Polit BehavAn experiment <strong>in</strong> which students kept diaries <strong>of</strong> the television programm<strong>in</strong>g theywatched and later answered standard ANES survey questions about their televisionview<strong>in</strong>g habits revealed not only a pervasive tendency <strong>to</strong> overestimate <strong>in</strong> surveyresponses the time spent watch<strong>in</strong>g television but also large discrepancies <strong>in</strong>estimates <strong>of</strong> television watch<strong>in</strong>g among different survey methods. I then showed asimilar pattern <strong>of</strong> differences <strong>in</strong> a large random sample <strong>of</strong> adults, among estimates<strong>of</strong> television view<strong>in</strong>g <strong>of</strong> the k<strong>in</strong>d commonly used <strong>in</strong> survey research <strong>in</strong>corporat<strong>in</strong>gCMAG data (the 1998 ANES pilot survey, which asked about television view<strong>in</strong>ghabits with three different question word<strong>in</strong>gs). These discrepancies are not random;<strong>in</strong>dica<strong>to</strong>rs <strong>of</strong> political sophistication, such as political knowledge, are systematicallyassociated with larger discrepancies. As a result, estimates <strong>of</strong> the relationshipbetween exposure <strong>to</strong> ads, political sophistication, and political behavior are unstableand h<strong>in</strong>ge on the questions used <strong>to</strong> gauge television view<strong>in</strong>g habits.I have also <strong>of</strong>fered two potential, partial solutions. First, by tak<strong>in</strong>g the log <strong>of</strong>estimated exposure (which has the two advantages <strong>of</strong> account<strong>in</strong>g for the decreas<strong>in</strong>gmarg<strong>in</strong>al effects <strong>of</strong> additional exposure <strong>to</strong> advertis<strong>in</strong>g and, by ‘‘compress<strong>in</strong>g’’estimated exposure, reduc<strong>in</strong>g the <strong>in</strong>flated estimates <strong>of</strong> exposure that appear endemic<strong>to</strong> these questions) we can dim<strong>in</strong>ish those differences <strong>in</strong> results that are artifactsprimarily <strong>of</strong> question word<strong>in</strong>g. This approach will not elim<strong>in</strong>ate the problem,however. Even us<strong>in</strong>g logged estimates, I have shown that researchers can drawsharply contrast<strong>in</strong>g normative <strong>in</strong>ferences from how exposure <strong>to</strong> negative advertis<strong>in</strong>g<strong>in</strong>fluences the propensity <strong>to</strong> talk about politics. My analysis could be used <strong>to</strong> suppor<strong>to</strong>ne picture <strong>in</strong> which exposure <strong>to</strong> negative ads makes low political sophisticatesbehave like high political sophisticates but also the opposite view <strong>in</strong> which exposure<strong>to</strong> negative ads exacerbates the differences between low and high politicalsophisticates. In either case, the <strong>in</strong>terpretation is purely an artifact <strong>of</strong> the questionsused <strong>to</strong> gauge television view<strong>in</strong>g habits. A second potential solution uses multiplemeasures <strong>to</strong> gauge ad exposure. Bartels (1996, 2), who is <strong>of</strong>ten cited <strong>in</strong> support <strong>of</strong> the‘‘shows’’ method <strong>of</strong> gaug<strong>in</strong>g view<strong>in</strong>g habits, is similarly circumspect about us<strong>in</strong>g as<strong>in</strong>gle set <strong>of</strong> measures; he suggests that we should weigh the net benefit <strong>of</strong> <strong>in</strong>vest<strong>in</strong>g‘‘entirely <strong>in</strong> specific exposure items’’ aga<strong>in</strong>st the advantages <strong>of</strong> us<strong>in</strong>g ‘‘somecomb<strong>in</strong>ation <strong>of</strong> specific exposure items, general exposure items, and quiz items.’’The implications for research on the impact <strong>of</strong> advertis<strong>in</strong>g are pr<strong>of</strong>ound. Themixture <strong>of</strong> f<strong>in</strong>d<strong>in</strong>gs <strong>in</strong> experiments and surveys may be the result <strong>of</strong> much more thanbasic differences <strong>in</strong> research design. Survey estimates <strong>of</strong> the effects <strong>of</strong> ad exposureare themselves highly unstable. Any attempt <strong>to</strong> estimate exposure <strong>to</strong> televisionshould be wary <strong>of</strong> <strong>in</strong>dividual sensitivity <strong>to</strong> even the most subtle changes <strong>in</strong> questionword<strong>in</strong>g that can have vast effects on <strong>in</strong>ferences. It is no wonder that carefullyconducted studies <strong>of</strong>fer the conflict<strong>in</strong>g <strong>in</strong>terpretations that negative advertis<strong>in</strong>g is aboon or a burden <strong>to</strong> American democracy. Perhaps a comb<strong>in</strong>ation <strong>of</strong> approaches, <strong>in</strong>which we return <strong>to</strong> multiple measures <strong>of</strong> ad exposure <strong>in</strong> order <strong>to</strong> be more certa<strong>in</strong> <strong>of</strong>the stability <strong>of</strong> relationships, while also evaluat<strong>in</strong>g effects based on a s<strong>in</strong>gle commonoperationalization <strong>of</strong> exposure, such as logged estimates, will po<strong>in</strong>t the way forward.Acknowledgementsand suggestions.Thanks <strong>to</strong> Barbara Allen, Andrew Seligsohn, and the edi<strong>to</strong>rs for helpful comments123
