![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
I argue that the issue-handling reputations that underlie the theory of “issue ownership” affect the favorability of news coverage toward U.S. presidential candidates. A large-scale content analysis of newspaper coverage from the 1992, 1996, and 2000 elections shows that candidates are covered more positively when the news focuses on issues their party “owns” than on “opposition” issues. Democrats benefit particularly from news about social welfare topics. Republicans, meanwhile, receive the most favorable coverage in defense and tax stories. The differences are modest, but consistent, across the 3 election years. The findings suggest that candidates have an additional incentive to focus on owned issues and that the news media play a role in perpetuating issue ownership.
I am indebted to Daron Shaw for providing the data on which this study is based, as well as for his feedback on the article. I also thank Bruce Buchanan, Sharon Jarvis, John Sides, John Petrocik, James Endersby, Chappell Lawson, three anonymous reviewers, and the editor for helpful comments and suggestions.
Notes
1. While the data were collected as part of a partisan campaign, there is little reason to be suspicious of their scientific validity. The RNC was not using the data to promote George H. W. Bush or attack Bill Clinton, but rather to get an accurate picture of the nature of news coverage in the campaign. Moreover, CARMA used rigorous content analysis procedures to collect and code the data.
2. Seventeen papers were initially included in the 2000 coding project but, for various reasons, did not result in a usable sample of stories. I subsequently eliminated those five papers from my analysis, leaving the dozen newspapers listed in Appendix A.
3. To be sure, the lack of consistency among the 3 years—with respect to sample size and frequency of coding—raises legitimate questions about the comparability of the data. In 1992, the sample population numbers more than 45,000 stories, whereas in 1996 and 2000 those figures are roughly 2,100 and 1,900, respectively. Given these discrepancies, one may wonder if it makes sense to compare media coverage across the 3 years. As reported in the Discussion, I have run some of the analyses with a subset of papers as a way of gauging the amount of bias induced by having a different number of papers and different news outlets in the 3 years. For the most part, they do not run counter to the conclusions based upon the larger collection.
4. In 1992 and 2000, the coding project was conducted across the entire campaign, from the spring through the fall. But to create a comparable time period of study for all three elections, I focus only on the coding of news coverage for the last 3 months.
5. For example, in 1996, Bob Dole fell off the stage at a campaign rally in Chico, California. The resulting photograph, which appeared in newspapers across the country—and four columns wide on the front page of the Washington Post—made it appear as if Dole was gravely injured (Canellos & Scales, 1996). The visual may have had an impact beyond the words, whatever their tone or source, that accompanied the photo.
6. As described in Appendix B, there are slight differences between the 1992 and 1996 favorability measures, on one hand, and the 2000 measure, on the other.
7. A list is available from the author upon request.
8. In 1992 and 1996, a randomly selected 5% of all stories were double coded each week. In 2000, one round of reliability tests was conducted during the project. Cohen's kappa for intercoder reliability for the focus of the stories was .85 in 1992, .90 in 1996, and .81 in 2000. The kappa statistics for the favorability measures were also within acceptable ranges of reliability (Banerjee et al., 1999). In 1992, reliability for Bush favorability was .75, and .73 for Clinton. In 1996, the kappa statistic for Dole favorability was .79, and .77 for Clinton. In 2000, kappa was .74 for Bush and .75 for Gore. This is consistent with other studies analyzing media favorability (Neuendorf, 2002).
9. Some recent work has challenged the durability of these reputations. Sides (2006) shows that the party advantages on some issues, rather than remaining stable, have fluctuated considerably over the last two decades. Holian (2004) demonstrates that Clinton's efforts to reframe the debate over crime succeeded in eroding the historic Republican advantage on law and order topics. Still, I choose to rely, in large measure, on Petrocik's definitions of Republican and Democratic issues on the assumption that journalists' orientations to the parties' credibility on issues are based on tradition—the conventional wisdom reflected in Petrocik's categorizations—rather than recent fluctuations in public opinion. The categorization scheme reflects long-held beliefs about the issue-handling abilities of the political parties, which are the same reputations I argue serve as journalistic shortcuts in reporting on political issues. Even if public opinion waxes and wanes, the general reputation of the parties is likely to remain static among elites, which includes journalists. My assumption is that these traditional perceptions of the parties, not the short-term fluctuations in public opinion, guide coverage.
