Você está na página 1de 17

This article was downloaded by: [University of Pennsylvania] On: 15 August 2012, At: 12:17 Publisher: Routledge Informa

Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Communication Monographs
Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/rcmm20

A meta-analysis of the effects of viewing U.S. presidential debates


William L. Benoit, Glenn J. Hansen & Rebecca M. Verser Version of record first published: 24 Jun 2010

To cite this article: William L. Benoit, Glenn J. Hansen & Rebecca M. Verser (2003): A metaanalysis of the effects of viewing U.S. presidential debates, Communication Monographs, 70:4, 335-350 To link to this article: http://dx.doi.org/10.1080/0363775032000179133

PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: http://www.tandfonline.com/page/terms-andconditions This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.

A Meta-analysis of the Effects of Viewing U.S. Presidential Debates


William L. Benoit, Glenn J. Hansen, and Rebecca M. Verser
Televised debates are now an expected component of the American presidential election campaign. A meta-analysis was used to cumulate the research on the effects of watching presidential debates. General campaign debates increase issue knowledge and issue salience (the number of issues a voter uses to evaluate candidates) and can change preference for candidates issue stands. Debates can have an agenda-setting effect. Debates can alter perceptions of the candidates personality, but they do not exert a signicant effect on perceptions of the candidates competence (leadership ability). Debates can affect vote preference. Primary debates increase issue knowledge, inuence perceptions of candidates character, and can alter voter preferences (the effect sizes for these variables are larger in primary than general debates). The effect sizes for the dependent variables with signicant effects were heterogeneous (except for effects of debates other than the rst on vote preference). No support was found for several possible moderator variables on issue knowledge, character perceptions, candidate competence, and vote preference: nature of subject pool (students, nonstudents), study design (pretest/posttest, viewers/nonviewers), number of days between debate and election, or data collection method (public opinion poll or experimenter data). The rst debate in a series had a larger effect on vote preference than other debates, but was not a moderator for other dependent variables. The possibility that other moderator variables are at work cannot be rejected.

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

elevised presidential debates have become an expected feature of the American political landscape (Friedenberg, 1994). They have occurred in eight general campaigns (1960, 19762000) and, although the records are not as clear, it appears that debates have been broadcast in the primary phase of the campaign in thirteen American presidential elections, beginning in 1948 with a radio debate featuring Thomas Dewey and Harold Stassen (Benoit et al., 2002a). Presidential debates have occurred in several other countries including Australia, Canada, Germany, Israel, Korea, and New Zealand (see, e.g., Coleman, 2000). Debates are a key element of a campaign. A citizens vote represents a choice between (or among) competing candidates. Debates present the leading candidates side by side, discussing more or less the same topics, which helps voters compare the options available to them and assists them in making their vote choice. Moreover, the face-to-face confrontation of debates allows candidates an opportunity to respond to opponents and such clashes can yield benets for voters. Furthermore, the length of presidential debates (90 minutes for general election debates after the four debates of 1960) provides voters with an extended opportunity to hear the candidates discuss issues. Although there are more ads than debates, each debate is far longer than any individual ad (most of which are 30 seconds long in recent campaigns). Although debate viewership varies from campaign to campaign as well as from debate to debate within a campaign, debates usually attract larger audiences than other campaign events (Carlin, 1994). The huge size of the presidential debate
William L. Benoit (Ph.D., Wayne State University, 1979) is Professor, and Glenn J. Hansen (M.A., Bethel College, 1999) and Rebecca M. Verser (M.A., University of Arkansas, 2003) are Doctoral students at the University of Missouri, Columbia.
Communication Monographs,

Copyright 2002, National Communication Association DOI: 10.1080/0363775032000179133

Vol. 70, No. 4, December 2003, pp 335350

336

COMMUNICATION MONOGRAPHS

audience means that capacity for inuence is considerable. Finally, the candidates prepare extensively for the debates, but unexpected questions or statements from opponents may provide voters with a more spontaneous (and possibly less contrived) impression of the candidates than they can obtain from other messages. The Racine Group (2002) concluded, while journalists and scholars display varying degrees of cynicism about the debates, few deny that viewers nd them useful and almost no one doubts that they play an important role in national campaigns (p. 201). Some have voiced concerns about limitations on the power of debates. Jamieson and Birdsell (1988, p. 161) asserted, debates dont very often convert partisans on one side to the other. Many voters are not partisans, however, which means that they are susceptible to inuence from debates. At the start of the 2000 presidential campaign, 38% of voters considered themselves independent, 34% identied with the Democratic Party, and 28% with the GOP (Gallup, 1999). Furthermore, a substantial proportion of partisans do vote for the nominee of the other party. Nie, Verba, and Petrocik (1999) reported that between 14% and 27% of party members defect to the opposing party. Although they do not explain the reasons for these vote defections, and they do not attribute them to debates, many partisans do vote for the candidate of the other political party. This fact means that a sizeable number of partisans could be susceptible to inuence from debates. Presidential debates serve other important functions besides conversion of partisans, e.g., they can help undecided voters make a decision. Additionally, debates can increase viewers condence in their vote choice (Benoit, McKinney, & Holbert, 2001), which could rm up the support of partisans and prevent some potential voters from defecting. This process also results in increased turnout of that candidates supporters on election day. Finally, debates may provide voters with more or better information on which to base their decision without necessarily changing their vote choice (Holbert, Benoit, & McKinney, 2002b). Quantitative studies of the effects of presidential debates on voters have occurred in the United States since the rst presidential debates between Richard Nixon and John Kennedy in 1960 and have been conducted in every campaign since the debates resumed in 1976. The recent essay by the Racine Group (2002) lays out important issues and summarizes that work in a narrative review of the literature. The literature has now accumulated a substantial body of research across a number of campaigns, which means there is a solid base of data for conducting a meta-analysis on the effects of debates on voters. Functional Theory (Benoit, Blaney, & Pier, 1998a) argues that citizens can base their vote choices on policy and character, and they explain that two components of character are personality and competence (or leadership). Using these variables (policy or issues, character, competence) from Functional Theory and drawing on ndings in the existing literature (e.g., Benoit et al., 2001; Racine Group, 2002), seven hypotheses are proposed for this meta-analysis:1
H1. Presidential debate watching increases knowledge of candidates issue stands. H2. Presidential debate watching increases issue salience (number of issues used to evaluate candidates). H3. Presidential debate watching affects preference for one candidates issue positions over anothers. H4. Presidential debate watching has an agenda-setting effect (inuences perceptions of the relative importance of issues). H5. Presidential-debate watching affects perceptions of the candidates character (e.g., honesty, compassion, morality).