Polit BehavAppendixCod<strong>in</strong>g <strong>of</strong> VariablesDaypart Questions. Question Word<strong>in</strong>g: Th<strong>in</strong>k<strong>in</strong>g about this past week, about howmany hours did you personally watch television on a typical weekday morn<strong>in</strong>g/afternoon, from [6:00 <strong>to</strong> 10:00 AM/ 10:00 AM <strong>to</strong> 4:00 PM/4:00 PM <strong>to</strong> 8:00 PM/8:00 PM <strong>to</strong> 11:00 PM/11:00 PM <strong>to</strong> 1:00 AM]. Th<strong>in</strong>k<strong>in</strong>g about this past weekend,about how many hours did you personally watch television from 6:00 AM <strong>to</strong>7:00 PM? Cod<strong>in</strong>g: The <strong>to</strong>tal number <strong>of</strong> weekday hours (multiplied by 5) werecomb<strong>in</strong>ed with the <strong>to</strong>tal number <strong>of</strong> weekend hours <strong>to</strong> estimate the <strong>to</strong>tal number <strong>of</strong>hours <strong>of</strong> TV watched per week.Typical Week Questions (from ANES 1998 Pilot). Question Word<strong>in</strong>g: On atypical weekday, about how many hours <strong>of</strong> television do you watch dur<strong>in</strong>g themorn<strong>in</strong>g and afternoon? About how many hours <strong>of</strong> television do you watch on atypical weekday even<strong>in</strong>g? On a typical weekend day, about how many hours <strong>of</strong>television do you watch dur<strong>in</strong>g the morn<strong>in</strong>g and afternoon? Cod<strong>in</strong>g: The <strong>to</strong>talnumber <strong>of</strong> weekday hours (multiplied by 5) were comb<strong>in</strong>ed with the <strong>to</strong>tal number <strong>of</strong>weekend day hours (multiplied by 2).Show Questions (ANES 1998 Pilot). Question Word<strong>in</strong>g: How many days/times<strong>in</strong> the past week have you watched [The Today Show/The Rosie O’Donnell Show/daytime soap operas like General Hospital or Days <strong>of</strong> Our Lives/Jeopardy or Wheel<strong>of</strong> Fortune/a sports event/local news]? Cod<strong>in</strong>g: The sum <strong>of</strong> all six genres (eachgenre was rescaled from zero <strong>to</strong> one) divided by six.Show Questions (Experiment). Question Word<strong>in</strong>g: How many times <strong>in</strong> a typicalweek do you watch [Jeopardy/Wheel <strong>of</strong> Fortune/morn<strong>in</strong>g news programs such asToday, Good Morn<strong>in</strong>g America, or The Early Show/daytime television shows suchas Oprah W<strong>in</strong>frey or Jerry Spr<strong>in</strong>ger/national network on news/local TV news shows,either <strong>in</strong> the late afternoon or early even<strong>in</strong>g]?.Efficacy. Question Word<strong>in</strong>g: Please tell me how much you agree or disagree withthese statements ... agree strongly, agree somewhat, neither agree nor disagree,disagree somewhat, disagree strongly, don’t know? Public <strong>of</strong>ficials don’t care whatpeople like me th<strong>in</strong>k; Sometimes politics seems so complicated that a person likeme can’t really understand what’s go<strong>in</strong>g on; People like me don’t have any sayabout what the government does. Cod<strong>in</strong>g: The average response on the 1 <strong>to</strong> 5 scale.Number <strong>of</strong> Days <strong>in</strong> the Past Week Talked About Politics. Question Word<strong>in</strong>g:How many days <strong>in</strong> the past week did you talk about politics with family or friends?Number <strong>of</strong> Issues Recognize that Candidates Have Talked About. QuestionWord<strong>in</strong>g: For each issue we would like <strong>to</strong> know if you th<strong>in</strong>k either one <strong>of</strong> thecandidates, both, or neither is talk<strong>in</strong>g about these issues (private school vouchers,abortion, gun-related crimes, campaign contributions from PACs, protect<strong>in</strong>g thequality <strong>of</strong> the air and water, improv<strong>in</strong>g discipl<strong>in</strong>e <strong>in</strong> schools). Cod<strong>in</strong>g: Total <strong>of</strong>number <strong>of</strong> issues each candidate is talk<strong>in</strong>g about.Intention <strong>to</strong> Vote. Question Word<strong>in</strong>g: (Half sample 1) So far as you know, doyou expect <strong>to</strong> vote <strong>in</strong> the elections this com<strong>in</strong>g November? Would you say that youare def<strong>in</strong>itely go<strong>in</strong>g <strong>to</strong> vote, probably go<strong>in</strong>g <strong>to</strong> vote, or are you just lean<strong>in</strong>g <strong>to</strong>wards123