Journal of Politics 2006 Althaus

8/3/2019 Journal of Politics 2006 Althaus

1/18

Priming Effects in Complex InformationEnvironments: Reassessing the Impact of NewsDiscourse on Presidential Approval

Scott L. Althaus University of Illinois at Urbana-ChampaignYoung Mie Kim The Ohio State University

This paper revisits the original psychological literature on priming in order to assess new possibilities for research onpriming effects stimulated by news discourse. We detail some important theoretical limitations of existing primingresearch; propose a method for studying the dynamics of priming effects in real-world, complex informationenvironments; and illustrate the usefulness of this approach with a case study of opinion change during the 199091

Persian Gulf Crisis. This case study documents for the first time the daily dynamics of priming effects in a complexinformation environment and confirms that priming effects are not merely a function of changes in the volume ofnews coverage about a given topic. Our findings suggest that news priming effects can be produced by changes in theapplicability of relevant knowledge constructs, rather than merely by their temporary accessibility in long-termmemory, and by cumulative as well as recent exposure to news coverage.

Although the mass media were once widelythought to have little impact on peoples politi-cal opinions (Klapper 1960), research on

priming effects over the past two decades has demon-strated that mass media can influence opinions in pro-

found ways (e.g., Iyengar and Kinder 1987; Krosnickand Brannon 1993; Krosnick and Kinder 1990; Millerand Krosnick 1996, 2000). Priming research has shownthat although media coverage by itself is unlikely tochange the political attitudes of many people, mediacoverage can influence which attitudes people use tomake political judgments. Because priming can alterthe criteria citizens use to evaluate political issues,leaders, and events, priming effects can have importantconsequences for the outcomes of elections, the emer-gence of public support for policy initiatives, and theapproval ratings of political leaders.1

Although it appears increasingly likely thatpriming is responsible for a wide range of mediaeffects, priming research conducted by political com-munication scholars has enjoyed only limited successin drawing connections between observed effects and

underlying causes. While the effects of news priminghave been most clearly identified in laboratory experi-ments (following the path breaking work on agenda-setting effects by Iyengar and Kinder 1987; Iyengaret al. 1984; Iyengar, Peters, and Kinder 1982), only afew studies have examined such effects outside thelaboratory setting (Iyengar and Simon 1994; Krosnickand Brannon 1993; Krosnick and Kinder 1990; Panand Kosicki 1997). We argue below that the findings ofall of these nonexperimental studies share commonfeatures that obscure both the underlying dynamics ofnews priming and the causal factors that shape

1For example, campaigners in the recent 2004 presidential election seemed less interested in persuading voters so much as priming themby influencing the criteria on which voters would choose between candidates. National surveys conducted during the campaign showedthat majorities favored Senator Kerrys positions on health care and the economy, while majorities also favored President Bushs leadershipstyle and his handling of the War on Terrorism. The Kerry campaign therefore attempted to focus public attention on the lack of jobgrowth over the presidents first term while the Bush campaign reminded voters of the 9/11 attacks and the need for strong leadership inthe face of continued threats. Each campaign attempted to produce priming effects by increasing the salience of attitudes that favored theirrespective candidates rather than pursue the more difficult task of trying to talk voters out of their established opinions.

The Journal of Politics, Vol. 68, No. 4, November 2006, pp. 960976

2006 Southern Political Science Association ISSN 0022-3816

960


2/18

priming effects in realistically complex informationenvironments.

This article attempts to expand our knowledge ofnews priming effects by addressing these limitations ofprevious work. We revisit the original psychologicalliterature on priming to assess new possibilities for

priming effects research conducted by political scien-tists. We then propose a method for studying thedynamics of priming effects in real-world, complexinformation environments and illustrate the useful-ness of this approach with a case study of opinionchange during the 199091 Persian Gulf Crisis. Thiscase study documents for the first time the dailydynamics of priming effects in a complex informationenvironment and confirms that priming effects are notmerely a function of changes in the volume of newscoverage about a given topic. Priming effects on presi-dential job approval were influenced not only by

recent exposure to news about the Persian Gulf Crisisbut also by cumulative exposure to relevant news dis-course and changes in the evaluative tone of that dis-course. Moreover, our findings suggest that newspriming effects can be produced by changes in theapplicability of relevant knowledge constructs ratherthan merely in their temporary accessibility in long-term memory. By expanding the study of newspriming to include applicability as well as accessibilityeffects, we illuminate the dynamic, short-term evolu-tion of news priming effects that has gone unobservedin prior research.