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

PRESIDENTIAL DEBATES

337

H6. Presidential-debate watching affects perceptions of the candidates competence (leadership ability, experience). H7. Presidential-debate watching affects vote choice.

Procedures will be described next, followed by results and discussion of the implications of this meta-analysis.

Method
Procedure
Literature search and coding. Suggestions by Stock, Okum, Haring, Miller, Kinney, and Williams (1982) were employed to locate and code the articles for the study. The rst step was to locate studies to include in the meta-analysis. Recent review articles were consulted rst (e.g., Benoit et al., 2001; Racine Group, 2002; Yawn & Beatty, 2000). Then various indices (CommSearch, ComIndex, J-Stor) and Loudens useful web-page bibliography (2003) were examined to locate other studies. Each time a pertinent study was located its references were examined for other studies. This approach generated an initial pool of 87 publications that reported empirical studies of presidential debates. The literature indicates that primary debates are likely to have greater inuence on viewers than general debates because voters have less information about the candidates at that point in the campaign (Benoit, McKinney, & Stephenson, 2002a; Best & Hubbard, 2000). For example, in 2000 voters surely knew less about primary contenders such as Orrin Hatch or Gary Bauer than about the two major party nominees who participated in the general debates (Governor George W. Bush and Vice President Al Gore). In fact, voters probably knew less about Bush and Gore during the primary season than they knew about these candidates in the Fall. Thus, primary debates are likely to have greater effects than general debates. Accordingly, the results for general and primary debates will be reported separately (studies located permitted tests of H1, H5, and H7 for primary debates).2 Second, a coding sheet was developed that listed the citation for the study, the information sought about the nature of the study (Which campaign and debate was being studied? Did the study use a student sample? Was the design pretest/posttest or posttest only? Did the study use public opinion poll data?), the dependent variable(s) tested in the hypotheses (e.g., issue learning, issue salience, preference for candidate policy stands, agenda setting, candidate character, candidate competence, and vote preference), sample size, and results. The coding sheet and coding procedures were rened by jointly coding nine studies. Finally, the studies were divided among the authors and each individually coded one third of the studies. When questions arose about coding a particular article, all authors met to agree on the coding for that study. Some studies could not be included in the meta-analysis. For example, Kraus and Smith (1962) investigate differences between Republicans and Democrats. Although interesting, they do not report either pretest/posttest data for debate watchers or compare debate watchers with nonwatchers. Gallup (1987) reports the percentages supporting each presidential candidate before and after the 1980 debate, but does not provide the sample size for the surveys. Studies that employed regression were omitted unless they also provided zero-order correlations for the variables of interest. Regression was not used because there is no accurate method of deriving the effect size from beta-weights (b or ; or from the alpha levels associated with the

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

338

COMMUNICATION MONOGRAPHS

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

beta-weights). This decision is consistent with most other meta-analytic literature (Allen & Burrell, 2002; Hunter & Schmidt, 1990). Also, Holbrook (1999) used a dependent variable considered to be a measure of knowledge but which is more accurately conceptualized as a measure of issue salience (see Brians & Wattenberg, 1996; Holbert, Benoit, Hansen, & Wen, 2002a). Accordingly, this study was not included in this analysis. The studies used a variety of dependent variables, which were clustered into the variables embodied in the hypotheses. For example, a single character variable was employed. Nevertheless, some studies asked multiple character questions about such personality traits as honesty, compassion, courage, morality, or sociability. When a study reported data from multiple questions about a concept considered to be a single dependent variable, like character, results for those related items were averaged. For example, if a study reported the effects of watching a debate on honesty, compassion, courage, morality, and sociability, the data on these ve character traits were averaged to create a single score for the character variable. Had the results for each question been included separately, that would have articially inated the N for the meta-analysis (that is, the n for that study would have been counted ve times, one for each of the character traits). When the questions asked about the candidates competence, experience in government, or leadership ability, those items were considered to operationalize the competence variable. If a study provided results for four character variableslike competence, honesty, experience, couragethe data for honesty and courage were averaged (and used to test H5 on character), whereas data for competence and experience were averaged (and used to test H6 on competence). Similarly, some studies reported the results for several questions about issue knowledge (e.g., education, taxation, economy). Again, these questions were averaged into a single estimate of issue knowledge from that sample.
Statistical analysis. This analysis employed the correlation coefcient r as opposed to other measures of effect size (e.g., Cohens d). Hunter and Schmidt (1990) recommended r as the most useful measure of effect size in meta-analysis (they also prefer r over r2; see also Beatty, 2002). Formulas for converting the statistics reported in the coded articles to r were obtained from several sources. Hunter and Schmidt (p. 272) provided the formula for converting t to r. This formula was used to convert F to r after extracting the square root of F (the numerator df must equal 1). Wolf (1986, p. 35) provided similar formulas including the formula for converting 2 to r. Finally, Glass, McGaw, and Smith (1981, p. 139) supplied the Probit transformation table used to convert differences in proportions to effect sizes. One correction was made to the effect sizes. Sampling error was corrected by weighting the average overall effect size by the number of subjects in the study. Hunter and Schmidt (1990) noted that if the population correlation is assumed to be consistent across all studies then the best estimate of that correlation is not the simple mean across studies but a weighted average in which each correlation is weighted by the number of persons in that study (p. 100). Similarly, Wolf (1986) indicated, it can be argued that not all studies synthesized in a meta-analytic literature review should be given equal weight (p. 39). All things being equal, studies with larger sample sizes provide a better estimate of the population parameter being measured. Hunter and Schmidt supplied the formula used to correct for sampling error (p. 100). Effect size mean condence intervals were used to determine whether the average

PRESIDENTIAL DEBATES
TABLE 1 KNOWLEDGE OF CANDIDATE ISSUE POSITIONS IN GENERAL DEBATES Author(s) Bishop, Oldendick, & Tuchfarber, 1978 Graber, 1978 Chaffee, 1978 Abramowitz, 1978 Benoit & Hansen, in press Mulder, 1978 Lang & Lang, 1978 Benoit & Hansen, in press Wall, Golden, & James, 1988 Benoit & Hansen, in press Lanoue, 1991 Zhu, Milavasky, & Biswas, 1994 Jamieson & Adasiewicz, 2000 Holbert et al., 2002a Benoit et al., 1998b Benoit & Hansen, in press Holbert et al., 2002a Benoit & Hansen, in press Mean weighted effect size, total N Year 1976 1976 1976 1976 1976 1976 1976 1980 1984 1984 1988 1992 1996 1996 1996 1996 1996 2000

