congruent information about the opposition will lead to negative evaluations of the opposition.  For example, as column 1 shows, strong Democrats had a 0.88 probability of reporting an unfavorable evaluation of the Republican candidate when exposed to negative, congruent information about that candidate.  Similarly, strong Republicans had a 0.81 probability of reporting an unfavorable Democratic evaluation when the content was negative and congruent to the Democratic candidate.  Overall, the results supported Hypothesis 1. 

Hypothesis 1 only addressed the effect of negative content.  However, as Table 2 shows, positive information also affected how respondents evaluated candidates.  This effect is not as large as that from negative content. For example, as column 1 reveals, strong Democrats had a 0.90 probability of reporting a favorable candidate evaluation for the Democratic candidate when content was favorable to the Democratic candidate.  Strong Republicans had a 0.68 probability of reporting a favorable Republican candidate evaluation when content favored the Republican candidate.

Hypothesis 2 states that disparate information will lead to favorable like-minded candidate evaluations and unfavorable opposition candidate evaluations.  Therefore, not only the content’s tone, but also how the content is presented affects evaluations.  To address this, Table 3 displays the changes in the predicted probabilities (based on Table 2) for strong partisans exposed to the balanced articles compared to those who read the disparate articles.  A negative result supports the hypothesis that disparate content lead to more favorable and unfavorable evaluations.

 

TABLE 3. Changes in Predicted Probabilities of Candidate Evaluations For Strong Partisans

 Comparing Biased Articles to Neutral Articles

 

 

 

 

 

 

 

 

Disparate Newspaper Article Content

 

 

+Democrat, -Republican

-Democrat, +Republican

 

 

 

 

 

 

 

 

Change in Probability for Balanced Articles:

 

Candidate

+ Democrat

- Democrat

+ Democrat

- Democrat

Party ID

Evaluation Made

+ Republican

- Republican

+ Republican

- Republican

 

 

 

 

 

 

Strong

Favorable

-0.13

-0.36

-0.06

-0.29

Democrats

Democratic

 (0.08)

 (0.12)

(0.11)

 (0.14)

 

 

 

 

 

 

 

Unfavorable

-0.47

-0.26

-0.15

 0.06

 

Republican

 (0.14)

 (0.12)

(0.19)

 (0.10)

 

 

 

 

 

 

Strong

Favorable

-0.10

-0.15

-0.01

-0.06

Republicans

Republican

 (0.15)

 (0.17)

(0.14)

 (0.16)

 

 

 

 

 

 

 

Unfavorable

-0.08

-0.26

-0.08

-0.26

 

Democratic

 (0.15)

 (0.18)