What We Still Dont Know aboutNews Priming

Over the past 25 years the priming literature in socialpsychology has undergone a theoretical renaissance,but during this same period the basic theoretical ori-entations and measurement strategies used in thepolitical science literature on priming have remainedstable. Empirical research on news priming has tendedto focus on replicating the findings of two founda-

tional studies (Iyengar and Kinder 1987; Krosnick andKinder 1990), with a recent emphasis on identifyingmoderating variables that condition priming effects.As a consequence, the insights generated by recentsocial psychological work on priming have yet to befully incorporated into research on news primingeffects conducted by political communication scholars(Price and Tewksbury 1997). Models of news primingeffects therefore remain relatively underspecified com-pared to priming research in such topical domainsas aggression and racial stereotyping (e.g., Anderson

1997; Power and Murphy 1996). Some outside observ-ers even suggest that political communication researchuses the term priming more as a metaphor than afully specified concept (Roskos-Ewoldesen, Roskos-Ewoldesen, and Dillman Carpentier 2002).

As a consequence of its limited engagement with

developments in social psychology, the news primingliterature has had difficulty providing satisfactoryanswers to long-standing questions about the natureof priming effects. First, the political science literaturehas struggled to explain what causes priming effectsstimulated by news coverage, in part because the stan-dard research designs used in this literature obscurethe relative contributions of two causal mechanismsknown to produce priming effects. Second, the newspriming literature tends to measure priming effects atonly one or two points in time. As a consequence, ithas been unable to adjudicate the sharp disagreement

between experimental and survey-based primingstudies regarding the duration of priming effects.Laboratory experiments conclude that priming effectstend to be short-lived, largely disappearing withinminutes or hours of exposure to the priming stimulus,while survey-based studies using secondary analysis ofpanel data conclude that priming effects are highlypersistent, often lingering months or even years afterstimulus exposure. No convincing attempt has yetbeen made to reconcile these conflicting but consistentfindings produced by different strands of the researchliterature. Third, social psychological research has

documented a variety of stimulus attributes known toinfluence priming effects, but previous news primingstudies have reduced the complexity of news discourseinto a simple measure of topical prominence withinthe agenda of news coverage (e.g., Pan and Kosicki1997; Zaller 1992). Because such big message indi-cators do not control for other features of newscontent likely to stimulate priming effects, the politicalcommunication literature may be drawing misleadinginferences about the reasons why news coverage pro-duces priming effects.

How Are Priming Effects Produced?

Priming is the activation of knowledge stored in long-term memory following exposure to a stimulus.Priming can produce a variety of judgmental effects,but research conducted by political communicationscholars has focused almost exclusively on what areknown as accessibility effects. Accessibility can bedefined as the potential that knowledge stored inmemory will be activated for potential use in a judg-mental task (Higgins 1996, 134). The temporary ease

priming effects in complex information environments 961


3/18

with which a construct may be called up into workingmemory can increase the likelihood that the activatedconstruct will be used as the basis for making subse-quent evaluations of various kinds.

The social psychological literature emphasizesthat accessibility alone does not produce knowledge

activation and that increasing the accessibility of aconstruct does not ensure that it will be used as acriterion in subsequent judgments. Rather, accessibil-ity is one of two primary factors moderating the acti-vation of stored knowledge: the other is the degreeto which a stimulus and a stored knowledge constructare perceived as applicable to one another. Even if aknowledge construct has a small chance of being acti-vated due to its diminished accessibility, its chances ofbeing activated are increased by its perceived applica-bility or relevance to a judgmental task (Higgins1996).

