The Slant of the News: How Editorial Endorsements Influence Campaign Coverage and Citizens' Views of Candidates

One of the essential elements of an impartial press in the United States is the “wall of separation” between the editorial pages and the pages devoted to the news. While the political beliefs of newspaper owners and editors are clearly articulated on opinion pages, their views are not supposed to infiltrate the reporting of the news. The analyses presented in this paper raise questions about this claim. We examine newspaper coverage of more than 60 Senatorial campaigns across three election years and find that information on news pages is slanted in favor of the candidate endorsed on the newspaper's editorial page. We find that the coverage of incumbent Senators is most affected by the newspaper's endorsement decision. We explore the consequences of “slanted” news coverage by showing that voters evaluate endorsed candidates more favorably than candidates who fail to secure an editorial endorsement. The impact of the endorsement decision on voters' evaluations is most powerful in races receiving a great deal of press attention and among citizens who read their local newspaper on a daily basis.

T he First Amendment and scores of Supreme Court decisions accord newspapers broad leeway about what information to print concerning politics. Not surprisingly, with so few restrictions, the press has remade itself several times during the last 250 years. Today, newspaper coverage of political events and public policy conforms to the canons of "professional journalism." These include, for example, a code of ethics concerning the rights and responsibilities of the press in a democracy, the formation of structured education for aspiring journalists, and a series of norms and rules that define the general practice of gathering the news.
This constellation of characteristics is supposed to produce the fair and balanced reporting of the news. Today, the professional journalist assumes "the role of a politically neutral adversary, critically examining all sides of an issue and thereby assuring the impartial coverage of the broadest range of issues" (Bennett 1988, 120). It is expected that professional journalists subordinate their political leanings in pursuit of professionalism (Davis 1996).
One of the essential elements of an impartial press is the impenetrable wall between the editorial pages and the pages devoted to the news. This "wall of separation" is critical to the establishment of a fair and impartial press. While the political beliefs of newspaper owners and editors are articulated clearly on opinion pages, their views are not supposed to infiltrate the reporting of the news. According to journalistic norms, the information on the news pages is reported objectively and free from pressure or direction by the people who own and run the newspapers. According to the executive editor of the Washington Post, Leonard Downie, Jr., the news department at the Post follows the paper's formal ethics policy that "the separation of news columns from the editorial pages . . . is solemn and complete" (Seib 1994, 119). A routine practice at most major newspapers in this country is for editors and reporters who cover the daily news to have nothing to do with endorsement decisions or other opinions printed on the editorial page. Likewise, the editorial page editors have no involvement in the coverage of the news (Seib 1994).
The concept of the "wall" is preached faithfully in journalism classes across the country and is a widely held norm within the newspaper industry. In this paper, we examine the integrity of the wall by investigating the relationship between newspapers' editorial decisions and the tone of their news coverage. 1 We examine newspaper coverage of more than 60 Senatorial campaigns across three election years.
Our analysis has important implications for the role of the modern press during campaigns. Historically, the press has played two crucial roles during elections. First, it has been a conduit of information between citizens and candidates. Indeed, most of what citizens know about candidates comes from the news media (e.g., Graber 1997;Paletz 1999). Furthermore, candidates often look to the press to disseminate their messages to potential supporters (e.g., Just et al. 1996;Kahn and Kenney 1999).
Second, the press structures the discourse of political campaigns by emphasizing certain topics over others (e.g., Iyengar and Kinder 1997;McCombs and Shaw 1972). Along with candidates and parties, the press has the ability to help define the nation's problems and identify possible solutions. While the press may sometimes echo the discussions of political elites, at other times it acts as a dissenting or alternative voice in the political arena.
The press has been performing these roles in the United States for the better part of two centuries. However, how the press performs these functions has changed dramatically. In the early to mid-1800s, the press was unabashedly partisan. Newspapers were viewed not as objective purveyors of information, but as sources of political propaganda. The avowed practice of today's press, in contrast, is to limit displays of political preferences to the editorial pages, while objectively presenting the news in the remainder of the paper.
The editorial influence on news coverage has electoral ramifications. In particular, the endorsement decisions of newspapers, reflected in the news coverage, may affect citizens' attitudes about competing candidates. Citizens may be unaware that the coverage in their daily newspapers is slanted, making them especially susceptible to persuasion (Milburn 1991). Because the modern press claims to be an objective source of information, unlike the partisan press of the past, any bias in the news section of today's newspapers is particularly insidious.

SEARCHING FOR ENDORSEMENT EFFECTS IN CAMPAIGN COVERAGE
We conduct an extensive content analysis of campaign coverage to see whether endorsement decisions affect news coverage. We then turn to individual-level survey data to determine whether citizens develop more favorable impressions of candidates who are endorsed.
We focus on U.S. Senate elections. Senate campaigns generate much more coverage in the local press. Senate races generate more coverage than House races because of the more efficient "media-market" fit in Senate contests (Clarke and Evans 1983;Goldenberg and Traugott 1984). Most metropolitan newspapers need to cover only one Senate race because all (or most) readers reside in the same state. However, most newspapers are read by voters residing in several House districts. In addition, House races receive less coverage than Senate campaigns because reporters and editors view House elections as less newsworthy (Cook 1989). Because Senate races generate so much media coverage, differences in press patterns in Senate campaigns may have a profound influence on voters' views of Senate candidates (Kahn and Kenney 1999). Examining Senate races is also preferable to an examination of presidential races because, in a given election year, more than 30 Senate races are available for study. However, to examine 30 presidential campaigns, we would need to go back to the nineteenth century. Of course, press coverage of presidential campaigns has changed dramatically over this extended period.