339

r
0.440 0.410 0.243 0.381 0.174 0.235 0.146 0.143 0.089 0.206 0.522 0.252 0.414 0.118 0.160 0.204 0.101 0.429 0.256

N
898 21 164 139 458 91 308 842 203 588 142 53 1013 722 352 393 413 402 7202

Ss1 N N N N N N S N S N S S N N S N N N

Design2 V V P P V V P V P V P V V V V V V V

Debate # all all all 1 all 1 1 all 2 all 2 1 1 1 1 all 2 all

Days3 n/a n/a n/a 41 n/a 41 41 n/a 9 n/a 20 24 31 31 31 n/a 21 n/a

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

Notes. 1Student or nonstudent. 2Pretest/posttest or viewers/nonviewers. 3Number of days between the debate and election day. None of these studies utilized public opinion poll data.

r differs signicantly from zero. If the condence interval of an effect size includes zero then that effect was considered to be no different from zero. The homogeneity of the variance of each effect size was tested using the formula in Hunter and Schmidt (1990, p. 110). A nonsignicant 2 indicates that the variance of the effect size is within the limits of sampling error and that the studies form of homogeneous group (so differences in effect size are due to random error rather than a moderating variable). A signicant result indicates that the differences in effect sizes are larger than what would be expected from sampling error alone (DAlessio & Allen, 2000; Hunter & Schmidt, 1990). The average effect size when the variance is heterogeneous should be viewed cautiously (Allen & Burrell, 2002). When this test indicated heterogeneous variance, potential moderator variables were examined by calculating correlations between the potential moderator variable and the dependent variable (Hunter & Schmidt, p. 486).

General Debates

Results

Knowledge of candidate issue positions. Candidate issue knowledge measures whether debate viewing increases viewers knowledge of the candidates issue positions. Table 1 indicates that 18 measurements of this variable were obtained from 13 studies. The total N for this group of studies was 7202.3 The mean weighted effect size was .256 (SD .136). The 95% condence interval for the effect size was .195 .323. Thus, watching debates has a positive effect on issue knowledge (H1). The test of homogeneity of effect size was statistically signicant, 2(15) 152.62, p .001, which could indicate the presence of one or more moderator variables.

340

COMMUNICATION MONOGRAPHS
TABLE 2 ISSUE SALIENCE IN GENERAL DEBATES Author(s) Benoit & Hansen, in press Benoit & Hansen, in press Benoit & Hansen, in press Holbert et al., 2002a Holbert et al., 2002a Benoit & Hansen, in press Mean weighted effect size, total N Year 1976 1984 1996 1996 1996 2000
r N

Debate # all all all 2 1 all

Days1 n/a n/a n/a 21 31 n/a

0.201 0.223 0.183 0.181 0.076 0.309 0.187

555 623 410 413 722 430 3153

Notes. 1Number of days between the debate and election day. None of these studies utilized public opinion poll data.

All of these studies compared viewers with nonviewers and all studies used nonstudents.

Because the variance in this effect size was heterogeneous, potential moderator variables were examined. The effect size was not related to student (coded low) versus nonstudent samples, r .119, p .64. Nor was effect size related to use of viewers/nonviewers (coded low) or pre/posttest design, r .079, p .76. There was no larger effect size for the rst debate (coded low) compared with subsequent debates, r .246, p .33.4 Because debates held further away from (or, alternatively, closer to) election day could have had greater effects, a correlation was calculated between the days between the debate and the election and the effect size. Again, this potential moderator variable did not correlate substantially with the effect size, r .138, p .70. Thus, none of the moderator variables tested were signicant inuences on effect size.5 Issue salience. Hypothesis two concerned issue salience. The sample of studies offers six data points for the effects of watching debates on issue salience from two studies with a total unique N of 3153 (see Table 2). The mean weighted effect size was .187 (SD .07). The 95% condence interval was .134 .258, indicating that watching debates has an effect on issue salience. The test for homogeneity of effect size was statistically signicant, 2(5) 16.59, p .01. The characteristics of these six studies did not allow additional analyses for potential moderator variables (e.g., none of these studies used students as subjects). Issue preference. Four studies examined the effects of debates on preference for candidates policy positions (see Table 3), the third hypothesis. The total N was 625. The average weighted effect size for this variable was .136 with SD .10. The 95% condence interval was .027 .319, suggesting that this effect is different from zero. The test of homogeneity was not statistically signicant, 2(3) 6.49, p .09. No further moderator tests were possible with these studies. Agenda-setting effect. Three studies (total N of 216) examined the agenda-setting effect of presidential debates (see Table 4). The mean weighted agenda-setting effect size was .291 (SD .168). The 95% condence interval did not include zero (.215 .441), indicating that watching debates has an agenda-setting effect (H4). No additional tests for moderator variables were conducted on these three studies. Perceptions of candidates character. The sample included 10 studies with 17 effect sizes and a combined N of 5426 studied perceptions of candidate character (see Table 5), the fth hypothesis. The mean weighted effect size was .266 (SD .156) and the 95% condence interval ranged from .182 to .350, which suggests this effect is

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

PRESIDENTIAL DEBATES
TABLE 3 ISSUE PREFERENCE IN GENERAL DEBATES Author(s) Kraus & Smith, 1962 Abramowitz, 1978 Becker, Sobowale, Cobbey, & Eyal, 1978 Benoit et al., 2001 Mean weighted effect size, total N Year 1960 1976 1976 2000
r N

341

Ss1 N N N S

Design2 P P V P

Debate # 1 1 all 1

Days3 37 41 n/a 29

0.206 0.006 0.124 0.355 0.136

131 139 297 58 625

Notes. 1Student or nonstudent. 2Pretest/posttest or viewers/nonviewers. 3Number of days between the debate and election day. None of these studies utilized public opinion poll data.

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

TABLE 4 AGENDA SETTING IN GENERAL DEBATES Author(s) McLeod Durall, Ziemke, & Bybee, 1979 Swanson & Swanson, 1978 Benoit et al., 2001 Mean weighted effect size, total N Year 1976 1976 2000
r N

Ss1 N S S

Debate # all 1 1

Days2 n/a 41 29

0.110 0.490 0.385 0.291

97 61 58 216

Notes. 1Student or nonstudent. 2Number of days between the debate and election day. None of these studies utilized public opinion poll data. All of these studies used a pretest/posttest design.

