all knowledge is not created equal: knowledge effects and the 2012 presidential debates

21
All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates JEFFREY A. GOTTFRIED Pew Research Center BRUCE W. HARDY Annenberg Public Policy Center University of Pennsylvania KENNETH M. WINNEG Annenberg Public Policy Center University of Pennsylvania KATHLEEN HALL JAMIESON Annenberg Public Policy Center University of Pennsylvania Decades of research have confirmed that debates increase viewer knowledge about the issue stands of the candidates. However, the conditions under which viewers learn are less well understood. In this article, we examine how differences in the context of information in presidential debates affect both who learns from such debates and what they learn. Consistent with previous literature, we report that watching the debates increased knowledge of campaign issues and related matters discussed in the debates. We also found that knowledge based on accurate information that was uncontested in the debate was gained at a greater rate than knowledge based on information that was presented by one candidate but contested by the other. And consistent with confirmation bias, learning based on information that was contested in the debate was influenced by viewers’ candidate preferences more often than not. Jeffrey A. Gottfried is a research associate for the Pew Research Center for the People & the Press and the Pew Research Journalism Project. Bruce W. Hardy is a senior researcher in political communication at the Annenberg Public Policy Center of the University of Pennsylvania. Kenneth Winneg is the managing director of Survey Research at the Annenberg Public Policy Center of the University of Pennsylvania. Kathleen Hall Jamieson is the Elizabeth Ware Packard Professor at the Annenberg School for Communication and director of the Annenberg Public Policy Center of the University of Pennsylvania. Presidential Studies Quarterly 44, no. 3 (September) 389 © 2014 Center for the Study of the Presidency

Upload: kathleen-hall

Post on 30-Mar-2017

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

All Knowledge Is Not Created Equal:Knowledge Effects and the 2012

Presidential Debates

JEFFREY A. GOTTFRIEDPew Research Center

BRUCE W. HARDYAnnenberg Public Policy Center

University of Pennsylvania

KENNETH M. WINNEGAnnenberg Public Policy Center

University of Pennsylvania

KATHLEEN HALL JAMIESONAnnenberg Public Policy Center

University of Pennsylvania

Decades of research have confirmed that debates increase viewer knowledge about theissue stands of the candidates. However, the conditions under which viewers learn are less wellunderstood. In this article, we examine how differences in the context of information inpresidential debates affect both who learns from such debates and what they learn. Consistentwith previous literature, we report that watching the debates increased knowledge of campaignissues and related matters discussed in the debates. We also found that knowledge based onaccurate information that was uncontested in the debate was gained at a greater rate thanknowledge based on information that was presented by one candidate but contested by the other.And consistent with confirmation bias, learning based on information that was contested in thedebate was influenced by viewers’ candidate preferences more often than not.

Jeffrey A. Gottfried is a research associate for the Pew Research Center for the People & the Press and the PewResearch Journalism Project.

Bruce W. Hardy is a senior researcher in political communication at the Annenberg Public Policy Center of theUniversity of Pennsylvania.

Kenneth Winneg is the managing director of Survey Research at the Annenberg Public Policy Center of theUniversity of Pennsylvania.

Kathleen Hall Jamieson is the Elizabeth Ware Packard Professor at the Annenberg School for Communication anddirector of the Annenberg Public Policy Center of the University of Pennsylvania.

Presidential Studies Quarterly 44, no. 3 (September) 389© 2014 Center for the Study of the Presidency

Page 2: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

Decades of research have confirmed that debates increase viewer knowledge aboutthe issue stands of the candidates (e.g., Benoit, Hansen, and Verser 2003). However, theconditions under which viewers learn are less well understood. In this article, we examinehow differences in the context of information in presidential debates affect both wholearns from such debates and what they learn. We ask three research questions regardinglearning from exposure to the 2012 general election presidential debates: (1) Did thesedebates increase knowledge of the content discussed in them? (2) Were viewers morelikely to learn about issues and matters relevant to the 2012 presidential election whenone candidate challenged the other’s view or when the candidate’s statement was notcontested? (3) Does learning occur at the same rate regardless of viewer political predis-positions, or did confirmation bias determine whether and which individuals learnedfrom the debates?

Issue Knowledge and “Correct” Voting

For a representative democracy to work, citizens must be able to cast a “correctvote” (Lau and Redlawsk 1997), one in line with their policy preferences. In a world inwhich presidential campaigns do at times mislead (see Jackson and Jamieson 2007),citizens may vote for a candidate on the supposition that their positions align when infact they do not (see Waldman and Jamieson 2006). In 2008, for example, the Obamacampaign used targeted radio ads to deceptively cast Senator John McCain as opposed tofederal funding for stem-cell research (see Kenski, Hardy, and Jamieson 2010). If astem-cell issue voter voted against McCain based on this deceptive message, then an“incorrect vote” was cast.

Because most of us do not personally know national political actors, we rely onmedia and campaign events to provide the necessary information to cast a “correct vote.”One important function of presidential campaigns, then, is to inform voters of candidatepolicy stands and of other factual information that is relevant to the election. In oursystem of government, this function is central because, as Gans (2003) wrote, “Thecountry’s democracy may belong directly or indirectly to its citizens, but the democraticprocess can only be truly meaningful if these citizens are informed” (1). Debates offer aunique opportunity for voters to gain such information.

Debates and Learning

Since incumbent president and Republican nominee Gerald R. Ford challengedDemocrat Jimmy Carter to a series of nationally televised debates in 1976, broadcastdebates have become standard campaign events in U.S. politics. These quadrennial eventsdemand an unmatched level of accountability of those seeking public office. Researchon debate viewing effects has focused primarily on political learning and shifts invote intentions (e.g., Benoit and Hansen 2004; Chaffee 1978; Jamieson and Adasiewicz2000). We focus on the former.

390 | PRESIDENTIAL STUDIES QUARTERLY / September 2014

Page 3: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

Past studies have consistently linked debates viewing with higher levels of poli-tical knowledge. A meta-analysis of 18 studies on debates from 1976 to 2000 (totalN = 7,202) found a mean weighted correlation coefficient of 0.256 between debates andpolitical knowledge, suggesting that on average, debate viewing has a substantial impacton viewer learning of the candidates’ policy positions (Benoit, Hansen, and Verser 2003).Some research suggests that debates increase issue salience and have an agenda-settingeffect (Becker et al. 1978; Benoit and Hansen 2004). Benoit and Hansen (2004) exam-ined the 1976, 1984, 1996, and 2000 data in the American National Election Studies(ANES) and found that debate watchers cited more issues in their evaluations of thecandidates than did nondebate watchers. Consistent with these findings, we hypothesize,

Hypothesis 1: Exposure to the 2012 presidential debates will increase accurate knowledgeof the issues and matters discussed in the debates themselves.

Debate Learning by Type of Information

This study adds to the existing literature by examining whether and how thecontext in which information is discussed within a presidential debate affects rates oflearning. In debates, information can be divided into two categories: that which iscontested and uncontested. Contested information is presented by one candidate andchallenged by the other. Uncontested information is presented without challenge. Wewill treat these information structures as two different informational contexts.

