Issue Framing and Engagement: Rhetorical Strategy in Public Policy Debates

A similar logic informs treatments of campaign strategy. Candidates, it is thought, highlight issues that they “own” (Petrocik 1996) or dimensions on which they are advantaged (e.g., Carsey 2000; Simon 2002).

Riker’s argument implies that the loudness (or frequency) of an appeal is the primary criterion when it comes to rhetorical strategy. Recent work by Chong and Druckman (2008) illustrates the importance of argument strength (or quality). I discuss the implications of this research for my findings in a later section.

See Schrott and Lanoue (1992, p. 463) on the value of rebuttals in candidate debates (also see Brewer 2003 or Brewer and Gross 2005).

The AP serves as an agenda setter for other papers around the nation (Graber 2001). As such, it is one of the most comprehensive sources for the substantive arguments made for and against health care reform (for a similar use of the AP see Jacobs and Shapiro 2000, p. 160).

The content analysis does not include paid television advertising, such as the “Harry and Louise” spot. Rebuttals to ads such as this one regularly appeared in the news, however (e.g., King 1994), and therefore are included, at least indirectly, in the media data (see West et al. 1996 for more on the role of advertising in the health care reform debate).

Several proposals were discussed during the debate, but Clinton’s bill (the Health Security Act) received most of the attention (Jamieson and Cappella 1998). Accordingly, terms such as “reform proponents” and “reform opponents” refer to proponents and opponents of the Clinton plan.

I used the secondary literature to develop an initial list of the most important considerations in the health care reform debate. Additional categories were added after reading the entire debate several times. As a result of this iterative process, nearly all the text in the news stories could be coded along a substantive dimension. The categories in Table 1 also correspond to several of the health policy metaphors identified by Lau and Schlesinger (2000). This reduces the concern that the empirical results are a product of arbitrary coding decisions. See Appendix A for details.

Multiple coders were used. An intercoder reliability analysis on a randomly selected sample of AP reports (36%) shows agreement between coders. The correlation between ratings was .83 (p < .001). The value of kappa, a more conservative measure of agreement, was .67 (p < .001). Kappa values between .61 and .80 represent “substantial agreement” (Landis and Koch 1977).

One final concern with using the media data to establish the frequency of engagement is the norm of objectivity and the tendency for journalists to seek out opposing views (Bennett 2005). Even if the objectivity norm is operating, we should observe balance in the number—as opposed to the substance—of arguments made by both sides. Data at the daily level show precisely this pattern. Reform proponents made nearly 3 arguments a day; the corresponding figure for reform opponents was 2. While it is true that journalists solicit opposing views, political actors have discretion over which aspects of an issue they choose to talk about.

I exclude descriptive statements and group appeals from the rest of the analysis because these categories are qualitatively different than the others. Descriptive statements consisted of explanations of the alternatives being discussed (i.e., they were intended to be informative, not persuasive). Group appeals included statements that were targeted at many different groups, making the comparison between proponents and opponents difficult. Excluding these categories has little effect on the empirical analysis or the conclusions I draw from it. The pattern in Fig. 1 looks similar when I examine other periods of the debate (e.g., early vs. late).

Kanji (1999) states this test is appropriate when sample size is sufficiently large (i.e., n ≥ 30).

Speculation as to why opponents denied there was a crisis is post hoc, but the pattern that engagement took on this subject is informative. Supporters of the Clinton plan had been making assertions about a health care crisis since the first day of the debate, and yet, opponents did not make their first rebuttal until eight weeks later. Much like the pattern for security and status quo, reform opponents attempted to ignore the subject of a health care crisis. According to journalistic accounts (e.g., Anderson 1994), opponents became worried that the idea of a crisis was beginning to gain traction with the public. This concern likely prompted them to “ride the wave” (Ansolabehere and Iyengar 1994) and to respond, albeit belatedly, to proponent arguments about a health care crisis. Relatively few arguments were made about a health care crisis by either side (see Fig. 1), indicating that this topic was a minor element of the debate.

I measure public support for the Clinton plan by combining responses to 42 opinion polls administered throughout the 11-month debate. See Appendix A for more details.

This is an Autoregressive Distributed Lag (ADL) Model. The term “autoregressive” refers to the fact that lagged values of Y appear as a regressor. The term “distributed” indicates that the effect of X is spread over multiple time periods (Charemza and Deadman 1997). In the analyses below, I include a single lag. This implies that support for the Clinton plan at time t was influenced by rhetoric in t−1 (the previous week). It is plausible that opinion might change in response to arguments made even earlier (say, at t−2). In analyses not shown here, higher order lags were insignificant. Stationarity tests were conducted on all variables (see Appendix B).

Presidential approval might influence support for health care reform (i.e., rising satisfaction with the president’s job performance might make the public more willing to embrace health care reform). In analyses not reported here, presidential approval consistently had no effect on support for the Clinton bill. Given the potential for an endogenous relationship between approval and the dependent variable, I excluded this variable from the analysis.

Separate models were estimated because measures of opponent (proponent) framing and engagement are highly correlated (r = −.51 and −.64, respectively; p < .001). Similar results are obtained when a single model is estimated (e.g., p < .10 for Proponent Engagement _t−1).

OLS estimates can be biased in the presence of outliers, so I examined the data for influential observations. I calculated DFBETAs for the models in this analysis and identified one influential observation. When I replaced the outlier with the mean value for the variable in question, the conclusions from Table 2 remain unchanged. I obtain identical results when I include dummy terms for survey organization (these terms were insignificant so they are not included in Table 2). Finally, using count data in Table 2 (instead of proportions) yields similar results.

Here I rely on the Gallup data because there were more surveys by this organization than any other. Gallup also distinguishes pure independents from those that lean toward one of the major parties (and leaners are closer to partisans in their opinions and behaviors; Keith et al. 1992).

In a related finding, Koch (1998) uses NES data to show that it was the moderately aware who were mostly likely to change their opinions on national health insurance between 1992 and 1994. As Zaller argues (1992, p. 124), this group has the highest probability of both receiving and accepting the messages they receive in the mass media (compared to either the least or most aware). Unfortunately, the surveys used in the present study did not include the types of questions that often are used to create a measure of attentiveness (e.g., political knowledge items). When I examine opinion change across education groups, the average change for people with a high school diploma was twice the size of the change for those with a college degree or more (6 vs. 3 percentage points), a pattern that seems consistent with Koch’s (1998) results.

Prior studies have discerned argument strength empirically (e.g., by asking experimental subjects to rate a message in terms of its effectiveness; see Eagly and Chaiken 1993). With that approach, the precise factors that make an argument strong or weak remain unspecified (see Chong and Druckman 2008, n. 4). The preceding discussion suggests that a policy’s design has some bearing on the strength of the various arguments that are used to shape public opinion.

In a related study, Cobb and Kuklinski (1997) examine the effect of political arguments on opinion about health care reform and the North American Free Trade Agreement (NAFTA). They found more opinion change in response to an experimental frame on NAFTA (an issue on which most subjects were initially neutral) than on health care (an issue on which most subjects had preexisting and polarized views).

Following Petrocik (1996, p. 837) I excluded statements about legislative strategy (also called “game frames”). About 10% of the arguments could not be classified into one of the categories in Table 1 and were placed in a residual category (also excluded from analysis).