TABLE 5 PERCEPTIONS OF CANDIDATES CHARACTER IN GENERAL DEBATES Author(s) Lang & Lang, 1962 Atkin, Hocking, & McDermott, 1979 Carter, 1962 Benoit & Hansen, in press Becker et al., 1978 Becker et al., 1978 Becker et al., 1978 Mulder, 1978 Graber, 1978 Benoit & Hansen, in press Jamieson & Adasiewicz, 2000 Yawn & Beatty, 2000 Yawn & Beatty, 2000 Benoit et al., 1998b Benoit & Hansen, in press Benoit & Hansen, in press Benoit et al., 2001 Mean weighted effect size, total N Year 1960 1976 1976 1976 1976 1976 1976 1976 1976 1984 1996 1996 1996 1996 1996 2000 2000
r N

Ss1 N N N N N N N S N N N N S S N N S

Design2 P P P V V V V V V V V P P V V V P

Debate # 1 3 all all 1 all all 1 all all 1 2 2 1 all all 1

Days3 12 12 n/a n/a 41 n/a n/a 41 n/a n/a 31 21 21 29 n/a n/a 29

0.411 0.496 0.481 0.079 0.407 0.182 0.220 0.070 0.450 0.162 0.434 0.002 0.002 0.159 0.295 0.225 0.441 0.266

91 466 107 512 297 256 233 310 21 606 1013 59 215 355 402 425 58 5426

Notes. 1Student or nonstudent. 2Pretest/posttest or viewers/nonviewers. 3Number of days between the debate and election day. None of these studies utilized public opinion poll data.

statistically signicant. The test of homogeneity of the effect sizes on character was statistically signicant, 2(16) 150.8, p .001. The nature of these studies permitted several tests for possible moderator

342

COMMUNICATION MONOGRAPHS
TABLE 6 PERCEPTIONS OF CANDIDATE COMPETENCE IN GENERAL DEBATES

Author(s) Ben-Zeev & White, 1962 Ben-Zeev & White, 1962 Lang & Lang, 1962 Ben-Zeev & White, 1962 Mulder, 1978 Benoit et al., 1998b Mean weighted effect size, total N
Notes. 1Student

Year 1960 1960 1960 1960 1976 1996

Ss1 N N N N S S

Design2 P P P P V V

Debate # 3 4 all 2 1 1

Days3 21 12 n/a 27 41 31

0.016 0.023 0.411 0.108 0.028 0.139 0.096

164 169 91 175 310 355 1264

or nonstudent. 2Pretest/posttest or viewers/nonviewers. 3Number of days between the debate and election day. None of these studies utilized public opinion poll data.

variables. There was no relationship between students (coded low) and nonstudents, r .323, p .21. There was no moderating effect for those studies that used viewers/nonviewers (coded low) versus a pre/posttest design, r .175, p .50. No relationship was detected between the rst debate (coded low) and all other debates, r .041, p .876. The correlation between days between the debate and election day was not signicant, r .134, p 730. No other potential moderator variables could be examined. Perception of candidates competence. The sample included four studies (six effect sizes) investigating perceptions of candidates competence (see Table 6). The N for this dependent variable was 1264 and the mean weighted effect size was .096 (SD .102). The 95% condence interval included zero (.003 .245), so this effect is not signicant (H6). The test for homogeneity of effect size was signicant, 2(5) 29.54, p .001. However, no moderator variable could be identied in these data: subjects (students coded low), r .192, p .716; design (viewer/nonviewers coded low), r .192, p .716; debate order (rst coded low), r .523, p .287; and days before election, r .283, p .645. Candidate vote preference. Hypothesis 7 predicted that debate watching inuences vote preference, the nal dependent variable examined. This variable had 25 data points from 14 studies of general election debates and included 8876 subjects (see Table 7). The weighted average mean was .149 (SD .123). The 95% condence interval ranged from .099 to .211, indicating that the effect was signicant. The test of homogeneity was statistically signicant, 2(24) 176.39, p .001. These studies did allow tests of several possible moderators. There was no relationship between student (coded low) and nonstudent samples, r .290, p .159. There was no signicant relationship for design, viewers/nonviewers (coded low) versus pretest/posttest, r .171, p .413. No relationship existed between effect size and whether the study used public opinion poll data or local samples (coded low), r .093, p .657. There was no signicant relationship between the days between the debates and election day and effect size, r .112, p .112. There was, however, a moderator effect between rst (coded low) and all other debates, r .554, p .021. The rst debate had a larger effect on vote preference compared to subsequent debates. In order to make certain that no moderators were present in these subsamples, rst and other debates were considered separately. The weighted mean effect size for watching the rst debate in a campaign on vote preference was .252 (SD .112).

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

PRESIDENTIAL DEBATES
TABLE 7 CANDIDATE VOTE PREFERENCE IN GENERAL DEBATES Author(s) Lang & Lang, 1962 Ben-Zeev & White, 1962 Desutschmann, 1962 Ben-Zeev & White, 1962 Ben-Zeev & White, 1962 Benoit & Hansen, in press Becker et al., 1978 Becker et al., 1978 Geer, 1988 Becker et al., 1978 Rose, 1976 Becker et al., 1978 Becker et al., 1978 McLeod et al., 1979 Lang & Lang, 1978 Jacoby, Troutman, & Whittler, 1986 Geer, 1988 Geer, 1988 Wall et al., 1988 Benoit & Hansen, in press Payne, Golden, Marlier, & Ratzan, 1989 Payne et al., 1989 Lanoue, 1991 Benoit & Hansen, in press Benoit & Hansen, in press Mean weighted effect size, total N
Notes. 1Student

343

Year 1960 1960 1960 1960 1960 1976 1976 1976 1976 1976 1976 1976 1976 1976 1976 1980 1984 1984 1984 1984 1988 1988 1988 1996 2000

Ss1 N N N N N N N N N N N N N N S S N N S N S S S N N

Design2 P P P P P V V V P V V V V P P V P P P V P P P V V

Debate # all 3 all 4 2 all 3 2 1 1 all 3 1 all 1 1 2 1 1 all 1 2 2 all all

Days3 n/a 21 n/a 12 27 n/a 12 n/a 41 41 n/a 12 41 n/a 41 8 9 23 23 n/a 38 20 20 n/a n/a

Poll4 N N N N N N N N P N N N N N N N P P N N N N N N N

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

0.349 0.000 0.381 0.040 0.020 0.061 0.136 0.184 0.215 0.018 0.048 0.107 0.411 0.120 0.303 0.367 0.177 0.177 0.418 0.013 0.065 0.129 0.070 0.024 0.054 0.149

95 149 159 165 155 345 277 168 1137 190 1416 162 387 97 694 30 514 510 302 544 498 182 98 285 317 8876

or nonstudent. 2Pretest/posttest or viewers/nonviewers. 3Number of days between the debate and election day. 4Public opinion poll data/nonpublic public opinion poll data.