The structure of campaign information can influence the motivation and the abilityof individuals to process it (Rahn, Aldrich, and Borgida 1994). For example, a candidatespeech is a simple structure that is often person-centered and invites easy candidateevaluation. Debates, on the other hand, are more complex, presenting information on anumber of different facets from two or more candidates in an alternating format. Rahn,Aldrich, and Borgida (1994) characterize debates as a dimension-center format andconclude that the more person-centered information contexts, such as the candidatespeeches, require less cognitive effort to process, whereas in the more dimension-centeredcontexts, such as debates, the complex presentation of the information may interfere withthe ability to make sense of the information (Rahn, Aldrich, and Borgida 1994). Thisfinding is consistent with Just, Crigler, and Wallach’s (1990) finding that political ads aremore likely to increase candidate knowledge than debates.

Like the candidate speech, uncontested information provides a single viewpoint ordimension. By contrast, contested information creates a more complex context. Amongother things, the presence of two or more sides to an argument can lead to confusion.Thus, we further hypothesize,

Hypothesis 2: Information introduced by one candidate and not rebutted by the other—uncontested information—is learned at the greater rate than information that is presentedby one candidate but challenged by the opponent—contested information.

Similar to Rahn, Aldrich, and Borgida (1994), we hypothesize that motivation andability to process information in these two different contexts will vary based on viewers’

Gottfried, Hardy, Winneg and Jamieson / KNOWLEDGE EFFECTS | 391

Page 4: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

political sophistication. Because political sophisticates have greater “cognitive dexterity,they should be less constrained by the structure of the information imposed on them”(Rahn, Aldrich, and Borgida 1994, 195). Likewise, since processing contested informa-tion requires greater cognitive effort than uncontested information (see Hypothesis 2) andsince political sophisticates have greater cognitive dexterity, they should be less hobbledby complex information contexts than political nonsophisticates. Thus,

Hypothesis 3: The rate of learning in general and learning from contested exchanges inparticular will be greater among political sophisticates than political nonsophisticates.

Previous studies have found that one’s political beliefs will influence how oneinteracts with and processes political information because of selective attention andexposure (Graf and Aday 2008; Stroud 2008, 2011). But when viewers are exposed to asource that presents information that is both consistent and inconsistent with theirpolitical beliefs, viewers will likely rely on heuristics, such as candidate preference, indetermining which side is accurate. The confirmation bias hypothesis suggests thatpeople bolster “hypotheses or beliefs whose truth is in question,” create “one-sided casebuilding process[es],” and interpret “evidence in ways that are partial to existing beliefs,expectations, or a hypothesis in hand” (Nickerson 1998, 175). When exposed to infor-mation in a debate, particularly information that is contested, confirmation bias shouldbe actively and forcefully at play. The same may not be true of uncontested informationbecause the accuracy of this information is not in question. Specifically, viewers mayrespond to uncontested claims using a simple heuristic that assumes if the statement wasnot accurate, the side disadvantaged by the information would flag the inaccuracy.Therefore, we hypothesize,

Hypothesis 4: When information in a debate is contested, viewers will be more likely toaccept the interpretation offered by their preferred candidate, but uncontested informationwill be accepted regardless of candidate preference.

Method

We analyze data from the Annenberg Public Policy Center’s (APPC) Institutions ofDemocracy 2012 Political Knowledge Survey, a six-wave national cross-sectional tele-phone survey of U.S. adults, 18 years or older, conducted during and after the 2012presidential election. In this article, we use waves 3 and 4 since they were conductedimmediately following the second and third presidential debates. Wave 3 was fieldedbetween October 17 and October 23, 2012, and wave 4 between October 24 and October29, 2012. Throughout the analyses in this article, wave 3 is used to test the hypothesesonly for the second presidential debate, and wave 4 is used to test the hypothesis only forthe third presidential debate, since each wave was fielded immediately following eachrespective debate.

Under contract to APPC, Social Science Research Solutions (SSRS) completed atotal of 1,233 interviews for wave 3 and 1,248 for wave 4 with randomly selected adults

392 | PRESIDENTIAL STUDIES QUARTERLY / September 2014

Page 5: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

contacted through random-digit dialing (RDD) of cell phones (wave 3: N = 422; wave 4:N = 430) and landline telephones (wave 3: N = 811; wave 4: N = 818).1 We oversampledhouseholds in battleground states; 53.0% of the sample (N = 653) in wave 3 and 49.1%of the sample (N = 613) in wave 4 is residents in a battleground state.2

Dependent Variables

Total Knowledge of the Second and Third Presidential Debates. Our focus inthis article is not on recall of what was said in the debates but on the accuracy of theknowledge gained from exposure. Consequently, we rely on a series of knowledge ques-tions that asked respondents about content discussed in the second and third presidentialdebates. In wave 3, respondents were asked 16 items about the second presidentialdebate, and in wave 4, they were asked six items about the third presidential debate. Allsix items asked about the third presidential debate in wave 4 were also asked in wave 3.3

Table 1 presents the wave 3 items, their correct answers, and the percent of the samplethat answered each item correctly;4 Table 2 presents the same for the items in wave 4.Each item was first coded to indicate whether a respondent answered it correctly or not.5

Scales were then created to indicate the percent of total items answered correctly aboutthe second presidential debate from wave 3 (mean = 55.24%; SD = 18.62) and the thirdpresidential debate from wave 4 (mean = 51.38%; SD = 21.58).

Contested and Uncontested Debate Knowledge. The knowledge items were cate-gorized into two groupings: knowledge based on information that was contested andknowledge based on accurate information that was not contested in either debate. The

1. All interviews were conducted by live interviewers using computer-assisted telephone interview-ing (CATI) software. In households reached by landline, an adult was selected at random based on their age(or most recent birthday). The questionnaire was delivered in Spanish for those who selected that option.Nonresponsive phone numbers were contacted up to six times, and in cases where initial attempts were metwith soft refusal (i.e., abrupt hang-ups), refusal-conversion attempts were made. Accounting for designeffects produced by survey weights, the margin of error is +/-3.7 percent for wave 3 and +/-3.6 percent forwave 4. The response rate was 13% for wave 3 and 11% for wave 4. Response rate uses the standard AAPORRR3 formula.

2. These percentages are not weighted; every other raw percent in this article is weighted by nationalpopulation parameters. Battleground states were defined as Colorado, Florida, Iowa, Michigan, Nevada, NewHampshire, North Carolina, Pennsylvania, Ohio, Virginia, and Wisconsin. The total share of adults residingin these states is 28.9% based on the 2012 March Supplement of the U.S. Census Bureau’s CurrentPopulation Survey (CPS) (http://www.census.gov/cps).

3. Questions were chosen for inclusion in the debate knowledge scales only if they were directly andexplicitly discussed by either candidate in the second and third presidential debates. If a question was notdirectly and explicitly discussed in the debate, then it was not included in the scale for that debate. Therewere no questions that met this requirement that were omitted from the analysis. The difference in thenumber of questions used for the scales between the second and third debates is due to there being fewerquestions asked that met this requirement for the third debate.