A CONTENT ANALYSIS OF CAMPAIGN COVERAGE
We examine all Senate races contested between 1988 and 1992-79 races involving incumbents and 17 open races. 2 To see whether newspaper endorsement decisions influenced press coverage in these races, we selected the largest circulating newspaper in each state for analysis, because this newspaper reaches more potential voters than any other paper in the state. 3 In 12 of the 79 incumbent races, the state's largest newspaper did not endorse a Senate candidate. These races were dropped from our sample, leaving 67 incumbent races available for analysis. 4 Newspapers endorsed a candidate in each of the 17 open campaigns.
For each newspaper in our sample, we examined news coverage between September 1 and Election Day. 5 We examined all articles that mentioned either candidate in the first, state, or editorial section of the newspaper. In total, we examined 5,529 articles. 6

Considering Rival Explanations
To assess the impact of endorsement decisions on campaign coverage, it is necessary to take the political context into account. Endorsed candidates may receive more positive coverage because news personnel favor the candidates they endorse, showering the endorsed candidates with favorable press attention. On the other hand, endorsed candidates may receive more positive coverage because they happen to be superior in terms of experience, popularity, resources, and other criteria (Fenno 1996). We thus employ a series of indicators of the political context. 2 We decided to analyze open races separately from incumbent campaigns for two reasons. First, prior research has documented that these races vary considerably along several dimensions, such as substance of campaign message, tone of campaign message, and voter recognition of candidates (see Abramowitz 1988, Jacobson 2001, and Kahn and Kenney 1999. Second, the number of open races is small during this period (i.e., 17), making an analysis of open races more suggestive than definitive. Nonetheless, we present the findings for the open races in footnote 16. 3 By concentrating on the larger and therefore more professional papers, we may be underrepresenting the relationship between editorial decisions and news content, the idea being that more professional papers are more likely to separate editorial opinions from news. 4 Because we are interested in examining whether endorsement decisions influence news content, we eliminate races in which an endorsement decision is not made. Although not making an endorsement may represent an endorsement decision, we cannot be certain. Some newspapers have a policy of not endorsing candidates. In our study, incumbents obtain the bulk of the newspaper endorsements (70%). Furthermore, our data suggest that endorsements of Senate candidates does not simply follow partisan lines. Newspapers endorse the Senate and the presidential candidate of the same party only 55% of the time. 5 For specific sampling details about the content analyses, see Kahn and Kenney 1999, 36-8. 6 In all, 20 trained coders participated in the project. Intercoder reliability was assessed repeatedly during the coding process. On average, there was 92% agreement across all the content codes. More specifically, intercoder agreement was 0.90 for the tone of the article, 0.94 for the tone of the headline, 0.92 for the tone of the front page, 0.90 for attributed criticisms, 0.88 for unattributed criticisms, 0.92 for the tone of issue coverage, 0.90 for the tone of trait coverage, and 0.96 for the tone of horserace coverage. These measures of tone are discussed below.
Characteristics of the Candidates. To assess the quality of the candidates, we include several measures. First, we assess the seniority of the incumbent seeking election. Senior senators may be more successful in generating favorable coverage than their junior counterparts. 7 In addition to seniority, involvement in scandals and political controversies may influence press coverage of the sitting senators (Abramowitz 1988). Therefore, we include a measure assessing the candidates' involvement in a scandal.
We expect challengers with political "experience" and candidates who are more "skillfull" campaigners to capture more positive press coverage than their less experienced and less skillful counterparts. With regard to experience, others (Green and Krasno 1988;Squire 1989;Squire and Smith 1996) have developed measures that predict challengers' abilities to raise campaign funds and capture votes. Political experience should also explain whether challengers capture the attention of the press. Challengers with electoral and governing experience can rely on established relationships with reporters and editors. These candidates may also be more adept at generating news releases and holding news conferences.
Beyond experience, some challengers are skillful campaigners and are savvy at garnering the media's attention. As Squire and Smith (1996, 236-7) point out, "Campaign skills refer to the personal characteristics of a candidate that strengthen his or her appeal irrespective of party membership or stands on the issues." To measure the quality of the challenger, we combine the elective experience measure with the indicator of campaign skill.
Campaign Spending. Reporters and editors react to campaign activity when preparing stories (e.g., Clarke and Evans 1983). These activities include press releases by the candidates, press conferences, interviews with reporters, and campaign rallies. To stage these events, candidates need money (Cook 1989). Given the expected relationship between campaign spending and coverage, we control for incumbent and challenger spending when examining patterns of news treatment.
Closeness of the Races. The competitiveness of the campaign also influences news coverage. Candidates receive substantially more coverage when the race is perceived as competitive (e.g., Goldenberg and Traugott 1984;Kahn and Kenney 1999;Westlye 1991). In addition, as races become more "hard-fought," press coverage becomes more negative. In competitive races, newspapers publish more criticisms of the candidates and are more likely to use negative traits to describe the candidates. Given the strong relationship between competition and coverage, we control for the closeness of the race by relying on preelection polls published prior to October 1.

THE RELATIONSHIP BETWEEN ENDORSEMENTS AND COVERAGE
Reporters and editors can slant news coverage of campaigns in various ways-some subtle and difficult for casual readers to detect, others unambiguous and easily identified. We borrow liberally from Page's (1996) summary of more common techniques.

Tone of Coverage
We begin by looking at the relationship between endorsements and the overall tone of press coverage. Not all news stories are neutral in tone. Instead, the content of many campaign stories can be characterized as favorable toward one candidate and unflattering to another candidate. We scored each article as positive, negative, mixed, or neutral in tone and then calculated an average tone score for the articles written about each candidate. 8 The ordinary least-squares (OLS) regression results in Table 1 indicate a link between the newspapers' endorsement and how reporters write about sitting senators. In the model for incumbents, the parameter estimate for endorsements is 0.24, with a standard error of 0.09. With all control variables held constant at their means, endorsed incumbents can expect an average tone score of 0.38 (on a scale ranging from −1 to 1), whereas unendorsed incumbents can expect an average tone score of 0.14. In contrast, challengers who secure endorsements do not receive more favorable press treatment than nonendorsed challengers, as indicated by the coefficient of only −0.01, with a standard error of 0.12, compared to nonendorsed challengers.
Among the control variables, competition influences the tone of coverage for incumbents; as races become more competitive, incumbents can expect more critical coverage. 9 In the challenger model, as challengers become more experienced and more skillful, they receive more favorable coverage.