Perceptions of relevance are themselves con-structed through priming, for by increasing the acces-sibility of a construct in the context of a judgmenttask, priming also increases the odds that the constructwill be used again when a person is faced with a similar

judgment task. The priming of a construct initiallyproduces accessibility effects, but subsequent primescan produce applicability effects, since each primingepisode generates associations between the primedconstruct and other constructs, stimuli, or judgmentalcontexts (Higgins 1996; Higgins and King 1981).Applicability effects can also be influenced by the

framing of the task (e.g., Trope and Ginossar 1988),the degree of attention to stimuli (e.g., Kim 2005;Martin and Achee 1992), as well as by the motivations(Kruglanski 1989; Thompson et al. 1994), predisposi-tions (Fein and Spencer 1997), prior value commit-ments (Domke, Shah, and Wackman 1998) and socialnorms (Mendelberg 2001; Valentino, Hutchings,and White 2002) used by individuals to processinformation.

Seen in this way, the social psychological literatureenvisions priming as something like a two-stageprocess: the priming stimuli should influence the

accessibility of some knowledge constructs more thanothers, but whether people use those primed con-structs as evaluative criteria depends on the degree towhich they are perceived as applicable to the judgmen-tal task. Against this theoretical background, thetendency for news priming research to focus on acces-sibility effects alone is a notable limitation.

News priming research has begun to explore someimplications of associative network theory (Lodge andTaber 2000; Miller and Krosnick 1996; Valentino1999), but the dynamic construction of applicability

effects has never been directly observed. What little weknow of applicability effects in the news priming lit-erature comes from tests of the gradient hypothesis(Iyengar and Kinder 1987; see also Miller and Kros-nick 1996), which confirmed that news coverage logi-cally irrelevant to target attitudes does not tend to

prime those attitudes. But if accessible constructs gen-erate priming effects only when those constructs arerelevant to the task, then the standard finding thatpriming is a function of short-term changes in con-struct accessibility must be only part of the story. Incontrast, the social psychology literature shows howcumulative exposure can generate strong accessibilityand applicability effects even in the absence of recentprimes (Higgins 1996; Lau 1989). Media primingstudies in other domains have likewise demonstratedapplicability effects generated by frequent priming(e.g., Shrum and OGuinn 1993). Since applicability

effects are constructed over time as a consequence ofaccessibility effects, the tendency for news primingresearch to focus narrowly on short-term changesin construct accessibility stimulated by recent newsexposure not only captures just part of the action, butmay have led previous research to misspecify thecausal mechanisms underlying news priming effects.

How Long Do Priming Effects Last?

Our uncertainty regarding the relative influence ofaccessibility and applicability for explaining priming

effects is heightened by the type of research designstypically employed to study news priming. Experi-ments tend to study the immediate effects of recentstimulus primes, often within minutes of stimulusexposure. Those using response latency measuresassess the effects of priming within milliseconds afterexposure. Experimental designs tend to find primingeffects that are short-lived, disappearing soon afterexposure to the stimulus prime. At the other extremeare survey-based designs, which test for effects inwhich the stimulus primes occur months or even yearsbefore the target attitudes are measured. In contrast to

the findings of experimental work, these survey-basedstudies usually conclude that priming effects aredurable and can be surprisingly long-lived.

No attempt has yet been made to reconcile thesestarkly divergent claims regarding the duration ofpriming effects. Doing so requires an analysis of theevolving dynamics of priming. The social psychologyliterature posits that priming effects are produced intwo ways: recent primes can temporarily increase theaccessibility of a construct, and frequent primes cangradually increase the applicability of a construct in