4. To determine whether a candidate’s statement about his or his opponent’s future plans wereaccurate in the debates, we relied on candidate statements in speeches and on the candidates’ web site. Manyof the issues and claims about past actions or the content of speeches asked about in our survey were discussedby the four major national fact-checking organizations (Washington Post’s Fact Checker, Politifact, FactCheck.org,and the Associated Press). For those questions in which the issues and claims were discussed by theseorganizations, we adopted their conclusions to determine the correct answer to each of our questions.

5. Do not know responses were coded as not correct. Refusals were coded as missing.

Gottfried, Hardy, Winneg and Jamieson / KNOWLEDGE EFFECTS | 393

Page 6: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

TABLE 1Knowledge Items Asked of Second Presidential Debate (Wave 3)

Question Wording Correct Answer % Correct

Knowledge Based on Contested Information1 Mitt Romney’s tax plan will increase the deficit by over 4 and a half trillion

dollars. How accurate do you think that statement is? Would you say it is veryaccurate, somewhat accurate, not too accurate, or not accurate at all?

Not Accurate 46.9%

2 Has oil production on federal lands increased, decreased, or stayed the sameunder President Obama?

Increased 30.3%

3 Governor Romney wanted to take Detroit auto manufacturers into bankruptcywithout providing any form of federal aid. How accurate do you think thatstatement is? Would you say it is very accurate, somewhat accurate, not tooaccurate, or not accurate at all?

Not Accurate 24.2%

4 No new jobs have been created as a result of the stimulus which was passedin the first year of Obama’s presidency. How accurate do you think thatstatement is? Would you say it is very accurate, somewhat accurate, not tooaccurate, or not accurate at all?

Not Accurate 56.7%

Knowledge Based on Accurate Uncontested Information5 Which candidate would push for passage of the assault weapons ban? Barack

Obama, Mitt Romney, both, or neither?Barack Obama 49.2%

6 Which candidate favors branding China as a currency manipulator andimposing tariffs on Chinese goods? Barack Obama, Mitt Romney, Both,or Neither?

Mitt Romney 58.0%

7 In early October, the Bureau of Labor Statistics announced that the nationalunemployment rate dropped below eight percent. How accurate do you thinkthat statement is? Would you say it is very accurate, somewhat accurate, nottoo accurate, or not accurate at all?

Accurate 65.5%

8 Which candidate has promised to increase military spending? Barack Obama,Mitt Romney, Both, or Neither?

Mitt Romney 58.1%

9 Which candidate proposes raising federal income taxes on households earning250 thousand dollars or more per year? Barack Obama, Mitt Romney, both, orneither?

Barack Obama 57.9%

10 Which candidate says he will keep the Bush tax cuts in place permanently?Barack Obama, Mitt Romney, both, or neither?

Mitt Romney 55.3%

11 Which candidate favors building the Keystone pipe line without delay? BarackObama, Mitt Romney, both, or neither?

Mitt Romney 65.5%

12 Mitt Romney has investments in Chinese companies. How accurate do youthink that statement is? Would you say it is very accurate, somewhat accurate,not too accurate, or not accurate at all?

Accurate 77.3%

13 Which candidate supports the Dream act which provides a path to permanentresidence status for young undocumented immigrants who were brought tothis country illegally as children. Barack Obama, Mitt Romney, both, orneither?

Barack Obama 73.6%

14 Which candidate favors eliminating funding for the Public BroadcastingService, also known as PBS? Barack Obama, Mitt Romney, both, or neither?

Mitt Romney 72.9%

15 The Obama administration cut banks out as the middlemen in the studentloan program and put the money saved back into student loans. How accuratedo you think that statement is? Would you say it is very accurate, somewhataccurate, not too accurate, or not accurate at all?

Accurate 63.2%

Based on Neither Contested nor Accurate Uncontested Information16 While in foreign countries, Barack Obama has repeatedly apologized for

America. How accurate do you think that statement is? Would you say it isvery accurate, somewhat accurate, not too accurate, or not accurate at all?

Not Accurate 30.9%

394 | PRESIDENTIAL STUDIES QUARTERLY / September 2014

Page 7: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

categorization of each item is dependent only on whether the information was contestedor uncontested in the second presidential debate for the wave 3 items and in the thirdpresidential debate for the wave 4 items, and not whether it was contested or uncontestedin any other form of communication, including other debates. A question based oncontested information asked, “Has oil production on federal lands increased, decreased,or stayed the same under President Obama?” In the debate, Obama stated, “So here’swhat I’ve done since I’ve been president. We have increased oil production to the highestlevels in 16 years.” Romney followed up this claim by arguing, “As a matter of fact, oilproduction is down 14 percent this year on federal land, and gas production is down 9percent. Why? Because the president cut in half the number of licenses and permits fordrilling on federal lands and in federal waters.” A question drawing on accurate uncon-tested information from the second debate is “Which candidate proposes raising federalincome taxes on households earning 250 thousand dollars or more per year? BarackObama, Mitt Romney, Both, or Neither?” In this debate, Obama clearly says that heplans to raise federal income taxes on households earning $250,000 a year, and thestatement was not challenged by Romney:

PRESIDENT OBAMA: “[I]n addition to some tough spending cuts, we’ve also got to makesure that the wealthy do a little bit more. So what I’ve said is your first $250,000 worth ofincome, no change. And that means 98 percent of American families, 97 percent of small

TABLE 2Knowledge Items Asked of Third Presidential Debate (Wave 4)

Question Wording Correct Answer % Correct

Knowledge Based on Contested Information1 While in foreign countries, Barack Obama has repeatedly apologized for

America. How accurate do you think that statement is? Would you say itis very accurate, somewhat accurate, not too accurate, or not accurate atall?

Not Accurate 36.7%

2 Mitt Romney’s tax plan will increase the deficit by over 4 and a halftrillion dollars. How accurate do you think that statement is? Wouldyou say it is very accurate, somewhat accurate, not too accurate, or notaccurate at all?

Not Accurate 45.1%

3 Governor Romney wanted to take Detroit auto manufacturers intobankruptcy without providing any form of federal aid. How accurate doyou think that statement is? Would you say it is very accurate, somewhataccurate, not too accurate, or not accurate at all?

Not Accurate 36.0%

Knowledge Based on Accurate Uncontested Information4 Which candidate has promised to increase military spending? Barack

Obama, Mitt Romney, Both, or NeitherMitt Romney 58.7%

5 Which candidate favors branding China as a currency manipulator andimposing tariffs on Chinese good? Barack Obama, Mitt Romney, Both,or Neither?

Mitt Romney 58.4%

6 Mitt Romney has investments in Chinese companies. How accurate doyou think that statement is? Would you say it is very accurate, somewhataccurate, not too accurate, or not accurate at all?

Accurate 73.8%

Gottfried, Hardy, Winneg and Jamieson / KNOWLEDGE EFFECTS | 395

Page 8: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

businesses, they will not see a tax increase. I’m ready to sign that bill right now. . . . Butwhat I’ve also said is for above 250,000 (dollars), we can go back to the tax rates we hadwhen Bill Clinton was president.”