The 95% condence interval for this effect size is .137 .355, which means it should be considered statistically signicant. These effect sizes were not homogeneous, 2(7) 102.56, p .001, which could signal the presence of one or more additional moderator variables. None of the potential moderators tested were signicant, however: subject (students coded low), r .289, p .488; design (viewers/nonviewers coded low), r .100, p .813; days before election, r .390, p .340; and poll (nonpoll data coded low), r .204, p .628. The weighted mean effect size for other debates (excluding studies that examined all debates as a group) was .116 (SD .06). The 95% condence interval for this effect size was .052 .140, which indicates that it is unlikely to be different from zero. These effects were homogeneous, 2(8) 7.69, p .45, which means that it is unlikely that a moderator variable affects these results.
Primary Debates Knowledge of candidate issue positions.

Two studies examined candidate issue knowledge gained from viewing primary debates (see Table 8). These studies had ve data points and 187 subjects. The weighted mean was .833 (SD .101). There is a 95% probability that the mean ranges between .697 and .897, indicating that this is a signicant effect (H1). The test of homogeneity was statistically signicant,

344

COMMUNICATION MONOGRAPHS
TABLE 8 KNOWLEDGE OF CANDIDATE ISSUE POSITIONS IN PRIMARY DEBATES Author(s) Pfau, 1988 Pfau, 1988 Pfau, 1988 Benoit et al., 2002a Benoit et al., 2002a Mean weighted effect size, total Year 1984 1984 1984 2000 2000
N r N

Design1 P P P V V

0.688 0.673 0.792 0.902 0.931 0.833

24 26 25 78 34 187

Notes

. 1Pretest/posttest or viewers/nonviewers. All studies used student subjects; none used public opinion poll data.
p

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

.001. Given the lack of variability across the moderator variables no tests for potential moderator variables could be conducted. Two studies with three unique ndings examined perceptions of the candidates character in primary debates (see Table 9). A total of 169 subjects were studied with a mean weighted effect size of .799 ( .082). The 95% condence interval for the mean effect size was between .707 and .959. This supports hypothesis ve in primary debates. The test of homogeneity of variance was statistically signicant, 2(2) 8.69, .001, but given the small number of studies no tests for moderator variables were conducted. Vote preference in primary debates was studied by four different researchers (ve unique data points) using 386 subjects. The mean weighted effect size was .541 ( .133) with a 95% condence interval of .432 .756. This effect indicates that watching primary debates inuences vote preference (H7). Finally, the test of homogeneity of variance was statistically signicant, 2(4) 25.46, .001, but no moderator tests could be conducted for this variable.
Perceptions of candidates character. SD p Candidate vote preference. SD p

2(4) 19.67,

TABLE 9 PERCEPTIONS OF CANDIDATES CHARACTER IN PRIMARY DEBATES Author(s) Benoit et al., 2002a Benoit et al., 2002a Benoit & Stephenson, in press Mean weighted effect size, total
Notes

Year 2000 2000 2000


N

0.946 0.730 0.769 0.799

34 78 57 169

. All studies used student subjects and a pretest/posttest design; none used public opinion poll data. TABLE 10 CANDIDATE VOTE PREFERENCE IN PRIMARY DEBATES Author(s) Best & Hubbard, 1999 Benoit et al., 2002a Benoit et al., 2002a Benoit & Stephenson, in press McKinney, Kaid, & Robertson, 2001 Mean weighted effect size, total
N

Year 1996 2000 2000 2000 2000

0.481 0.840 0.434 0.735 0.481 0.541

126 34 78 57 91 386

Notes

. All of these studies used student subjects and a pretest/posttest design; none used public opinion poll data.

PRESIDENTIAL DEBATES

345

Discussion
The rst four hypotheses pertain to issues or policy. The sample of issue learning studies provides 18 effect sizes encompassing more than 7000 subjects. The data support the prediction that debates engender learning about the issue positions of candidates with a signicant mean effect size of .275. Presidential debates offer an extended opportunity (90 minutes in each debate after 1960) for candidates to present their views. Benoit et al. (2002b) summarized content analyses of every general presidential debate, reporting that 75% of utterances in general debates discuss policy. This outcome means that viewers have an extended opportunity to hear the candidates discuss the pros and cons of their policy stands. It is quite reasonable, therefore, to nd that citizens issue knowledge increases from watching debates. Six studies with more than 3000 subjects produced a signicant mean effect size of .196 for issue salience. Those who watch debates tend to use more issues to evaluate presidential candidates. The fact that presidential candidates emphasize policy in these debates can explain why a greater number of issues are used to evaluate these candidates by those who watch debates. Of course, it is possible that those who have higher levels of issue salience are more likely to watch debates. Four studies including more than 600 participants supported the prediction that watching a presidential debate can inuence issue preference, or which candidates issue position appears superior to voters (or closer to voters own issue positions). The mean effect size for this variable was .138, which is signicant. Debates not only engender knowledge about the candidates policy proposals, but they also inuence the policy preferences of viewers. Three studies involving more than 200 subjects yielded a signicant mean effect size of .291 for agenda setting in general debates. The issues discussed in debates are perceived as more important by viewers. Two hypotheses addressed character. Seventeen studies of almost 5500 subjects produced a mean effect size for candidate personality perceptions of .270, which was signicant. Although content analysis reveals that candidates emphasize policy (Benoit et al. 2002b), presidential candidates did discuss character in 25% of general debate utterances. Thus, there is a clear opportunity for debate viewing to inuence perceptions of the candidates personality (of course, character perceptions could be inuenced by nonverbal factors as well as verbal behavior). Six studies with more than 1200 subjects investigated the effects of debates on perceptions of candidate competence (leadership ability). The mean effect size of .096 was not statistically signicant. Content analysis of general debates (Benoit et al., 2002b) reveals that discussion of leadership ability accounts for only 6% of the remarks in these encounters. Therefore, it is not surprising that no effect was found for perceptions of candidate competence. Finally, what is surely the bottom-line in studies of effects of watching presidential debates concerns vote preference. Twenty-ve studies with more than 8500 subjects yielded a mean effect size for debate watching on vote preference of .149, which was signicant. It is reasonable to place more condence in some of these ndings than others because of the number of studies and subjects involved. Three of these conclusions are based on stronger support: vote preference (25 studies, 8876 Ss), issue knowledge (18 studies, 7202 Ss), and candidate character (17 studies, 5426 Ss). The remaining