Tone of Prominent Coverage
We turn now to the tone of prominent press coverage. Many newspaper readers are attracted to stories because of the content of the headlines or the placement of stories. Front page stories, for instance, are much more likely to be read than articles buried near the back of the newspaper. Newspapers hoping to hurt or help candidates may slant the tone of headlines and the tone of front page stories. To examine the tone of prominent coverage, we calculate an average tone measure for the front page stories about a candidate and 8 In Appendix B, we present several examples of news content, accompanied by our coding decisions. 9 Throughout this section, we include two additional control variables: a variable measuring each newspaper's endorsement of the presidential candidates in the two previous presidential elections and a variable measuring each newspaper's endorsement decision in the most recent Senate race. These variables failed to reach statistical significance in 15 of the 16 equations in Tables 1-4. Note: The dependent variable is the average tone score for all the articles written about the candidate during the campaign. The tone score averaged 0.26(SD = 0.32) for incumbents and −0.03(SD = 0.33) for challengers. Endorsement is a binary variable coded 1 if the incumbent is endorsed and 0 if the challenger is endorsed. Competition ranges from 24 (in the least competitive races) to 100 (in the most competitive races). Incumbent and challenger spending is logged to base 10 and divided by the voting age population in the state. Scandal is a binary variable where 1 = incumbent involved in scandal and 0 = otherwise. Seniority is measured by number of years in U.S. Senate. Challenger quality is a scale ranging from 1 to 27. All p values are two-tailed. * * * p < 0.01.
an average tone measure for the headlines about a candidate.
In Table 2A, we present incumbent and challenger models predicting the tone of front page articles, and in Table 2B, we present models predicting the tone of headlines for incumbents and challengers. The results indicate that endorsements influence the tone of coverage given to incumbents but not challengers. With all remaining variables fixed at their means, endorsed incumbents can expect an average tone score of 0.27 for all front page articles, compared to an average tone score of 0.06 for nonendorsed incumbents. The same pattern holds for headlines. The tone of headlines is more favorable for endorsed than for nonendorsed incumbents. Other factors being equal, the average tone of headlines for endorsed incumbents is 0.17 higher than the average tone of headlines for nonendorsed incumbents.
Beyond the endorsement effect, the closeness of the race influences the tone of prominent coverage for incumbents. As in the earlier analysis, the most important explanatory variable in the challenger models is the quality of the challenger, which influences the tone of front page articles and the tone of newspaper headlines.

Number of Criticisms
Newspapers can also slant coverage by explicitly criticizing canditates. Journalists often turn to quotations as a way of supporting their point of view. Furthermore, according to the professional norms of newswriting (Cappon 1991), reporters need to identify sources when they offer criticisms of candidates in news stories. However, they sometimes stray from this ideal, criticizing a candidate in a news story without citing a source. When criticisms are linked to sources, readers can employ well-known political cues (e.g., the party identification or ideological view of the source) to place the criticisms in political context. On the other hand, when reporters make critical comments about candidates without identifying a source, readers cannot depend on common heuristics to help "counterargue" the information. Table 3 shows two sets of regression results for criticisms of incumbents and challengers. In Table 3A, we develop incumbent and challenger models to explain the number of attributed criticisms published about the candidates. In addition to controlling for the closeness of the race, the amount of spending by the candidates, and the characteristics of the candidates, we include the number of paragraphs written about the candidate during the course of the campaign. This is necessary because as the total coverage increases, so do criticisms.
The results in Table 3A indicate that endorsed incumbents are treated more favorably than Senators who fail to secure endorsements, receiving, on average, nearly 12 fewer attributed criticisms, holding all remaining variables constant. The endorsement decisions of newspapers do not affect coverage of challengers.
For the remaining variables in the model, the number of paragraphs written about the candidates is by far the most powerful factor influencing the number of attributed criticisms. The level of competition influences coverage in both models, with candidates in competitive races receiving more attributed criticisms. In addition, quality challengers who spend more money are successful in creating more attributed criticisms of incumbents. 10 Note: Front page tone is the average tone score for all the front page articles about the candidate during the campaign. The average front page tone was 0.20(SD = 0.29) for incumbents and 0.03(SD = 0.29) for challengers. Tone of headlines is the average tone score for all the headlines mentioning the candidate during the campaign. The average headline tone was 0.14 for incumbents(SD = 0.24) and −0.06(SD = 0.28) for challengers. Endorsement is a binary variable coded 1 if the incumbent is endorsed and 0 if the challenger is endorsed. Competition ranges from 24 (in the least competitive races) to 100 (in the most competitive races). Incumbent and challenger spending is logged to base 10 and divided by the voting age population in the state. Scandal is a binary variable where 1 = incumbent involved in scandal and 0 = otherwise. Seniority is measured by number of years in U.S. Senate. Challenger quality is a scale ranging from 1 to 27. All p values are two-tailed. * * * p < 0.01; * * p < 0.05.
In Table 3B, we focus on unattributed criticisms. The dependent variable in these models is the number of unattributed criticisms published about the candidate divided by the total number of criticisms written about the candidate. Endorsements continue to influence the tone of coverage for incumbents, but endorsements fail to influence the coverage of challengers. 11