For the second presidential debate (wave 3), four items were categorized as based oninformation that was contested and 11 were based on accurate information that wasuncontested. One item was categorized as neither based on contested nor on uncontestedaccurate information and thus not included in either scale. For the third presidentialdebate (wave 4), three items were classed as based on information that was contested inthe debate and three on accurate information that was uncontested. The categorizationof each item is also presented in Tables 1 and 2. Scales were created in both waves thatindicate the percent of items based on contested information answered correctly (wave 3:mean = 39.52%, SD = 22.97; wave 4: mean = 39.21%, SD = 26.72) and based on accurateuncontested information answered correctly (wave 3: mean = 63.21%, SD = 23.61; wave4: mean = 63.62%, SD = 31.98).

Independent Variables

Second and Third Presidential Debate Viewership. Respondents were asked inwave 3 how much of the second presidential debate they watched, and respondents inwave 4 were asked how much of the third presidential debate they watched. These twovariables are on a four-point scale ranging from 0 (none) to 3 (all) (wave 3: mean = 1.73,SD = 1.25; wave 4: mean = 1.48; SD = 1.27).

Political Sophistication. In the same vein as Rahn, Aldrich, and Borgida (1994),we created a composite index from several different measures for our measure of politicalsophistication. We standardized and took the mean of three measures: total news mediaconsumption, how closely respondents were following the election, and basic campaignknowledge. Total news media consumption was measured by taking the mean of fiveitems that indicate the days/week that individuals watch local news, broadcast nationalnightly news, their most frequently watched cable news channel, listened to talk radio,or read a newspaper. The scale ranges from 0 (very low news media consumption) to7 (very high news media consumption) (wave 3: mean = 3.14, SD = 1.68; wave 4:mean = 2.97, SD = 1.69). Closely following was measured with a single item that askedrespondents “How closely are you following the 2012 presidential campaign?” rangingfrom 1 (not closely at all) to 4 (very closely) (wave 3: mean = 3.31, SD = 0.89; wave 4:mean = 3.26, SD = 0.96). Basic campaign knowledge was measured with two items thatasked respondents to name the current vice president of the United States and theRepublican vice presidential candidate. A three-point scale ranging from 0 to 2 correctanswers was created (wave 3: mean = 1.45, SD = 0.80; wave 4: mean = 1.47, SD = 0.76).The standardized scale from these three items was split at the median to determinewhether respondents were political sophisticates or nonsophisticates. The measure variedsomewhat from Rahn, Aldrich, and Borgida’s (1994); notably, we did not include ameasure of participation in political activities.

396 | PRESIDENTIAL STUDIES QUARTERLY / September 2014

Page 9: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

Candidate Favorability. This variable was constructed by subtracting a feelingthermometer ranging from 0 to 10 for Mitt Romney from that of Barack Obama, creatinga scale ranging from -10 (very high favorability of Romney over Obama) to 10 (veryhigh favorability of Obama over Romney) (wave 3: mean = 0.58, SD = 7.16; wave 4:mean = 0.54, SD = 7.07).

Analyses

First, for the test of Hypothesis 1, multiple linear regression models predictingtotal debate knowledge from debate viewing were created. Next, for the tests of Hypoth-esis 2, linear regression models were conducted to predict the scales for knowledge basedon information that was contested in the debates and knowledge based on accurateinformation that was uncontested in the debates. For the test of Hypothesis 3, multiplelinear regression models were constructed to predict each of the three knowledge scalesfrom debate viewing, political sophistication, and the interaction between debateviewing and political sophistication. Finally, for the test of Hypothesis 4, two logisticregression models were conducted for each individual knowledge item to test whetherthe effect of debate viewing varies by candidate favorability; one model predictedcorrect knowledge of each item from debate viewership, and a second included candi-date favorability and an interaction between debate viewership and candidatefavorability. In addition to the independent variables already discussed, each modelcontrols for total news media consumption, basic campaign knowledge,6 residence in abattleground state, political party identification, political ideology, years of education,race, gender, and age.7

Results

Table 3 presents the models that predict each of the three knowledge scales of thesecond presidential debate (wave 3)—total knowledge, knowledge based on informationthat was contested, and knowledge based on accurate information that was uncontested.Table 4 presents the same for the third presidential debate (wave 4). The first model inboth Tables 3 and 4 tests Hypothesis 1, which predicts that debate exposure will increasetotal knowledge of content discussed in the debate. Consistent with the hypothesis, wefind that reported debate viewership increased total knowledge in both debates; reportingwatching all of the second debate increased the percent of questions answered correctly by

6. The models for Hypotheses 3 do not include total news media consumption and basic campaignknowledge since the measure of political sophistication already includes these two measures.

7. The following is the demographic breakdown of wave 3: political party identification—25.8%Republican, 33.6% Democrat; political ideology (1 = very liberal; 5 = very conservative)—mean = 3.15,SD = 1.26; years of education—mean = 13.82; SD = 2.51; race—12.4% black, 12.5% Hispanic; gender—51.4% female; age—mean = 47.78, SD = 17.80. The following is the demographic breakdown of wave 4:political party identification—22.6% Republican, 34.5% Democrat; political ideology (1 = very liberal;5 = very conservative)—mean = 3.27, SD = 1.19; years of education—mean = 13.84; SD = 2.50; race—13.4% black, 13.5% Hispanic; gender—51.7% female; age—mean = 47.14, SD = 17.87.

Gottfried, Hardy, Winneg and Jamieson / KNOWLEDGE EFFECTS | 397

Page 10: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

11.9% compared to not watching any of the debate (t = 9.88, p < .001), and reportedviewing all of the third debate increased the number of questions answered correctly by15.2% compared to not watching any of that debate (t = 10.53, p < .001).8

It should first be noted that respondents answered more of the questions based onaccurate uncontested information correctly than those based on contested information;63.2% (wave 3) and 63.6% (wave 4) of the items based on accurate uncontested infor-mation were answered correctly compared to 39.5% (wave 3) and 39.2% (wave 4) ofitems based on information that was contested.

The second and third models in Tables 3 and 4 test Hypothesis 2, which predicts thatdebate exposure will increase knowledge based on information that was uncontested inthe debates more so than knowledge based on information that was contested. First, wefind strong evidence from both waves that viewing the debates increased accurate infor-mation that was uncontested. Watching all of the second debate increased knowledge ofthe accurate uncontested information by 16.0% (t = 10.53, p < .001) compared to notwatching any of that debate, and watching all of the third debate increased accurate

8. For the models in Tables 3 and 4, we did not control for closely following the election because whilethis variable may be a confounding variable, it also theoretically can serve as a mediating variable between debateviewership and knowledge. That said, we ran an additional set of models with closely following as a control.In these models the magnitude of the coefficient for debate viewership across models is slightly depressed, butin the same direction, and significant where they were significant in Tables 3 and 4.