10. To be sure, labor unions are a traditional Democratic constituency, and farmers have also historically supported Democrats. But a reading of the relative handful of news stories in my data set that dealt with either group revealed that rarely are these groups mentioned aside from economic issues. Stories that included reference to labor unions typically had more to do with unemployment, economic recession, or trade rather than anything substantive about the unions or labor rights themselves. Similarly, news mentioning farmers usually was framed in terms of broader economic trends, not with regard to subsidies or their position as a Democratic constituency. As such, it seemed appropriate to code these as economic, rather than partisan, issues. Ultimately, any effects of my tweaking of the categories are likely minimal: Just one-half of 1% of campaign stories in the sample focused on farmers, and about 1% of stories dealt with labor unions. The results are substantively unchanged when labor and farmers are coded as Democratic issues, though the labor stories—most of which came in 1992—do benefit Clinton and hurt Bush, as my hypothesis predicts. Since Clinton received better coverage on economic issues more generally, it is impossible to say whether this is a reflection of labor being an economic or Democratic topic.
11. All of the results presented here are restricted to stories about the campaign, defined as those that include a favorability rating for both candidates. Stories about the activities of the Bush presidency in 1992, for example, are not included. If, however, that story included a mention of Clinton that resulted in a favorability score, then it would be included in the analysis. The strategy has the effect of not allowing for an investigation of incumbent-only coverage during the election years, but also reduces the chances that the analysis will include stories that were not focused on the presidential campaign.
12. I ran two additional analyses to test the hypothesis. First, I calculated the percentage of each candidate's coverage on partisan issues that was negative (below 50), neutral, and positive (above 50). I expected that Democratic candidates would have a larger proportion of positive ratings in Democratic issue stories than in Republican ones, and a larger proportion of negative ratings in Republican stories compared to Democratic stories. The converse should be true for Republicans. Among the 12 relevant comparisons, 11 were in the expected direction, and 8 were statistically significant. In no case did the data contradict the hypothesis. Second, I examined the mean favorability ratings on the five partisan issues shown in . In general, the data on the specific issues fit the patterns described when Republican and Democratic issue stories are aggregated together.
13. The full results are available from the author upon request. For the most part, the control variables reveal no major surprises. Other than in 1992, a newspaper's endorsement does not seem to do much for the favored candidate, but it does produce negative coverage for his opponent on the order of 3 points. The effect of the poll standing variable, which indicates the Republican candidate's lead on the Democrat, suggests variations in public support during the last few months of the campaign may not be as influential as is commonly supposed, at least when controlling for other factors. While four of the six coefficients are significant, only two are pointed in the expected direction. In 1992, Bush's favorability scores declined as he approached Clinton in the polls, and in 2000, Gore received better coverage as Bush's lead grew. A candidate's coverage improves considerably during his party convention. For example, Bush in 1992 received a 16.5-point boost in favorability during the Republican convention. The role of debates in shaping favorability is minimal, as only 2 of the 12 debate coefficients are statistically significant. Interestingly, in two of the three elections, coverage becomes more favorable to the winning candidate and less favorable to the loser as election day nears.
14. The sample of stories in the models presents some estimation concerns. While the sample was drawn with a goal of maximizing generalizability—especially with regard to geography and political orientation—the stories in the analysis are ultimately the result of a convenience sample. This is a potential problem if the stories from any single newspaper share characteristics—perhaps because of the paper's editorial orientation or simply because the same reporter is writing many of the campaign stories. If there are high levels of intraclass correlation on the favorability measures—that is, correlation within stories from individual news outlets—then the models in may be underestimating the standard errors, which could result in a too generous interpretation of statistical significance. In fact, the levels of intraclass correlation within the news outlets are quite low, ranging from .03 to .12 on Republican and Democratic candidate favorability for all 3 years. Still, to be sure the regression models are not biased, I reran them, clustering the standard errors by the newspaper. In large measure, the results are the same as in the original tables, though in two cases the statistical significance of the issue-content variables drops out. The coefficient for Republican issues (primary focus) in the Gore model, significant in , narrowly misses statistical significance when the standard errors are clustered, as does the performance issue variable for Dole. Overall, however, the models do nothing to undermine the substantive conclusions drawn from the original analyses. Even when estimating the models with robust standard errors, the owned-issue content of a story affects favorability in the hypothesized ways. The results of the supplemental analyses are available from the author upon request.
15. These numbers assume the stories appear in a newspaper that did not endorse either candidate. When a newspaper gives a candidate its endorsement, the effects grow larger, from one-half of a point to about 3 points. The effects decrease, though less sharply, when a newspaper endorses the candidate's opponent.
16. Across the two sets of models—one primary issue content, and one for the cumulative issue content—there are 20 coefficients that test the effects of owned-issue content on favorability. Of those, 14 were signed in the expected direction, and only one—a negative effect for Clinton of Democratic issue news in 1996—was contradictory.