Substance of Coverage
We now turn to the substance of coverage, seeking to determine whether coverage of issues, traits, or the horserace varies with endorsement decisions. We begin by examining issue coverage, which far outstrips trait and horserace coverage in terms of press attention. On average, 154 paragraphs about issues are published during Senate campaigns, wheres only 54 paragraphs, on 11 We repeated the same analysis for unattributed criticisms with one change; we made the dependent variable the number of unattributed criticisms. The results are identical to those presented in Table 3. average, discuss the personal characteristics of candidates, and 40 paragraphs, on average, discuss the candidates' chances of victory. 12 We calculate an average tone measure for stories mainly about issues. 13 The results presented in the first and second columns in Table 4 indicate that endorsed incumbents could expect more positive coverage of their policy views compared to nonendorsed incumbents. The endorsement coefficient indicates that the average tone score for issue articles is 0.23 higher for endorsed incumbents compared to nonendorsed incumbents, all other things being equal. In contrast, Note: Number of attributed criticisms is the number of criticisms citing a source published about a candidate during the campaign. The mean for incumbents was 44.1(SD = 36) per race and 29(SD = 28) for challengers. Proportion of unattributed criticisms is the number of criticisms not referring to a source published about a candidate/total number of criticisms published about the candidate. The mean for incumbents was 0.12(SD = 0.12) per race and 0.22(SD = 0.25) for challengers. Endorsement is a binary variable coded 1 if the incumbent is endorsed and 0 if the challenger is endorsed. Competition ranges from 24 (in the least competitive races) to 100 (in the most competitive races). Incumbent and challenger spending is logged to base 10 and divided by voting age population in the state. Scandal is a binary variable where 1 = incumbent involved in scandal and 0 = otherwise. Seniority is measured by number of years in U.S. Senate. Challenger quality is a scale ranging from 1 to 27. Paragraphs about candidate is the number of paragraphs written about the a candidate during the course of the campaign. All p values are two-tailed. * * * p < 0.01; * * p < 0.05; * p < 0.10. endorsements failed to influence the tone of issue coverage for challengers.
We also look at patterns of trait coverage. While professional standards for modern journalists caution against using evaluative adjective (Cappon 1991), contemporary news coverage is replete with evaluative adjectives (Page 1996). For example, reporters routinely characterize candidates as inexperienced, erratic, out of touch, or ineffective.
We present results for models of the proportion of negative trait coverage devoted to the candidates in the third and fourth columns in Table 4. In a deviation from the pattern we have seen thus far, endorsement decisions are not consequential for incumbents. Reporters are not more likely to describe incumbents in unflattering terms when the incumbent fails to impress the newspaper's editorial board. In addition, and consistent with our previous results, critical coverage of the challengers' personal traits does not vary with endorsement decisions.
Finally, we examine patterns of horserace coverage in the fifth and sixth columns in Table 4. Political reporters spend a great deal of time describing the viability of the competing candidates (e.g., Bartels 1988;Patterson 1993). The news discussion of the candidate's viability can be consequential, influencing voters' impressions of the candidates and affecting the candidates' abilities to raise funds (Bartels 1988;Mutz 1995). Endorsements do alter how reporters describe the viability of sitting senators. Incumbents who are not endorsed by newspapers are described as less electable than incumbents who are endorsed, all other things being equal. Reporters are less likely to use phrases such as "way ahead," "big lead," and "safe margin" when covering nonendorsed incumbents. Consistent with all previous analyses in this paper, endorsed challengers do not receive the same "endorsement slant" as incumbents.
The closeness of the race influences the coverage of issues, traits, and horserace for incumbents. Competition also alters the tone of horserace coverage given Note: Tone of issue coverage is the average tone score for all articles mainly about issues for the candidate during the campaign The average tone for incumbents was 0.24(SD = 0.30) for incumbents and 0.01(SD = 0.31) for challengers. Tone of trait coverage is the number of negative traits about a candidate/the total number of traits (positive and negative) mentioned about the candidate. The mean for incumbents was 0.27(SD = 0.18) per race and 0.33(SD = 0.20) for challengers. Tone of horserace coverage is the average viability score based on all the articles written about the candidate. The mean viability score for incumbents was 4.1(SD = 0.65) and the mean for challengers was 2.0(SD = 0.63). Competition ranges from 24 (in the least competitive races) to 100 (in the most competitive races). Incumbent and challenger spending is logged to base 10 and divided by the voting age population in the state. Scandal is a binary variable where 1 = incumbent involved in scandal and 0 = otherwise. Seniority is measured by number of years in U.S. Senate. Challenger quality is a scale ranging from 1 to 27. All p values are two-tailed. * * * p < 0.01; * * p < 0.05; * p < 0.10.
to the challenger, 14 and the quality of the challenger influences the tone of issue coverage for challengers. 15 14 Competition has been related to the slant of coverage in virtually all analyses. Given its consistent influence, it is possible that the strength of the relationship between the tone of coverage and endorsements changes as competition changes. We conducted two separate tests to investigate this possibility. First, we added an interaction between competition and endorsement to each of the analyses in Tables 1-4. The interaction coefficient fails to reach traditional levels of statistical significance ( p < 0.05) in 14 of the 16 tests. Second, we partitioned competition into low, medium, and high levels and then reestimated all of the analyses in Tables 1-4 for each level. In highly competitive races, the endorsement variable reached statistical significance in eight of the 16 tests. In moderately competitive races, endorsement reached statistical significance in two of the 16 tests. And in races with low levels of competition, endorsement reached statistical significance in one of the 16 tests. These findings suggest a modest conditional relationship between competition and endorsement, with the impact of endorsement being the most powerful in the most competitive races. The results of these analyses, as well as other analyses discussed in this paper, are available from the authors upon request. 15 Although challenger quality has performed quite well, defeating the null hypothesis in four of the eight challenger models ( p < 0.01), In sum, our results suggest that newspaper endorsements shape newspaper coverage for incumbents. For every measure of coverage, save one, endorsed incumbents receive more favorable coverage than their nonendorsed counterparts. Coverage of challengers, on the other hand, is unaffected by endorsement decisions. Challengers do, however, reap indirect benefits from endorsements: When challengers capture endorsements, their opponents-the nonendorsed incumbents-receive more negative coverage. 16 the elected experience component of the challenger quality model is not a true interval measure. Therefore, we examined alternative operationalizations of challenger quality. First, we recoded the nine categories of experience into three binary variables. Second, we examined a binary experience variable (i.e., elected experience, no elected experience). These operationalizations of challenger quality did not perform as well as the original measure. 16 As discussed earlier, we replicated all the analyses in Tables 1-4 for open races. The endorsement variable fails to achieve statistical significance in these models. Nonetheless, as noted earlier, our conclusions about open races are only suggestive given the small number of cases.