TABLE 3Effect of Watching the Second Presidential Debate on Knowledge of Issues and MattersDiscussed in the Debate (Wave 3)

Total KnowledgeKnowledge Based on

Contested Information

Knowledge Based onAccurate Uncontested

Information

b std err b std err b std err

Amount of Second PresidentialDebate Watched

3.98*** 0.40 0.75 0.61 5.32*** 0.51

Total News Media Consumption 0.17 0.30 −0.09 0.45 0.25 0.38Basic Campaign Knowledge 7.41*** 0.69 1.03 1.04 10.33*** 0.87Lives in Battleground State 0.21 0.88 −2.18 1.32 1.04 1.10Republican −1.55 1.15 1.03 1.73 −1.67 1.44Democrat 2.69* 1.11 1.14 1.66 1.89 1.39Conservative −1.44*** 0.41 0.33 0.61 −1.13* 0.51Education (years) 0.94*** 0.19 0.40 0.29 1.07*** 0.24Hispanic −1.96 1.78 2.03 2.67 −3.63 2.23Black −1.07 1.53 0.23 2.29 −0.98 1.91Female −4.62*** 0.89 −1.25 1.33 −6.37*** 1.11Age −0.08** 0.03 −0.08 0.04 −0.08* 0.04

Constant 35.04 34.89 35.00N 1,108 1,108 1,108R2 0.32 0.02 0.34

* p < .05, ** p < .01, *** p < .001.Source: Annenberg Public Policy Center’s Institutions of Democracy 2012 Political Knowledge Survey.

398 | PRESIDENTIAL STUDIES QUARTERLY / September 2014

Page 11: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

uncontested knowledge by 22.2% (t = 10.12, p < .001) compared to not watching any ofthat debate. Additionally, we find evidence from wave 4 that debate viewing increasedknowledge based on information that was contested. Watching all of the third presiden-tial debate increased contested knowledge by 8.4% (t = 4.35, p < .001) compared to notwatching that debate at all. While the coefficient for the effect of watching the secondpresidential debate on knowledge based on contested information was also positive inwave 3, it does not meet conventional standards of statistical significance.

Further, consistent with the hypothesis, the coefficient of debate viewership sig-nificantly differs between the two knowledge scales for both debates; the effect ofdebate viewership was greater on knowledge based on accurate uncontested informa-tion than on knowledge based on contested information (wave 3: F = 36.95, p < .001;wave 4: F = 21.76, p < .001). The predicted values for the effects of debate watchingon the two knowledge scales for the second presidential debate are presented inFigure 1, and for the third presidential debate, they are presented in Figure 2. Theslope of the lines for both knowledge based on accurate uncontested information andknowledge based on contested information increases as the amount of debate viewingdoes (though not significantly for knowledge based on contested information for thesecond presidential debate), but at a greater rate for knowledge based on accurateuncontested information.

TABLE 4Effect of Watching the Third Presidential Debate on Knowledge of Issues and Matters Discussedin the Debate (Wave 4)

Total KnowledgeKnowledge Based on

Contested Information

Knowledge Based onAccurate Uncontested

Information

b std err b std err b std err

Amount of Third PresidentialDebate Watched

5.07*** 0.48 2.80*** 0.64 7.38*** 0.73

Total News Media Consumption 0.92* 0.38 0.50 0.51 1.30* 0.58Basic Campaign Knowledge 5.73*** 0.93 1.98 1.24 9.60*** 1.41Lives in Battleground State 0.10 1.08 0.46 1.45 −0.41 1.64Republican 3.22* 1.43 8.16*** 1.91 −1.79 2.17Democrat −2.40 1.37 −3.15 1.83 −1.45 2.08Conservative 0.62 0.52 1.64* 0.70 −0.30 0.79Education (years) 0.90*** 0.25 1.07*** 0.33 0.71 0.37Hispanic 0.59 1.94 6.15* 2.59 −5.32 2.94Black −1.38 1.99 −0.07 2.65 −2.79 3.01Female −3.03** 1.11 −2.72 1.49 −3.12 1.69Age −0.05 0.03 −0.02 0.05 −0.08 0.05

Constant 23.28 11.42 34.74N 1,112 1,112 1,111R2 0.25 0.10 0.22

* p < .05, ** p < .01, *** p < .001.Source: Annenberg Public Policy Center’s Institutions of Democracy 2012 Political Knowledge Survey.

Gottfried, Hardy, Winneg and Jamieson / KNOWLEDGE EFFECTS | 399

Page 12: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

Tables 5 and 6 present the models for the test of Hypothesis 3, which predict thatpolitical sophisticates will learn from the debates at a greater rate than nonsophisticates,particularly for the knowledge based on contested information. First, we find thatpolitical sophisticates were significantly more knowledgeable than nonsophisticates fortotal knowledge and knowledge based on accurate uncontested information in bothwaves. While also in the expected positive direction, the coefficients for being a politicalsophisticate in the models that predict knowledge based on contested information are notsignificant in either wave. Additionally, there is no evidence in any of the models that theeffect of debate viewership on total knowledge, knowledge based on contested informa-tion, or knowledge based on accurate uncontested information varied between politicalsophisticates and nonsophisticates, as seen with the nonsignificant interaction terms.Thus, we find no evidence for Hypothesis 3.

To begin understanding why there are the differences in the rates of learning of eachtype of knowledge, as found in the test of Hypothesis 2, we turn to the logistic regressionmodels in Table 7 that predict each knowledge item individually. There are two separatemodels for each item; the first column for each item is a model that predicts havinganswered the item correctly from debate viewing, and the second column includescandidate favorability and the interaction between debate viewing and candidatefavorability.

FIGURE 1. Effect of Watching the Second Presidential Debate on Knowledge Based onInformation Contested in Debate and Accurate Information Uncontested in the Debate.

400 | PRESIDENTIAL STUDIES QUARTERLY / September 2014

Page 13: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

From the models in the first column for each item, we find that two of the seven(28.6%) items based on contested information across the two debates (zero of four itemsfrom the second debate and two of the three from the third debate) were significantlyand positively predicted by debate viewing. Comparatively, 11 of the 14 (78.6%) itemsbased on accurate uncontested information (eight of 11 items from the second debateand all three from the third debate) were significantly predicted by debate viewing.Information complexity, then, provides one explanation for why the rate of learningcontested information is lower than uncontested accurate information. When accurateinformation is rebutted during a debate, complexity increases, and audiences are lesslikely to learn.9

These differences in the rates of learning based on contested compared to uncon-tested information also presumably occur because viewers assume that argumentsadvanced by their candidate are accurate regardless of whether or not they are, as stated

9. For the models in Table 7, we did not control for closely following the election for reasonspreviously mentioned. That said, we ran additional models with closely following as a control. The onlymentionable differences for the second debate items are that the main effect of debate viewing in thenoninteraction models of items 5 and 9 falls below significance, and the interaction between candidatefavorability and debate watching becomes significant for item 11. For the third presidential debate, theonly mentionable difference is that the main effect for debate viewership in the noninteraction model foritem 1 falls below significance.

FIGURE 2. Effect of Watching the Third Presidential Debate on Knowledge Based on InformationContested in the Debate and Accurate Information Uncontested in the Debate.