962 scott l. althaus and young mie kim


4/18

ways that make the construct chronically accessiblelong after exposure to the stimulus (e.g., Price andTewksbury 1997). Only the first mechanism has beendirectly tested in news priming studies, and thesestudies have tended to capture the effects of primingat only one point in time: nearly all of the published

studies on news priming have employed a standardpre/post treatment design that captures the impact ofa single stimulus treatment. Since the target attitudesfor stimulus primes in these studies were all real-worldattitudes about political issues that had been aroundfor a long time, the designs used by these studies areunable to distinguish whether observed effects areproduced by the recency or cumulative frequency ofprimes. Priming effects produced by exposing subjectsto stimuli about a familiar issue could be a function ofthe recent exposure itself. They could also be a func-tion of frequent preexperimental exposure to relevant

stimuli that has generated applicability effects whichthemselves trigger the priming response observed inthe laboratory. If frequent exposure before the experi-ment is really driving priming effects that seem toresult from recent exposure in the laboratory, then ourinability to control for the frequency of preexperimen-tal priming will lead us to overstate the apparentimpact of the experimental manipulation.

Differentiating between recency and frequency asthe mechanisms underlying observed effects wouldrequire the experimental manipulation of a com-pletely unfamiliar attitude construct over a long

period of time, where close controls over cumulativeexposure could clarify the relative contribution ofrecent and frequent exposure to priming stimuli. Ourpaper is the first to do this, using a quasi-experimentalanalysis of the evolving dynamics of priming effects inpresidential approval caused by the sudden onset of anunexpected and unfamiliar foreign policy crisis.

Is Priming Merely A Big Message Effect?

Political communication researchers have long pre-sumed that priming effects are produced by the rela-

tive salience of a topic in news coverage. Primingstudies conducted by political scientists have assumedthat news exposure inclines audiences to use the domi-nant agenda carried in the news as an evaluativecriterion. Such claims are justified in the case ofexperimental studies with strict controls to rule outother possible influences, but cannot be presumed instudies using naturally occurring news exposure. Sur-prisingly, no studies of priming conducted outside thelaboratory have examined directly the content andtone of media messages about a given topic as factors

that might moderate priming effects. The way moststudies of real-world priming operationalize mediacontent is merely by identifying the total amount ofnews coverage given to a topic without regard tocontent variation within that topical coverage (e.g.,Krosnick and Brannon 1993; Krosnick and Kinder

1990; Pan and Kosicki 1997). In research on primingeffects stimulated by the 1991 Gulf War, for instance,the rich complexity and dynamic changes in mediacontent about the Iraqi invasion of Kuwait and itsaftermath were reduced to a single, one-dimensionalcatchall categorycoverage about the warwhichwas then operationalized as simple counts of stories orlines of coverage about the war appearing in newspa-pers or on television.

This notion of a big message effect (Pan andKosicki 1997; Zaller 1992) imposes important restric-tions on priming research. While the big message

approach can capture the effect of a dominant, one-sided flow of information (i.e., Zallers mainstreameffect), it is of questionable value in explainingpriming effects stimulated by complex informationenvironments, where conflicting ideas compete withone another in discourse. The rare studies to analyzemedia effects in complex information environments(Druckman 2004; Druckman and Nelson 2003; Sni-derman and Theriault 2004) have concluded that sucheffects tend to disappear once strict controls on expo-sure to one-sided information flows are relaxed. Asidefrom raising the possibility that news priming effects

may be a phenomenon of the laboratory rather than ofthe everyday world, the results of these studies suggestthat the presumption of big message effects needs tobe reconsidered.

Recent developments in the social psychologicalliterature suggest that the big message approachcould lead researchers to miss important effects ofnews priming, as well as to misspecify the mechanismsby which such effects are produced. For instance, thebig message approach is at odds with the possibilityof applicability effects brought about by cumulativeexposure, since it presumes that priming is influenced

only by the temporary dominance of a particular topicin the overall news agenda. Moreover, research onapplicability effects tends to conclude that constructsin memory with a similar evaluative toneas good isto nice, or happy is to joyare more likely to activateone another than constructs with a dissimilar evalua-tive tone, as good is to evil or happy is to sad. Researchby social psychologists has confirmed that priming ismore likely to occur when the evaluative tone betweenstimulus primes and target evaluations are consistentwith one another (e.g., Anderson 1983). Such findings



5/18

suggest that news priming effects might vary as a func-tion of the evaluative tone of news discourse, such aswhen news about a given subject is lopsidedly criticalof the president. Recent theoretical studies have sug-gested that the evaluative tone of media contentshould have a bearing on the ways in which that

content primes subjects (Lodge and Taber 2000; Priceand Tewksbury 1997), but such relationships neverhave been tested in news priming research. This omis-sion is especially glaring given the long line of persua-sion research (e.g., Fan 1988; Zaller 1992) that findsevaluative tone to be a major influence on other typesof media effects.