ASSESSING TEMPORAL ORDER
We have argued that the views of newspaper editors influence how the paper covers Senate campaigns. However, it is conceivable that the causal direction is reversed. That is, what appears in the newspaper may influence the endorsement decisions of editors. However, prior research examining how endorsement decisions are made suggests that editorial boards do not simply rely on news reports (Clarke and Evans 1983;Seib 1994). In particular, endorsement decisions at virtually all newspapers come after detailed scrutiny by editorial board members of the competing Senatorial candidates, without the influence of reporters. Most editorial boards conduct face-to-face interviews with the Senatorial candidates, and reporters are not privy to these interviews. In addition, editorial boards routinely prepare detailed reports on potential candidates, intentionally excluding reporters. Nevertheless, to feel confident that endorsement decisions influence news coverage, and not vice versa, we conduct a test aimed directly at assessing temporal order. Our task is made easier because we know the precise publication date of each newspaper's endorsement.
We model coverage patterns over time and determine whether the endorsement decision of the newspaper alters the tone of coverage from early to late in the campaign. To assess the change in the tone of coverage before and after the endorsement decision, we specify and estimate a dynamic OLS model. Based on the date of each news article and the date of the endorsement decision, we develop a change model in which the tone of coverage after the endorsement decision is a function of (1) the tone of coverage before the endorsement decision, (2) the newspaper's endorsement decision, (3) changes in incumbent and challenger spending, and (4) changes in the closeness of the race. This enables us to estimate whether changes in the tone of coverage can be explained by the endorsement decision, controlling for prior coverage and other dynamic campaign forces. 17 To measure the tone of coverage, we combine the eight measures of coverage presented in Tables 1-4 into a single global measure of tone. 18 Thus, the global tone measure after the endorsement decision serves as the dependent variable. 19 The results presented in Table 5 indicate that the endorsement decision changes the tone of coverage. 20 Incumbents who capture the 17 As with the prior analyses, we control for the newspapers' prior endorsements. Consistent with our prior analyses, both measures of prior endorsements fail to reach statistical significance. 18 More specifically, we normalize each of the tone measures (e.g., number of attributed criticisms, tone of headlines) and then take the mean of these normalized scores. 19 In this section, we restrict our examination to incumbent coverage because we found no evidence of an endorsement effect for challengers and candidates in open races. 20 The inclusion of a lagged dependent variable may produce autocorrelation and bias the estimators. We employ a test recommended by Durbin (1970), which requires the following steps: First, we use the original OLS model to estimate the residuals. Next, we regress the residuals at time t on the errors at time t − 1 while including newspaper's endorsement receive nearly a three-point increase in positive coverage after the editorial decision is published, on average, compared to incumbents who are not endorsed. 21 These findings enhance our confidence that the editors' endorsement decisions are influencing the tone of news coverage. Still open, though, is the question of whether slanted coverage alters citizens' views of candidates. If slanted coverage falls on deaf ears, then the relationship between editorial opinion and news coverage is far less noteworthy. all the variables from the original model as additional independent variables. Finally, we conduct classical tests of significance for the coefficient estimating the errors at time t − 1. The results show that the standard error for the coefficient representing errors at time t − 1 is almost as large as the coefficient, indicating the lack of autocorrelation in the model. Therefore, no remedial action is needed. 21 We examined the temporal order question with one additional test. We compared the impact of the endorsement decision on the slant of coverage prior to and after the publication of the endorsement. We ran two OLS equations, one predicting the impact of the endorsement on global tone of coverage before the endorsement decision and the other predicting the impact of the endorsement on the global tone of coverage after the endorsement decision, controlling for all the variables in Tables 1-4. The impact of the endorsement decision on coverage increases sharply after newspapers print their endorsements. The unstandardized coefficient is 1.89 (standard error = 0.91) before the endorsement decision and 3.73 (standard error = 1.06) after the endorsement decision. This analysis suggests that once editors' preferences become known, the slant of new coverage in favor of endorsed candidates increases.

DO ENDORSEMENTS INFLUENCE CITIZENS' VIEWS OF CANDIDATES? A SURVEY ANALYSIS
People rely on and look to newspapers for information about political candidates (e.g., Goldenberg and Traugott 1984;Just et al. 1996). Therefore, it is reasonable to expect that slanted press coverage, driven by endorsement decisions, may influence the electoral fortunes of political candidates. To test for the electoral implications of newspaper endorsements, we integrate information about citizens' evaluations of the candidates with information about the newspapers' endorsement decisions, drawing on the 1988-92 National Election Studies/Senate Election Studies (NES/SES) for data on respondents' evaluations of U.S. Senate candidates. 22 We create a measure of comparative candidate evaluations. We also control for rival factors known to influence evaluations, such as political attitudes, perceptions of national conditions, candidate characteristics, and campaign characteristics. 23

Political Attitudes
We measure three political attitudes: party identification, ideological characteristics, and assessments of issues. According to Congressional scholars, these three forces consistently influence citizens' views of Senate candidates (e.g., Abramowitz and Segal 1992;Kahn and Kenney 1999). To measure party identification, we rely on the standard seven-point scale; for ideology, we construct a variable capturing the "comparative ideological distances" between the respondent and the two candidates, and we develop a measure of issue evaluations based on a series of spending questions included in the NES/SES.