Gottfried, Hardy, Winneg and Jamieson / KNOWLEDGE EFFECTS | 401

Page 14: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

in Hypothesis 4, which predicts that viewers would be more likely to learn the interpre-tation offered by their preferred candidate among the information that is contested in thedebates. The models for the test of this hypothesis are presented in the second column foreach item in Table 7. Starting with the knowledge based on accurate uncontestedinformation, there is only evidence for four of the 14 (28.6%) items that learning variedby respondents’ preference for one candidate over the other. Thus, for the items based oninformation that was accurate and uncontested in the debates, learning from debateviewing occurred for a vast majority of these items and most often occurred at the samerate regardless of which candidate the respondents supported.

While learning often did not vary by candidate preference among the knowledgeitems based on information that was accurate and uncontested in the debates, learningoften did vary by candidate preference among the knowledge items based on informationthat was contested in the debates. Among four of the seven (57.1%) knowledge itemsbased on contested information—two of the four from the second presidential debate andtwo of three from the third presidential debate—the rate of learning varied with candi-date preference. A positive interaction indicates that Obama supporters learned thatspecific piece of information at a greater rate than Romney supporters, while a negativeinteraction indicates that Romney supporters learned at a greater rate than Obamasupporters. Nonsignificant interaction terms indicate that the rate of learning, if any,does not differ by candidate preference. Of the two contested items that are significantly

TABLE 5Models Predicting Knowledge of Issues and Matters Discussed in the Second Presidential Debateby Debate Viewership and Political Sophistication (Wave 3)

Total KnowledgeKnowledge Based on

Contested Information

Knowledge Based onAccurate Uncontested

Information

b std err b std err b std err

Amount of Second PresidentialDebate Watched

4.13*** 0.53 1.01 0.77 5.51*** 0.67

Political Sophisticate 7.72*** 2.06 4.56 3.00 9.93*** 2.60Debate Watched * Sophisticate −0.36 0.87 −1.28 1.27 −0.26 1.10Lives in Battleground State 0.47 0.90 −2.12 1.31 1.41 1.14Republican −0.70 1.18 1.03 1.72 −0.45 1.49Democrat 2.68* 1.14 1.11 1.66 1.89 1.44Conservative −1.47*** 0.42 0.37 0.61 −1.19* 0.53Education (years) 1.35*** 0.90 0.41 0.28 1.66*** 0.24Hispanic −3.36 1.83 1.75 2.67 −5.56* 2.32Black −2.67 1.55 −0.08 2.26 −3.20 1.97Female −4.90*** 0.92 −1.20 1.33 −6.79*** 1.16Age −0.11*** 0.03 −0.09* 0.04 −0.13*** 0.04

Constant 39.77 35.51 41.52N 1,109 1,109 1,109R2 0.28 0.02 0.29

* p < .05, ** p < .01, *** p < .001.Source: Annenberg Public Policy Center’s Institutions of Democracy 2012 Political Knowledge Survey.

402 | PRESIDENTIAL STUDIES QUARTERLY / September 2014

Page 15: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

predicted by debate viewing, both vary by respondent candidate preference, suggestingthat when learning occurred overall, it occurred at a depressed rate. Further, among twoof the contested items in which learning overall did not occur, the effect of debateviewing varied by respondent disposition toward the candidates. In other words, for thesetwo items, learning of correct and incorrect answers both occurred, but the overall effectsof debate viewing on knowledge was cancelled out by partisan preference. Because bothof these items were from the second presidential debate, this explains why there was nosignificant effect of debate viewing on knowledge of information based on contestedinformation in the second debate. For the remaining three contested knowledge items,there is no evidence of learning of any kind. Overall, these results provide evidence forconfirmation bias in that learning information that is contested is largely affected bycandidate preference.

Further, the directions of the interaction terms suggest that those who preferredObama over Romney gained accurate knowledge about the contested items in which thecorrect answer was favorable to the Obama candidacy, and those who preferred Romneyover Obama gained accurate knowledge about the contested items in which the correctanswer was favorable to the Romney candidacy. For example, for the item that asks theaccuracy of the statement, “While in foreign countries, Barack Obama has repeatedlyapologized for America” (third debate, wave 4), the positive interaction suggests that asfavorability for Obama over Romney increased, so did the likelihood of reporting the

TABLE 6Models Predicting Knowledge of Issues and Matters Discussed in the Third Presidential Debateby Debate Viewership and Political Sophistication (Wave 4)

Total KnowledgeKnowledge Based on

Contested Information

Knowledge Based onAccurate Uncontested

Information

b std err b std err b std err

Amount of Third PresidentialDebate Watched

6.04*** 0.64 3.41*** 0.85 8.78*** 0.98

Political Sophisticate 7.41*** 2.21 1.67 2.93 13.04*** 3.35Debate Watched * Sophisticate −1.39 0.99 −0.44 1.31 −2.41 1.50Lives in Battleground State 0.18 1.10 0.77 1.46 −0.45 1.67Republican 3.32* 1.46 8.18*** 1.93 −1.66 2.21Democrat −2.80* 1.39 −3.33 1.84 −2.22 2.11Conservative 0.55 0.53 1.55* 0.70 −0.40 0.80Education (years) 1.18*** 0.24 1.15*** 0.32 1.20*** 0.37Hispanic −0.28 1.96 5.68* 2.60 −6.63* 2.98Black −1.94 1.98 0.25 2.63 −4.29 3.01Female −3.13** 1.13 −3.00* 1.49 −3.12 1.71Age −0.05 0.03 −0.01 0.04 −0.08 0.05

Constant 28.35 13.91 42.54N 1,116 1,116 1,115R2 0.23 0.10 0.19

* p < .05, ** p < .01, *** p < .001.Source: Annenberg Public Policy Center’s Institutions of Democracy 2012 Political Knowledge Survey.

Gottfried, Hardy, Winneg and Jamieson / KNOWLEDGE EFFECTS | 403

Page 16: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

TA

BL

E7

Log

isti

cR

egre

ssio

nM

odel

sP

red

icti

ng

Eac

hK

now

led

geIt

emfr

omD

ebat

eV

iew

ing

and

Can

did

ate

Fav

orab

ilit

y

Deb

ate

2It

ems

Bas

edon

Con

test

edIn

form

atio

n

1†2

34

Am

ount

ofSe

cond

Pre

side

ntia

lD

ebat

eW

atch

ed0.

030.

000.

120.

06−0

.05

−0.0

90.

030.

06C

andi

date

Fav

orab

ilit

y—

−0.1

4***

—.0

8**

—−0

.06*

—.1

1***

Fav

orab

ilit

y*

Deb

ate

Wat

ch—

−0.0

2*—

0.02

*—

−0.0

1—

0.00

Con

stan

t−1

.47

−0.4

4−0

.26

−0.9

1−0

.34

0.28

−0.4

6−1

.22

N1,

107

1,09

61,

106

1,09

51,

106

1,09

51,

106

1,09

5M

cFad

den

R2

.14

.24

.10

.16

.08

.10

.09

.14

Deb

ate

2It

ems

Bas

edon

Acc

urat

eU

ncon

test

edIn

form

atio

n

56

78

Am

ount

ofSe

cond

Pre

side

ntia

lD

ebat

eW

atch

ed0.