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

346

COMMUNICATION MONOGRAPHS

conclusions should be considered more tentative: issue salience (6 studies, 3153 Ss), candidate competence (6 studies, 1264 Ss), issue preference (4 studies, 625 Ss), and agenda setting (3 studies, 216 Ss). Primary debates have signicant effects on issue knowledge, candidate character, and vote preference (the effect sizes for primary debates were larger than for general debates, as was expected from the literature review, as indicated in footnote 2). Because much less research has been conducted on debates in this phase of the campaign, however, these conclusions must also be considered somewhat tentative: vote preference (5 studies, 386 Ss), issue knowledge (5 studies, 187 Ss), and candidate character (3 studies, 169 Ss). It is important to note that the effect sizes for these dependent variables were not homogeneous. This result could indicate the presence of moderator variables. Several potential moderator variables were examined for one or more of the dependent variables investigated: nature of subject pool (students, nonstudents),6 study design (pretest/posttest, viewers/nonviewers), order of debate (rst, other), number of days between debate and election, or data collection method (public opinion poll or experimenter data). The only moderator detected was that the rst general debate has a larger effect on vote preference than subsequent debates. Of course, the fact that effect sizes were heterogeneous could mean that other moderator variables are operating, although they could not be tested. It could also mean that the quality of these studies (e.g., wording of questionnaire items) varies, increasing variance in effect sizes. Thus, the mean effect sizes reported should be interpreted cautiously. This analysis suggests that research on debates has become more sophisticated over time, which could mean that the quality of research has increased.7 First, earlier studies of debates were less likely to conduct statistical analysis (to show that the contrasts observed are probably signicant differences rather than chance uctuations). For example, some early studies report percentages or frequencies without attempting any statistical analyses. Second, when statistical analyses are conducted, the complete results are more likely to be reported in more recent publications. Some early studies indicate which contrasts are signicant (and at what alpha level) but do not report the actual t or F values. Of the 87 potential studies located initially, only 33 reported data that could be included in the meta-analysis. The fact that only a few studies provided data on the reliability of their measures is lamentable. It also means that the estimates reported here for effect sizes may be conservative. Third, political scientists appear to have a proclivity for regression analysis. Although regression is a perfectly acceptable form of statistical analysis, beta-weights cannot be cumulated across studies. So, using regression analysis without providing zero order correlations impedes the cumulation of data through meta-analysis. The advice offered on research into presidential debates by Zhu et al. (1994) merits consideration. After reviewing the literature they offered recommendations for future research on debate watching:
The basic design should be a between-subjects design contrasting viewers and non-viewers, the dependent variables should be measured both pre- and post-debate, the results should be veried with content analysis of the debate to ascertain content-specic effects, and the data should be collected from both the forced viewing condition (experiment, to assure that exposure does take place) and the natural viewing condition (survey, to offset laboratory artifacts). (pp. 311312)

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

PRESIDENTIAL DEBATES

347

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

Research into the effects of debates must consider the content of debates (e.g., it would be a mistake to attempt to measure issue learning from a debate with questionnaire items concerning the candidates stands on abortion if abortion was not discussed in that debate). The search for potential moderator variables indicates that some of this advice is questionable. Design choice (pretest/posttest, viewer/nonviewer) is not a moderating variable for effect size for issue knowledge or candidate personality, the two dependent variables these authors discuss. Finally, this meta-analysis indicates where more research is needed. Research has clearly established that debates are capable of increasing issue knowledge. Although there are undoubtedly more specic questions concerning this variable that could be investigated (e.g., do Republicans, Democrats, and other voters learn equally from debates), the basic question of whether debates enhance knowledge can be answered afrmatively. Similar points could be made about two other variables, vote preference and character. On the other hand, other variables, like issue preference or agenda setting (four and three studies in this sample, respectively), deserve to be prioritized in future research. Closer examination of the Tables reported here reveals that, of the six studies using competence (leadership ability) as a dependent variable, four of those investigated the 1960 debates and one the 1976 debates. Future research might protably study this variable (perhaps more sophisticated research would nd an effect). Finally, relatively few studies have been conducted into the primary debates. Thus, this study might be used both to assess the present status of the study of presidential debates, as well as directions for future research.

Conclusion
This study employed meta-analysis to cumulate the quantitative research on the effects of watching televised presidential debates. Results indicate that general debates have several signicant effects (issue knowledge, issue salience, issue preference, agenda setting, candidate character, and vote preference), although they do not appear to inuence perceptions of candidate competence. Primary debates had signicant effects on the three variables with multiple studies: issue knowledge, candidate character, and vote preference (and these effects were larger than in general debates). Debates potentially can have both direct inuence (on voters who choose to watch debates) and indirect inuence (on voters who learn about debates from the news). A meta-analysis (DAlessio & Allen, 2002) conrmed Festingers (1957) prediction that people are more likely to expose themselves to consonant than dissonant information. Thus, prior attitudes and expectations about debates could inuence who watches debates and are thereby susceptible to these effects. Given the lack of media bias (DAlessio & Allen, 2000), those inuenced by a more indirect path (from news) are likely to receive a balanced report. The variance in effect size for all statistically signicant results was heterogeneous. This outcome could mean that other moderator variables are at work or that there are problems with the quality of some studies. For one or more dependent variables, analyses found no support for several potential moderator variables: nature of sample (students versus nonstudents), design (pretest/posttest versus a posttest-only design of viewers/nonviewers), and whether the data were from public opinion polls