Perceptions of National Conditions
Voters also consider the prevailing national conditions when casting their ballots (e.g., Ferejohn and Calvert 1984;Stein 1990), blaming candidates of the president's party if the national economy is getting worse or crediting candidates who share the president's partisan label if the economy seems healthy. Because voters may also use their vote for the president as a guide in their choice of Senate candidates (Campbell and Sumners 1990;Ferejohn and Calvert 1984), we include a measure of presidential vote choice in our model. In addition, voters may support the Senate candidate of the president's party if they approve of the president's job performance (Jacobson 2001;Kernell 1977), a possibility we examine. 22 We look exclusively at incumbent races because we failed to find that endorsements influenced coverage in open races. For the 67 campaigns examined in our analysis, 4,298 interviews were completed. 23 For information on the operationalization of these variables, see Appendix A.

Characteristics of the Candidates
We control for three characteristics of the candidates: the seniority of Senators, the quality of challengers, and whether incumbent Senators are considered controversial (Abramowitz 1988;Squire 1989). By including measures of candidate characteristics, we not only develop a more complete mdoel of citizens' evaluations, but also control for forces that covary with the newspaper's endorsement decision.

Campaign Spending
Citizens may develop more favorable impressions of challengers who spend more, compared to challengers who have few resources (e.g., Green and Krasno 1990;Jacobson 1980). In addition, levels of campaign spending, similar to candidates' characteristics, are likely to covary with the newspapers' endorsement decisions.

Competition
Finally, we measure the closeness of the race. Because editors may be more likely to endorse incumbents in lopsided races, it is necessary to control for competitiveness. Furthermore, people's impressions of candidates may vary with the closeness of race. For example, citizens may develop more favorable impressions of safe incumbents compared to vulnerable incumbents. Similarly, people may have more positive impressions of competitive challengers than challengers lagging far behind in the preelection polls.

THE IMPACT OF ENDORSEMENTS ON CITIZENS' VIEWS OF CANDIDATE
In Table 6, we examine the impact of slanted coverage, as conditioned by the amount of news attention devoted to the campaign. We employ a binary endorsement variable to capture the slant of coverage (1 = the incumbent is endorsed by the state's newspaper; 0 = the incumbent is not endorsed by the newspaper). By including the endorsement variable as a measure of slanted coverage, we capture the candidate preferences of the newspapers, thereby measuring the various ways endorsements influence the tone of coverage. We include an interaction term to estimate whether the impact of the newspaper's endorsement decision on candidate evaluations becomes more powerful as the amount of coverage increases.
The results in Table 6 indicate that the newspapers' endorsement decisions, by affecting patterns of press coverage, influence people's attitudes toward the candidates, holding all other factors constant. 24 As 24 We also examined whether the newspaper's endorsement decision, conditioned by the amount of coverage, influences the voting decision. In this analysis, the dependent variable is 1 = vote for the incumbent and 0 = vote for the challenger. The logit coefficient for the interaction between endorsement and the amount of coverage is 0.0005, with a standard error of 0.0002, suggesting that the slant Note: The dependent variable is the comparative feeling thermometer scores, where challenger scores are subtracted from incumbent scores. See Appendix A for the operationalization of the independent variables. All p values are two-tailed. * * * p < 0.01; * * p < 0.05; * p < 0.10.
expected, the impact of the endorsement decision increases significantly with the amount of news attention, as indicated by the statistically significant and positive interaction coefficient. 25 To illustrate these effects, we calculate point estimates on the comparative feeling thermometer. For example, when news coverage is at its observed maximum value (i.e., paragraphs = 3,070) and the incumbent is endorsed, the predicted score on the comparative feeling thermometers is 31.1, heavily in favor of the incumbent. However, when coverage is at the maximum value and the incumbent is not endorsed, the average score on the feeling thermometers drops more than 16 points, to 13.5, slightly in favor of the incumbent.
In contrast, the effect of endorsements on citizens' attitudes is virtually nonexistent given sparse coverage. With coverage at its observed minimum (i.e., of coverage, conditioned by the amount of coverage, significantly influences respondents' reported vote decision. The number of cases for the analysis is 2,623. 25 We examined the impact of slanted coverage with an alternative measure: the global measure of tone based on the specific measures of slanted coverage presented in Tables 1-4. The influence of the global tone measure on candidate evaluations escalates as the total amount of coverage increases. The interaction coefficient is positive and statistically significant (0.008, with a standard error of 0.003). paragraphs = 71) and the incumbent endorsed, the predicted score on the comparative feeling thermometers is 30.4, heavily in favor of the incumbent. When the incumbent fails to secure the newspaper's endorsement and coverage is at the minimum, the predicted score decreases only to 29.8, still heavily favoring the incumbent. These examples demonstrate that the influence of slant of coverage depends on the amount of campaign news. The effect of endorsements increases as coverage increases, while the effects of endorsements are barely detectable when coverage is sparse. 26 The control variables in Table 6 perform as expected. Virtually all of the rival explanations are signed in the hypothesized directions and reach statistical significance.
The amount of coverage notwithstanding, some people simply do not read the newspaper as often as others. Some people form a lifelong habit of reading the paper every day, while others pick up the newspaper far less frequently. People exposed to the slanted coverage on a daily basis, almost by definition, should be more influenced than people who pick up a paper less regularly.
To test this expectation, we established two categories of readers: everyday readers and less-thaneveryday readers. We then reestimated the equations in Table 6 for these two types of readers (see Table 7). The parameter estimate for the interaction term is statistically significant in the model for everyday readers but not in the model for people who read the paper less often. These findings indicate that the conditional relationship between slanted coverage and the amount of coverage is more powerful for people who read the paper every day. For citizens who do not read the newspaper daily, even large amounts of slanted coverage fail to influence their attitudes about the Senate candidates.
To illustrate the conditional relationship for everyday readers, we vary the amount of coverage and the endorsement decision. 27 When the incumbent is endorsed and the amount of coverage is at its peak (i.e., paragraphs = 3,070), the estimate for everyday readers is 17.1, favoring the incumbent. However, when the incumbent is not endorsed and the amount of coverage is at its highest point, the estimate for everyday readers is −4.9, favoring the challenger. In contrast, when information is at its lowest point, the estimate for everyday readers changes less than one point as a function of the endorsement decision. When coverage 26 The impact of endorsements may be conditioned by other aspects of the campaign environment, such as competition. In particular, the impact of slanted coverage may be greater in competitive races where citizens are more interested in the race and are more likely to seek out news coverage. To test for this possibility, we reestimated the model in Table 6 and substituted the interaction term endorsement * competition for endorsement * total incumbent coverage. The interaction coefficient for endorsement * competition is 0.28, with a standard error of 0.11. The positively signed and statistically significant coefficient demonstrates that the influence of slanted coverage on citizens' evaluations of the candidates increases as competition increases. 27 We do not illustrate the conditional relationship between amount of coverage and endorsement decisions for less-than-everyday readers because the interaction coefficient and the baseline coefficients are not statistically significant for these respondents.