20**

0.18

**0.

44**

*0.

47**

*0.

120.

14*

0.41

***

0.40

***

Can

dida

teF

avor

abil

ity

—−0

.04

—0.

00—

0.11

***

—0.

00F

avor

abil

ity

*D

ebat

eW

atch

—0.

02*

—−0

.04*

**—

0.00

—0.

01C

onst

ant

−0.9

6−0

.83

−2.1

1−1

.79

0.86

0.18

−1.8

0−1

.90

N1,

107

1,09

61,

107

1,09

61,

107

1,09

71,

107

1,09

6M

cFad

den

R2

.05

.05

.11

.14

.08

.11

.11

.11

Deb

ate

2It

ems

Bas

edon

Acc

urat

eU

ncon

test

edIn

form

atio

n

910

1112

Am

ount

ofSe

cond

Pre

side

ntia

lD

ebat

eW

atch

ed0.

19**

0.18

**0.

010.

00.5

3***

0.54

***

.46*

**.4

8***

Can

dida

teF

avor

abil

ity

—0.

05*

—−0

.01

—−0

.03

—0.

07**

Fav

orab

ilit

y*

Deb

ate

Wat

ch—

−0.0

1—

0.01

—−0

.02

—−0

.01

Con

stan

t−1

.93

−2.1

3−1

.23

−1.3

4−1

.59

−1.2

30.

870.

57N

1,10

71,

096

1,10

61,

095

1,10

71,

096

1,10

41,

093

McF

adde

nR

2.1

0.1

0.0

7.0

7.2

5.2

6.0

8.0

9

*p

<.0

5;**

p<

.01;

***

p<

.001

.N

otes

:A

llm

odel

sco

ntro

lfo

rto

tal

new

sm

edia

cons

umpt

ion,

basi

cca

mpa

ign

know

ledg

e,li

ves

inba

ttle

grou

ndst

ate,

part

yid

enti

ficat

ion,

poli

tica

lid

eolo

gy,

educ

atio

n,ra

ce,

gend

er,

and

age.

†Q

uest

ion

wor

ding

for

each

item

can

befo

und

inTa

ble

1.T

henu

mbe

rsco

rres

pond

toea

chre

spec

tive

item

inTa

ble

1.So

urce

:A

nnen

berg

Pub

lic

Pol

icy

Cen

ter’

sIn

stit

utio

nsof

Dem

ocra

cy20

12P

olit

ical

Kno

wle

dge

Surv

ey.

404 | PRESIDENTIAL STUDIES QUARTERLY / September 2014

Page 17: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

TA

BL

E7

Con

tin

ued

Deb

ate

2It

ems

Bas

edon

Acc

urat

eU

ncon

test

edIn

form

atio

n

13†

1415

Am

ount

ofSe

cond

Pre

side

ntia

lD

ebat

eW

atch

ed0.

030.

04.3

9***

0.37

***

0.34

***

0.38

***

Can

dida

teF

avor

abil

ity

—0.

02—

0.08

**—

0.07

**F

avor

abil

ity

*D

ebat

eW

atch

—0.

01—

−0.0

2*—

0.01

Con

stan

t−1

.74

−1.9

1−0

.99

−1.0

82.

161.

72N

1,10

71,

096

1,10

31,

092

1,10

51,

094

McF

adde

nR

2.0

8.0

8.1

9.1

9.1

0.1

4

Deb

ate

3It

ems

Bas

edon

Con

test

edIn

form

atio

n

12

3

Am

ount

ofT

hird

Pre

side

ntia

lD

ebat

eW

atch

ed0.

16*

0.10

0.06

0.04

.22*

**0.

16*

Can

dida

teF

avor

abil

ity

—0.

11**

*—

−0.1

3***

—−0

.09*

**F

avor

abil

ity

*D

ebat

eW

atch

—0.

05**

*—

−0.0

1—

−0.0

2*C

onst

ant

−1.5

0−2

.59

−2.4

3−1

.59

−1.9

1−1

.05

N1,

109

1,10

31,

112

1,10

61,

109

1,10

3M

cFad

den

R2

.25

.36

.18

.25

.14

.19

Deb

ate

3It

ems

Bas

edon

Acc

urat

eU

ncon

test

edIn

form

atio

n

45

6

Am

ount

ofT

hird

Pre

side

ntia

lD

ebat

eW

atch

ed0.

38**

*0.

39**

*0.

42**

*0.

46**

*0.

33**

*0.

37**

*C

andi

date

Fav

orab

ilit

y—

0.02

—0.

01—

0.04

*F

avor

abil

ity

*D

ebat

eW

atch

—−0

.01

—−0

.04*

**—

0.01

Con

stan

t−1

.82

−1.7

8−1

.69

−1.4

01.

571.

22N

1,11

01,

104

1,10

91,

104

1,10

51,

099

McF

adde

nR

2.1

2.1

2.1

2.1

2.0

6.0

7

*p

<.0

5;**

p<

.01;

***

p<

.001

.N

otes

:A

llm

odel

sco

ntro

lfo

rto

tal

new

sm

edia

cons

umpt

ion,

basi

cca

mpa

ign

know

ledg

e,li

ves

inba

ttle

grou

ndst

ate,

part

yid

enti

ficat

ion,

poli

tica

lid

eolo

gy,

educ

atio

n,ra

ce,

gend

er,

and

age.

†Q

uest

ion

wor

ding

for

each

item

can

befo

und

inTa

bles

1an

d2.

The

num

bers

corr

espo

ndto

each

resp

ecti

veit

emin

Tabl

e1

for

the

seco

ndpr

esid

enti

alde

bate

and

2fo

rth

eth

ird

pres

iden

tial

deba

te.

Sour

ce:

Ann

enbe

rgP

ubli

cP

olic

yC

ente

r’s

Inst

itut

ions

ofD

emoc

racy

2012

Pol

itic

alK

now

ledg

eSu

rvey

.

Gottfried, Hardy, Winneg and Jamieson / KNOWLEDGE EFFECTS | 405

Page 18: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

correct answer from viewing the third debate. This relationship can be seen in Figure 3.10

Conversely, for the statement that asks about the accuracy of the statement, “GovernorRomney wanted to take Detroit auto manufacturers into bankruptcy without providingany form of federal aid” (third debate, wave 4), the negative interaction suggests thatas support for Romney over Obama increased, the rate of reporting the correct answerincreased.11 This relationship can be seen in Figure 4.

Discussion

This study explores how the context of information in debates influences rates oflearning and who learns. First, consistent with previous literature, we report that watch-ing the debates increased knowledge of campaign issues and related matters discussed in

10. Figures 3 and 4 were created by plotting the predicted values for these two knowledge items atthree different intervals of candidate favorability: completely favorable of Obama over Romney (i.e., 10 onthe candidate favorability scale), completely favorable of Romney over Obama (i.e., -10 on the candidatefavorability scale), and not favorable of one candidate over the other (i.e., 0 on the candidate favorabilityscale). These three theoretical points were chosen specifically as examples of the extremes of candidatefavorability and a midpoint of no preference. Thus, this graph should not be interpreted as three groups ofindividuals, but instead as three lines of the ranges of the rates of learning since the candidate favorabilityscale is a continuum from -10 to 10. Based on the logistic regression models, the baseline respondents couldbe interpreted as those who did not favor one candidate over the other and did not watch any of the debate.