348

COMMUNICATION MONOGRAPHS

or not. The data do reveal that the rst debate tends to have a larger effect on vote preference than subsequent debates. Of course, debate watching is self-selected. It is possible that those who choose to watch debates also are more likely to expose themselves to other sources of campaign information, which means we cannot be certain that debates are responsible for these results. However, the three times that it was possible to test design as a potential moderator variable revealed that it was not a moderator. The fact that there was no difference in effect size between the pre/posttest design (which of course controls for knowledge levels before debates) and the viewer/nonviewer design, it seems unlikely that self-selection explains the results.
Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

might notice that one potential hypothesis is missing. Although some research investigates the question of who won a debate, it was decided to exclude that question from this analysis for two reasons. First, we are not certain what this variable means. Elections are ultimately about vote choice (H7). Approaches to political communication (e.g., Benoit, 1999; Kaid & Johnston, 2000) typically divide message content into the topics of issue or policy and character or image (H1H5). Winning a debate, however, does not seem to pertain directly to any of these concepts; instead, it seems to measure ability to perform in an articial campaign communication event (and one that seems to us to have little to do with being president). Second, given that Ross Perot was considered by viewers to be the winner of the rst and third debates of 1992 without winning a single electoral vote (Benoit & Wells, 1996), there are serious questions about the utility of this variable. 2T-tests were conducted to verify that the effect size for primary debates was signicantly larger than for general debates on each variable with multiple studies using primary debates. All were statistically signicant: knowledge, t(df 21) 7.83, p .001, primary m 0.797, SD 0.119, general m 0.273, SD 0.136; character, t(df 18) 5.40, p .001, primary m 0.833, SD 0.108, general m 0.268, SD 0.173; vote preference, t(df 28) 6.17, p .001, primary m 0.594, SD 0.181, general m 0.155, SD 0.138. 3All reported ns are the number of unique subjects for a particular dependent variable. For example, Holbert et al. (2002a) employed two different groups of subjects and effect sizes were obtainable for each, so this study appears twice in Table 1, once for each distinct group. 4Many studies (see Table 1) examined all debates for a given election in aggregate, so those studies could not be 5included in this analysis. Other potential moderators would have been tested (e.g., town hall versus other debate formats, partisans versus nonpartisans) if possible. The nature of the studies in the sample did not permit these tests, however (e.g., no study of issue knowledge in this sample concerned a town hall debate). 6Analysis of presidential television spots also found no signicant difference between studies using students and nonstudents (Lau, Sigelman, Heldman, & Babbitt, 1999). 7Sophisticated research should not be confused with complex research (Boster, 2002). *Abramowitz, A. I. (1978). The impact of a presidential debate on voter rationality. American Journal of Political Science, 22, 680690. Allen, M., & Burrell, N. (2002). The negativity effect in political advertising: A meta-analysis. In J. P. Dillard & M. Pfau (Eds.), The persuasion handbook: Developments in theory and practice (pp. 8396). Thousand Oaks, CA: Sage Publications. *Atkin, C., Hocking, J., & McDermott, S. (1979). Home state voter response and secondary media coverage. In S. Kraus (Ed.), The great debates: Carter vs. Ford, 1976 (pp. 429436). Bloomington, IN: Indiana University Press. Beatty, M. J. (2002). Do we know a vector from a scalar? Why measures of association (not their squares) are appropriate indices of effect. Human Communication Research, 28, 605611. *Becker, L. B., Sobowale, I. A., Cobbey, R. E., & Eyal, C. H. (1978). Debates effects on voters understanding of candidates and issues. In G. F. Bishop, R. G. Meadow, & M. Jackson-Beeck (Eds.), The presidential debates: Media, electoral, and policy perspectives (pp. 126139). New York: Praeger. Benoit, W. L. (1999). Seeing spots: A functional analysis of presidential television advertisements from 19521996. New York: Praeger. Benoit, W. L., Blaney, J. R., & Pier, P. M. (1998a). Campaign 96: A functional analysis of acclaiming, attacking, and defending. New York: Praeger.

1Readers

Footnotes

References

* Study contributed data for the meta-analysis.

PRESIDENTIAL DEBATES

349

*Benoit, W. L., & Hansen, G. J. (in press). Presidential debate watching, issue knowledge, character evaluation, and vote choice. Human Communication Research. *Benoit, W. L., McKinney, M. S., & Holbert, R. L. (2001). Beyond learning and persona: Extending the scope of presidential debate effects. Communication Monographs, 68, 259273. *Benoit, W. L., McKinney, M. S., & Stephenson, M. T. (2002a). Effects of watching campaign 2000 presidential primary debates. Journal of Communication, 52, 316331. Benoit, W. L., Pier, P. M., Brazeal, L. M., McHale, J. P., Klyukovksi, A., & Airne, D. (2002b). The primary decision: A functional analysis of debates in presidential primaries. Westport, CT: Praeger. *Benoit, W. L., & Stephenson, M. T. (in press). Effects of watching a presidential primary debate. Contemporary Argumentation and Debate. *Benoit, W. L., Webber, D. J., & Berman, J. (1998b). Effects of presidential debate watching and ideology on attitudes and knowledge. Argumentation and Advocacy, 34, 163172. Benoit, W. L., & Wells, W. T. (1996). Candidates in conict: Persuasive attack and defense in the 1992 presidential debates. Tuscaloosa, AL: University of Alabama Press. *Ben-Zeev, S., & White, L. (1962). Effects and implications. In S. Kraus (Ed.), The great debates: Background, perspective, effects (pp. 331337). Bloomington, IN: Indiana University Press. *Best, S. J., & Hubbard, C. (1999). Maximizing minimal effects: The impact of early primary season debates on voter preferences. American Politics Quarterly, 27, 450487. Best, S. J., & Hubbard, C. (2000). The role of televised debates in the presidential nominating process. In W. G. Mayer (Ed.), In pursuit of the White House 2000: How we choose our presidential nominees (pp. 255284). New York: Chatham House. *Bishop, G. F., Oldendick, R., & Tuchfarber, A. (1978). Debate watching and the acquisition of political knowledge. Journal of Communication, 28, 99113. Boster, F. J. (2002). On making progress in communication science. Human Communication Research, 28, 473490. Brians, C. L., & Wattenberg, M. P. (1996). Campaign issue knowledge and salience: Comparing reception from TV commercials, TV news, and newspapers. American Journal of Political Science, 40, 172193. Carlin, D. P. (1994). A rationale for a focus group study. In D. B. Carlin & M. S. McKinney (Eds.), The 1992 presidential debates in focus (pp. 319). Westport, CT: Praeger. *Carter, R. (1962). Some effects of the debates. In S. Kraus (Ed.), The great debates: Background, perspective, effects (pp. 253270). Bloomington, IN: Indiana University Press. *Chaffee, S. H. (1978). Presidential debatesAre the helpful to voters? Communication Monographs, 45, 330346. Coleman, S. (2000). Televised election debates: International perspectives. New York: St. Martins. DAlessio, D., & Allen, M. (2000). Media bias in presidential elections: A meta-analysis. Journal of Communication, 50, 133156. DAlessio, D., & Allen, M. (2002). Selective exposure and dissonance after decisions. Psychological Reports, 91, 527532. *Desutschmann, P. (1962). Viewing, conversation, and voting intentions. In S. Kraus (Ed.), The great debates: Background, perspective, effects (pp. 232252). Bloomington: Indiana University Press. Festinger, L. (1957). A theory of cognitive dissonance. Stanford, CA: Stanford University Press. Friedenberg, R. V. (1994). Patterns and trends in national political debates: 19601992. In R. F. Friedenberg (Ed.), Rhetorical studies of national political debates, 19601992 (2nd ed., pp. 235259). Westport, CT: Praeger. Gallup, G. (1987). The impact of presidential debates on the vote and turnout. In J. L. Swerdlow (Ed.), Presidential debates: 1988 and beyond (pp. 3442). Washington, DC: Congressional Quarterly. Gallup Poll. (1999, April 9). Independents rank as largest U.S. political group. Retrieved June 20, 2001, from http://www.gallup.com/poll/releases/pr990409c.asp *Geer, J. G. (1988). The effects of presidential debates on the electorates preference for candidates. American Politics Quarterly, 16, 486501. Glass, G. V., McGaw, B., & Smith, M. L. (1981). Meta-analysis in social research. Beverly Hills, CA: Sage Publications. *Graber, D. A. (1978). Problems in measuring audience effects of the 1976 debates. In G. F. Bishop, R. G. Meadow, & M. Jackson-Beeck (Eds.), The presidential debates: Media, electoral, and policy perspective (pp. 105125). New York: Praeger. *Holbert, R. L., Benoit, W. L., Hansen, G. J., & Wen, W.-C. (2002a). The role of political ad recall, news use, political discussion, and debate viewing in campaign issue knowledge and salience. Communication Monographs, 69, 296310. Holbert, R. L, Benoit, W. L., & McKinney, M. S. (2002b). The role of debate viewing in establishing enlightened preference in the 2000 presidential election. Seoul, Korea: ICA. Holbrook, T. M. (1999). Political learning from presidential debates. Political Behavior, 21, 6789. Hunter, J. E., & Schmidt, F. L. (1990). Methods of meta-analysis: Correcting error and bias in research ndings. Newbury Park, CA: Sage Publications.