A New Approach

As a consequence of these three limitations in previousresearch, we know surprisingly little about the nature

of news priming in realistically complex and dynamicinformation environments. Because the standardmethodological approaches used in this researchobscure the temporal dynamics of priming effectsdynamics that have important theoretical implicationsfor clarifying how priming effects workthe literaturehas been unable to settle the conflicting claims madeby experimental and survey-based priming studiesand is likely to have overstated the apparent impact ofrecent exposure to relevant information.

To address these limitations, we adopt a quasi-experimental approach that allows us to track daily

changes in priming effects on presidential jobapproval stimulated by an unanticipated but highlyvisible foreign policy crisis. On August 2, 1990, Iraqiforces launched a surprise invasion of oil-rich Kuwait,an event that precipitated an immediate series of high-profile responses by President George H. W. Bush thatled to the Persian Gulf War of 1991. By August 9, justone week after the invasion, the Gallup Organizationwas in the field with the first of 20 national surveyseventually covering the crisis period that askedrespondents about their exposure to news about thecrisis, their overall approval of the president, and their

support for the presidents handling of the crisis. TheGallup data allow us to estimate crisis-inducedpriming effects in presidential approval for 74 of the206 days of the crisis period.2 Coupled with a detailedcontent analysis of nightly news coverage given to thecrisis, these data allow us to track the unique influenceof recent and cumulative exposure to news about the

Persian Gulf on daily changes in the relative weight offoreign policy attitudes in the presidents overall jobapproval rating.

Data and MethodsNews Content Data

Measures of news coverage during the Gulf Crisiscome from content analysis data collected for everynightly news broadcast on each of the three majornetworks that appeared during the period of interest.Compiled from full-text transcripts, the data setincludes every news story relevant to the crisis withIraq that appeared on ABCs World News Tonight,CBSs Evening News, and NBCs Nightly News

(N=

3,854 stories).An unusually detailed coding protocol for thesedata was designed to capture not only the substance ofthe policy debate among government officials, jour-nalists, and other sources appearing on the nightlynews, but also the supporting arguments and framesof reference used by those sources to lend credibilityto their policy positions. The coding scheme forthis study follows prior studies (e.g., Gamson andModigliani 1989; Pan and Kosicki 1993) that identifydiscrete thematic elements in discourse correspondingto available policy opinions, the rationales for imple-

menting policies or the goals to be achieved by thosepolicies (Mermin 1996, 1999; Nelson 1999), andcontext discourse suggesting core problems or framesin which the policies might be located (Dorman andLivingston 1994; Kirton 1993; Lang and Lang 1994).Despite the complexity of this coding scheme (foradditional details and a complete listing of themes, seeAlthaus 2003), the resulting measures of news contentare highly reliable. Intercoder reliability tests were per-formed on a sample of 101 randomly selected ABCWorld News Tonight stories for all discourse themestaken together, which produced intercoder agreement

on the presence of specific themes in 88% of cases(Cohens kappa = .875; Brennan and Predigerskappa = .877).

The news content data used in the analysis wereaggregated to produce daily measures of news cover-age relevant to the Gulf Crisis. The measures usedbelow include the average daily number of GulfCrisis stories appearing on a network newscast, theaverage daily amount of policy discourse about theGulf Crisis contained in a nightly newscast, and theaverage daily evaluative tone of news discourse about

2As we define it, the Persian Gulf Crisis extended from the Iraqiinvasion until the start of the U.S.-led ground war that drove Iraqiforces out of Kuwait.