11. It should be noted that based on the predicted values in Figure 4, the rate of learning among thosewho completely favored Obama over Romney (i.e., 10 on the candidate favorability scale) is virtually flat.This is consistent with the notion that those who favored Romney learned at a greater rate about contestedknowledge that was favorable of Romney and that those who favored Obama would learn at a depressed rate,if at all.

FIGURE 3. Predicted Probability of Answering Question About Obama Apology Correctly ByDebate Viewing And Candidate Favorability.

406 | PRESIDENTIAL STUDIES QUARTERLY / September 2014

Page 19: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

the debates. We also found that knowledge based on accurate information that wasuncontested in the debate was gained at a greater rate than knowledge based on infor-mation that was presented by one candidate but contested by the other. Thus, whenaccurate information is presented in a debate, the context of the exchange influences therate of learning.

We did not find differences in the amount of knowledge political sophisticates andnonsophisticates gained from debates. Debates may provide a venue in which learning theissues and matters relevant to a specific presidential campaign occurs regardless of howpolitically sophisticated viewers are. We do realize, though, that there may be limitationin our measure of sophistication.

Consistent with confirmation bias, learning based on information that was con-tested in the debate was influenced by viewers’ candidate preferences more often than not.Yet, learning largely occurred at the same rate for knowledge that was based on accurateinformation that was uncontested in the debate. These findings suggest that in a debateformat, confirmation bias often, but does not always, influence whether viewers will gainaccurate knowledge when information is contested. Additionally, even though therewas uncontested information about both candidates, in most cases if the informationwas presented without rebuttal, it was learned equally among both Romney and Obamasupporters. This result confirms a civic function of debates. Specifically, in some contextsthey can increase viewer knowledge about both candidates regardless of candidatepreference.

Some notable limitations exist in this research. First, the analyses are based onobservational survey data, whose internal validity is less robust than experimentaldesigns. Replicating these results through an experimental design would be the next

FIGURE 4. Predicted Probability of Answering Question About Romney and the Auto IndustryCorrectly By Debate Viewing And Candidate Favorability.

Gottfried, Hardy, Winneg and Jamieson / KNOWLEDGE EFFECTS | 407

Page 20: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

step. Specifically, due to the nature of observational survey data, the reverse relationshipbetween debate viewership and knowledge is possible in that those who are moreknowledgeable about campaign content are more likely to have viewed the debates.Given the specificity of the questions asked about the debate content, the extensive listof controls, and the consistency of the findings with previous research, though, we suspectthat the direction of the relationship of exposure increasing knowledge is more plausible.Next, we recognize the research that suggests that people overreport their media andpresidential debate consumption (e.g., Prior 2009, 2012). However, since we examine theeffect of debate viewership on knowledge, overreported debate viewership would under-state effects because those who falsely reported viewing the debates were analyzed asviewing them, diminishing the actual knowledge effects of those who accurately reportedviewing the debates.

Overall, this study is important for three reasons. First, it confirms that debatesremain important events that foster learning about the candidates and their issue stands.Second, it details the influence of context (specifically whether information is or is notcontested) on learning in absorption of debate content. Finally, it underscores the findingthat confirmation bias is strongly at play among debate viewers but plays a lesser role inthe face of uncontested than contested information.

References

Becker, Lee B., Idowu A. Sobowale, Robin E. Cobbey, and Chaim H. Eyal. 1978. “Debates’ Effects onVoters’ Understanding of Candidates and Issues.” In The Presidential Debates: Media, Electoral, andPolicy Perspectives, eds. G. F. Bishop, R. G. Meadow, and M. Jackson-Beeck. New York: Praeger,126-39.

Benoit, William L., and Glenn J. Hansen. 2004. “Presidential Debate Watching, Issue Knowledge,Character Evaluation, and Vote Choice.” Human Communication Research 30 (1): 121-44.

Benoit, William L., Glenn J. Hansen, and R. M. Verser. 2003. “A Meta-analysis of the Effects ofViewing U.S. Presidential Debates.” Communication Monographs 70 (4): 335-50.

Chaffee, Steven. 1978. “Presidential Debates: Are They Helpful to Voters?” Communication Monographs45 (4): 330-46.

Gans, Herbert J. 2003. Democracy and the News. New York: Oxford University Press.Graf, Joseph, and Sean Aday. 2008. “Selective Attention to Online Political Information.” Journal of

Broadcasting & Electronic Media 52 (1): 86-100.Jackson, Brooks, and Kathleen Hall Jamieson. 2007. Unspun: Finding Facts in a World of Disinformation.

New York: Random House.Jamieson, Kathleen Hall, and Christopher Adasiewicz. 2000. “What Can Voters Learn from Election

Debates?” In Televised Election Debates: International Perspectives, ed. Stephen Coleman. New York:St. Martin’s Press, 25-42.

Just, Marion, Ann Crigler, and Lori Wallach. 1990. “Thirty Seconds or Thirty Minutes: What ViewersLearn from Spot Advertisements and Candidate Debates.” Journal of Communication 40 (3): 120-33.

Kenski, Kate, Bruce W. Hardy, and Kathleen Hall Jamieson. 2010. The Obama Victory: How Media,Money, and Message Shaped the 2008 Election. New York: Oxford University Press.

Lau, Richard R., and David P. Redlawsk. 1997. “Voting Correctly.” American Political Science Review91 (3): 585-98.

Nickerson, Raymond S. 1998. “Confirmation Bias: A Ubiquitous Phenomenon in Many Guises.”Review of General Psychology 2 (2): 175-220.

Prior, Markus. 2009. “Improving Media Effects Research through Better Measurement of NewsExposure.” Journal of Politics 71 (3): 893-908.

408 | PRESIDENTIAL STUDIES QUARTERLY / September 2014

Page 21: All Knowledge Is Not Created Equal: Knowledge Effects and the 2012 Presidential Debates

———. 2012. “Who Watches Presidential Debates? Measurement Problems in Campaign EffectsResearch.” Public Opinion Quarterly 76 (2): 350-63.

Rahn, Wendy M., John H. Aldrich, and Eugene Borgida. 1994. “Individual and Contextual Variationsin Political Candidate Appraisal.” American Political Science Review 88 (1): 193-99.

Stroud, Natlie Jomini. 2008. “Media Use and Political Predispositions: Revisiting the Concept ofSelective Exposure.” Political Behavior 30 (3): 341-66.

———. 2011. Niche News: The Politics of News Choice. New York: Oxford University Press.Waldman, Paul, and Kathleen Hall Jamieson. 2006. “Rhetorical Convergence and Issue Knowledge in

the 2000 Presidential Election.” Presidential Studies Quarterly 33 (1): 145-63.

Gottfried, Hardy, Winneg and Jamieson / KNOWLEDGE EFFECTS | 409