CONCLUSION
We have provided the most systematic evidence to date that newspaper coverage of campaigns is affected by editorial positions, which alter the tone of the news coverage given to incumbents. Moreover, endorsement-driven coverage affects the preferences of citizens. In races receiving a great deal of coverage, endorsed incumbents fare far better with potential voters than nonendorsed incumbents, even controlling for several forces known to shape the voting preferences of citizens. Individuals who routinely read their daily newspapers-the people who are most likely to voteare most affected by coverage patterns. We must emphasize that our study is limited to Senate campaigns over a six-year period. We encourage research on other races, different newspapers, and different years to assess the validity of our results.
Why do the newspapers' editorial decisions and news content coincide? We can only speculate. 29 The connec- 28 We replicated this analysis with the global tone measure and found the same pattern of results. The interaction coefficient for everyday readers is 0.0008 (with a standard error of 0.0004), easily reaching statistical significance. The interaction coefficient for less frequent readers fails to reach statistical significance (0.0001, with a standard error of 0.0008). 29 We do not investigate the causal mechanism that generates the connection between news content and editorial preferences, because tion may be the result of the organizational structure of the newspaper (Gans 1980;Sparrow 1999). As Gans (1980, 97-8) explains, "Because news organizations are assembly lines on which people must work together to manufacture a product against a deadline, they almost always generate conformity-insofar as news judgment is filled with uncertainty, and top editors must, by virtue of their position, resolve uncertainty [by setting] tones and sometimes precedents, which then require conformity." Reporters tend not to be independent observers of the political scene. Instead, they are "employees of complex organizations who see their copy go through layers of editors" (Sparrow 1999, 107-8). Accordingly, it seems inevitable that the views of the editors will shape the content of the news. Ambitious reporters learn what to write to please their editors and publishers, who ultimately decide what is to be the news and which reporters are to be hired and promoted (Sparrow 1999). Page (1996) explains that reporters who hold the same political views as their editors and publishers may be more likely to be hired by particular newspapers and may be more likely to succeed at these newspapers. On the other hand, reporters with opposing views, if hired, may be less satisfied with their job and may be more likely to leave. Sparrow goes a step further by contending that reporters who write stories that "threaten the institutional interests" of the news organization feel "the invisible hand" of the news we have no data concerning the inner workings of newspapers across the nation. room's social control. These reporters learn to regulate and censor themselves in order to be successful.
Finally, why do the editorial preferences of the newspaper influence news coverage only of sitting Senators? Senators, unlike challengers, are powerful figures whose position in government can promote or thwart the interests of editors. When editors view Senators as allies in terms of ideology or party, or even personally, they have clear incentives to foster a friendly and stable relationship with these Senators. If editors provide these Senators with favorable coverage, incumbents may be more likely to push policy agendas that resonate with the media elite. Furthermore, favorable press attention may encourage Senators to be friendly sources of information for reporters and editors as they write campaign and noncampaign stories.
"Unfriendly" Senators, in contrast, will not promote policy changes desired by the editors and publisher of the newspaper. Furthermore, Senators who are at odds with newspapers tend not to be cooperative with reporters and editors (Fenno 1996;Kahn and Kenney 1999). These uncooperative Senators may not hold as many press conferences, may not grant as many interviews, and may not be willing to offer comments on or "off the record." Challengers, in contrast, are not the recipients of slanted coverage. This may be because they do not muster the same amount of coverage as incumbents (Goldenberg and Traugott 1984;Kahn and Kenney 1999;Westlye 1991). Challengers, perhaps because of their general lack of coverage, escape the slanted coverage due to editorial preferences.
Should we be concerned that news coverage has a political slant? In many other democratic countries, the press is unabashedly partisan. In Great Britain, for example, the major newspapers' partisan preferences are well known and these partisan preferences influence the tone and quantity of campaign coverage (see Semetko, Scammell, and Nossiter 1994). In the United States, in contrast, the press ostensibly strives for impartiality. The newspaper industry collectively assures Americans that news coverage is professionally produced and is free from bias. In fact, one of the "canons of journalism" of the American Society of Newspaper Editors is "a clear distinction between news reports and opinion" (McQuail 1992, 38). Our study shows that this goal has not been achieved.
The "hidden" bias in the United States press may be more problematic than the open bias of many European newspapers. If Americans believe that the news is presented in an objective fashion, free from editorial pressure, they may be more susceptible to media manipulation. Given that citizens rely on the press to bring them the news of campaigns, especially in races for statewide and national office, slanted coverage can have dramatic effects on the outcomes of elections.