6/18

the Gulf Crisis relative to the Bush administrationsstated positions.3

Opinion Data

Opinion data for this study come from national

surveys administered by the Gallup Organizationbetween August 9, 1990 and February 10, 1991. Twentysurveys containing relevant variables were in the fieldbetween the Iraqi invasion of Kuwait and the start ofthe ground war in late February, administered over atotal of 74 days. Since our analysis is concerned withdaily changes in attitude structures corresponding toshifts in news content, we divide respondents in thesemultiday surveys by date of interview. It is importantto point out that the daily survey data analyzed beloware therefore not daily random samples, but rathercome from random samples in which interviews were

conducted over a period of three or four days. Strictlyspeaking, data for each of these days are not represen-tative of the American population as a whole. To helpcompensate for this limitation, all analyses reportedbelow include controls for party identification(dummy variables for Republican and Democraticidentifiers, leaving independents as the reference cat-egory), gender (males = 1, females = 0), race (non-whites = 1, whites = 0), and age (in years).

Several questions asked in identical or nearly iden-tical form in each of these surveys allow us to track thechanging attitudinal structure of President Bushs job

approval ratings. Our measure of job approval comesfrom the standard question, Do you approve or dis-approve of the way George Bush is handling his job aspresident? Approving respondents were assigned avalue of one, and all others were assigned a value ofzero. Three independent variables are used to predict

job approval:

Support for troop deployment, from the question Inview of the developments since we first sent ourtroops to Saudi Arabia, do you think the UnitedStates made a mistake in sending troops to Saudi

Arabia or not? Respondents saying the troopdeployment was not a mistake were assigned a valueof one, while all others were assigned a value of zero.

Support for offensive force, from responses to twosimilar questions asking respondents whether theyapproved of the use of military force to drive Iraqi

forces out of Kuwait.4 Respondents favoring offen-sive military action were assigned a value of one,while all others were assigned a value of zero.

Gulf Crisis approval, from the question Do youapprove or disapprove of the way George Bush ishandling this current situation in the Middle East

involving Iraq and Kuwait?

5

Approving respon-dents were assigned a value of one, and all otherswere assigned a value of zero.

Exposure to news coverage is an additional controlvariable commonly used in priming studies. Thesestudies typically treat news exposure as a moderatingfactor, since news exposure in one form or anothermust be a necessary condition for news to influenceattitude structures. Our analysis shares this standardexpectation, but the nature of our case requires us tooperationalize news exposure somewhat differentlythan other studies because public attention to news

about the Gulf Crisis was unusually widespreadthroughout the time period that these surveys werebeing administered. All of the Gallup surveys used inthe analysis include a measure of self-reported newsexposure that asks, How closely have you followednews about the situation involving the invasion ofKuwait by Iraq and the sending of U.S. troops to SaudiArabia? Would yousay youhave followedit veryclosely,fairly closely, not too closely, or not at all closely?Averaging across the 74 days of Gallup data, 89% ofrespondents reported paying either very close orfairly close attention to news coverage of the GulfCrisis(dailymax = 98%, daily min = 74%).The level ofpopular engagement with the Gulf Crisis was so highthat the daily numbers of inattentive respondents aretoo few to analyze with any confidence. Our solution isto analyze only those respondents who reported payingvery or fairly close attention to news about the crisis,omitting all others from the analyses that follow. Thisleaves us with a mean of 248 attentive respondents perday (max = 700, min = 43).6

3These measures all take the form of averages across the threenightly newscasts because occasional missing transcripts for oneor two networks makes it inappropriate to simply sum the totalnumber of news elements across the broadcasts.

4Before the start of the air war on January 16, the question read Ifthe current situation in the Middle East involving Iraq and Kuwaitdoes not change by January (15, 1991), would you favor or opposethe United States going to war with Iraq in order to drive the Iraqisout of Kuwait? The question asked in the surveys conducted afterthe start of the air war read Do you approve or disapprove of theUnited States decision to go to war with Iraq in order to drive theIraqis out of Kuwait?

5In two of the 20 surveys, this question was phrased slightly dif-ferently as Do you approve or disapprove of the way George Bushis handling this current situation in the Persian Gulf region?