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

Downloaded by [University of Pennsylvania] at 12:17 15 August 2012

350 COMMUNICATION MONOGRAPHS *Jacoby, J., Troutman, T. R., & Whittler, T. E. (1986). Viewer miscomprehension of the 1980 presidential debate: A research note. Political Psychology, 7, 297308. *Jamieson, K. H., & Adasiewicz, C. (2000). What can voters learn from election debates? In S. Coleman (Ed.), Televised election debates: International perspectives (pp. 2542). New York: St. Martins Press. Jamieson, K. H., & Birdsell, D. S. (1988). Presidential debates: The challenge of creating an informed electorate. New York: Oxford University Press. Kaid, L. L., & Johnston, A. (2000). Videostyle in presidential campaigns: Style and content of televised political advertising. Westport, CT: Praeger. *Kraus, S., & Smith, R. G. (1962). Issues and images. In S. Kraus (Ed.), The great debates: Background, perspective, effects (pp. 289313). Bloomington, IN: Indiana University Press. *Lang, K., & Lang, G. E. (1962). Reaction of viewers. In S. Kraus (Ed.), The great debates: Background, perspective, effects (pp. 313330). Bloomington, IN: Indiana University Press. *Lang, G. E., & Lang, K. (1978). Immediate and delayed responses to a CarterFord debate: Assessing public opinion. Public Opinion Quarterly, 42, 322341. *Lanoue, D. J. (1991). The turning point. American Politics Quarterly, 19, 8095. Lau, R. R., Sigelman, L., Heldman, C., & Babbitt, P. (1999). The effects of negative political advertisements: A meta-analytic assessment. American Political Science Review, 93, 851875. Louden, A. (2003). Political debates: Selected bibliography. Retrieved May 12, 2003, from http://www.wfu.edu/ %7Elouden/Political%20Communication/Bibs/DEBATES.html *McKinney, M. S., Kaid, L. L., & Robertson, T. A. (2001). The front-runner, contenders, and also-rans: Effects of watching a 2000 Republican primary debate. American Behavioral Scientist, 44, 22322251. *McLeod, J. M., Durall, J. A., Ziemke, D. A., & Bybee, C. R. (1979). Reactions of young and older voters: Expanding the context of effects. In S. Kraus (Ed.), The great debates: Carter vs. Ford, 1976 (pp. 348367). Bloomington, IN: Indiana University Press. *Mulder, R. D. (1978). The political effects of the CarterFord debate: An experimental analysis. Sociological Focus, 11, 3345. Nie, N. H., Verba, S., & Petrocik, J. R. (1999). The changing American voter (enlarged ed.). San Jose, CA: toExcel/Harvard University Press. *Payne, J. G., Golden, J. L., Marlier, J., & Ratzan, S. C. (1989). Perceptions of the 1988 presidential and vice-presidential debates. American Behavioral Scientist, 32, 425435. *Pfau, M. (1988). Intra-party political debates and issue learning. Journal of Applied Communication Research, 16, 99112. Racine Group. (2002). White paper on televised political campaign debates. Argumentation and Advocacy, 38, 199218. *Rose, D. D. (1976). Citizen uses of the FordCarter debates. Journal of Politics, 41, 214221. Stock, W., Okum, M., Haring, M., Miller, W., Kinney, C., & Williams, R. M. (1982). Rigor in data synthesis: A case study of reliability in meta-analysis. Educational Researcher, 11(6), 1020. *Swanson, L. L., & Swanson, D. L. (1978). The agenda-setting functions of the rst FordCarter debate. Communication Monographs, 45, 347353. *Wall, V., Golden, J. L., & James, H. (1988). Perceptions of the 1984 presidential debates and a select 1988 presidential primary debate. Presidential Studies Quarterly, 18, 541563. Wolf, F. M. (1986). Meta-analysis: Quantitative methods for research synthesis. Beverly Hills, CA: Sage Publications. *Yawn, M., & Beatty, B. (2000). Debate-induced opinion change: What matters? American Politics Quarterly, 28, 270285. *Zhu, J.-H., Milavasky, J. R., & Biswas, R. (1994). Do televised debates affect image perceptions more than issue knowledge? A study of the rst 1992 presidential debate. Human Communication Research, 20, 302333.

Received: January 5, 2003 Revised: July 7, 2003 Accepted: October 25, 2003

Você também pode gostar