APPENDIX A. CODING AND MEASUREMENT OF VARIABLES
Amount of News Attention: The number of paragraphs written about the candidates.
Attributed Criticisms: The number of attributed criticisms published about the canddate.
Campaign Spending: Spending for both candidates is logged to base 10 and divided by the state's voting age population. Campaign spending is measured early in the campaign (prior to October 1) to limit the chance that coverage patterns influence spending levels. The bulk of campaign coverage (73%) appears after September 30.
Coattail Voting: This variable is coded 1 (the respondent votes for a Republican for president and a Republican incumbent for Senate; the respondent votes for a Democrat for president and a Democratic incumbent for Senate) to −1 (the respondent votes for a Republican for president and a Democratic incumbent for Senate; the respondent votes for a Democrat for president and a Republican incumbent for Senate).
Comparative Feeling Thermometer: The arithmetic difference between the incumbent and the challenger scores on the NES feeling thermometer scales. Respondents who did not rate a candidate are placed at 50 on the feeling thermometer.
Competition: The difference in support of the candidates in preelection polls from 100, creating a scale ranging from 24 in the least competitive (100 − 76 = 24) to 100 in the most competitive race (100 − 0 = 100). Competition is measured early in the campaign (prior to October 1) to limit the chance that coverage patterns influence competition levels.
Controversial Senator: A Senator is controversial if (1) the Senator was involved in a well-publicized scandal, (2) the Senator ran unsuccessfully for president, (3) the Senator's advancing age was an important campaign issue, or (4) the Senator was newly appointed.
Economic Assessments: The wording of the sociotropic question is, "Now, thinking about the country as a whole, would you say that over the past year, the nation's economy has gotten better, stayed about the same, or gotten worse?" Economic assessments are measured on a five-point scale ranging from +2 (e.g., the nation's economy is much better and the incumbent is a Republican; the nation's economy is much worse and the incumbent is a Democrat) to −2 (e.g., the nation's economy is much better and the incumbent is a Democrat; the nation's economy is much worse and the incumbent is a Republican).
Global Tone Measure: Each of the tone measures in Tables 1-4 is normalized and averaged.
Ideological Placement: The absolute distance between the respondents' self-placement on the ideological scale and their placement of the incumbent and the challenger is calculated. Then the respondents' distance from the challenger is subtracted from their distance from the incumbent, indicating which candidate the respondents feel closer to ideologically. Respondents not answering the ideological questions are recoded to the middle of the scale.
Issue Evaluations: The issue scale is based on respondents' spending preferences for six federal programs (i.e., the environment, education, welfare, health care, child care, and defense). The candidates' positions on these six issues is estimated by calculating a mean position for all Democratic candidates and a mean position for all Republican candidates, based on descriptions of the candidates' positions in the press. The respondents' position is matched with the candidates' positions on each of the issues. The scale is then recoded so that respondents who hold positions identical to those of the incumbent are given the highest score, and respondents sharing all the same positions as the challenger are given the lowest score. We adopt this measure because the NES/SES does not ask respondents to place candidates on issue scales.
Newspaper Readership: The NES/SES asks respondents, "How many days in the past week did you read a daily newspaper?" Quality of Challenger: This measure is based on the challenger's experience and the challenger's campaign skills. Challenger experience is based on a nine-point scale: 9 = sitting governor, 8 = House members and statewide officials serving more than one term, 7 = first-term House members and first-term statewide officials, 6 = mayors of major cities, 5 = state legislative leaders, 4 = state legislators, 3 = other local office holders, 2 = celebrities, and 1 = noncelebrities. The candidates' campaign skills are based on descriptions in the CQ Weekly Report and the Almanac of American Politics. Candidates described as skillful campaigners are scored 3, candidates with a mixture of positive and negative reports are scored 2, and candidates described as poor campaigners receive a score of 1. The experience measure is multiplied by the skill measure to create the measure of challenger quality.
Party Identification: Respondents strongly identifying with the incumbent's party receive the highest score (+3); respondents strongly identifying with the challenger's party are given the lowest score (−3).
Presidential Approval: This scale ranges from +2 (e.g., the Senator is a Republican and the respondent strongly approves of the Republican president's performance, the Senator is a Democrat and the respondent strongly disapproves of the Republican president's performance) to −2 (e.g., the Senator is a Republican and the respondent strongly disapproves of the Republican president's performance, the Senator is a Democrat and the respondent strongly approves of the Republican president's performance).
Seniority: Years in office.
Tone of Articles/Tone of Front Page Articles/Tone of Issue Articles: Negative articles receive a score of −1, positive articles receive a score of +1, and neutral and mixed articles receive a score of 0.
Tone of Headlines: Negative headlines receive a score of −1, positive headlines receive a score of +1, and neutral and mixed headlines receive a score of 0.
Tone of Trait Coverage: The number of paragraphs describing the candidates' personalities negatively is divided by the total number of trait paragraphs. A variety of negative traits is examined, including dishonest, uninformed, inexperienced, ineffective, part of the Washington establishment, erratic, insensitive, weak leader, and not an independent thinker.
Unattributed Criticisms: The number of unattributed criticisms is divided by the total number of criticisms published about the candidate.

APPENDIX B. CODING OF NEWSPAPER ARTICLES
Listed below are illustrative examples of how we coded news content from three races: Wallop vs. Vinich, 1988, Casper Star Tribune;Helms vs. Gantt, 1990, The Raleigh News and Observer;and Daschle vs. Haar, 1992; Argus Leader (Sioux Falls).

Tone of Coverage
1. "The Wyoming and National Education Associations endorsed Democratic Senate nominee John Vinich Monday with the NEA giving him $5,000 for his campaign against twotime incumbent Sen. Malcolm Wallop." Casper Star Tribune, 9/20/88 (positive for Vinich, neutral for Wallop).
2. "[Harvey B. Gantt] said during a campaign swing through the Triangle that Mr. Helms had spent his 18 years in the U.S. Senate trying to stomp out such 'personal demons' as communists, liberals, and artists rather than worrying about issues that matter to North Carolina families." The News and Observer, 9/23/90 (negative for Helms, neutral for Gantt).