6Only 21 days contained fewer than 100 respondents after filteringaway inattentive citizens. See the tables in the web appendix(http://www.journalofpolitics.org) for daily counts of valid casesused in the first-stage regressions.

http://www.journalofpolitics.org/http://www.journalofpolitics.org/


7/18

During this period, the most common source ofnews was the nightly network newscasts. Nielsen view-ership ratings for this period reveal a combined directaudience for the three evening news broadcasts ofbetween a quarter and a third of American house-holds, split fairly evenly among the three networks

(Althaus 2002). This direct audience for the networkbroadcasts was substantially larger than for any othernational news source. Moreover, the content of thesebroadcasts should accurately reflect the daily agendaand tone of other national news sources during thisperiod (e.g., Danielian and Reese 1989; Entman 2003;Gans 1979; McCombs and Shaw 1972; Sparrow 1999).Thus, we use network news broadcasts as a proxy forthe daily news coverage about the 199091 Gulf Crisisthat may have influenced Americans paying attentionto news about the crisis.

Our analytical strategy for estimating daily

priming effects brought about by news coverage pro-ceeds in two stages. In the first stage, we estimate dailychanges in the size of individual-level priming effects.Our basic model takes the following form:

prob Y S C D

S D e

i i k ik l d

l i d i

=( ) = + + + +

( )+

1 1

where Yi is respondent is presidential approval rating,Si is respondent is support for the presidents GulfCrisis policies, Cik is respondent is score on the kth

control variable, Dd is a dummy variable taking a valueof 1 on date d (0 otherwise), Si Dd is the product ofrespondent is support for the presidents Gulf Crisispolicies multiplied by the dummy variable for therespondents date of interview d, and ei is the errorterm for the ith observation. In this equation, b1 is thecoefficient for the support variable, bk is the coefficientfor the kth control variable, bl is the coefficient for thelth date, and dl is the coefficient for the lth interactionterm. This model was used on a pooled data set con-sisting of all attentive Gallup respondents in the 20national surveys conducted during the period of inter-

est. The design of this model estimates a main effectfor each of the control variables over the entire timeperiod, but allows the effects of the models Gulf Crisissupport variable to vary by date of interview. Since the

job approval variable used in this analysis is dichoto-mous, logistic regression was used to estimate thismodel.

In the second stage of the analysis, we calculate thedaily priming effect of support for Gulf Crisis policieson presidential approval as the change in the probabil-ity Yi = 1 that comes from shifting the value of the

support variable from opposing to favoring the presi-dents policies. We then use these daily estimates ofpriming effects in an OLS model that regresses the sizeof the priming effect among respondents interviewedin day d on the amount of particular types of newscontent appearing in evening newscasts on day d.7 If

particular types of news content are influencing thesize of a priming effect during the period of interest,this OLS model should return statistically significantcoefficients for those news content variables.

This approach is obviously cruder than experi-mental designs for assessing priming effects, since dif-ferent types of individuals surely respond in differentways to the same news content. But our approach issuperior in several respects to the standard survey-based strategies for assessing priming effects. Multiplemeasures of key independent variables allow us toshed additional light on the question of whether

priming is mainly an accessibility or applicabilityeffect, since accessibility should tend to prime relatedattitudes in similar ways. If news coverage of the GulfCrisis produces similar priming effects across ourmultiple measures of policy support, this would bestrong evidence that accessibility rather than applica-bility was driving observed effects. We also estimatepriming effects in time periods that coincide withdaily news cycles, which is the naturally occurringduration of headline news in the United States. Noprevious survey-based study has studied primingeffects at such a fine level of detail regarding the time

sequencing of stimulus and attitudinal measures.

Findings

Figure 1 displays opinion trends from Gallup pollsthat document a gradual decline in public support forthe president following a rally that came immediatelyafter the Iraqi invasion of Kuwait. Four months ofdeclining support for the president were then followedby almost immediate surges in support after the startof the bombing campaign. While some accounts usingmonthly summaries of aggregate opinion (e.g.,

7We experimented with a number of alternative lag specificationsbefore concluding that news data from the date of interviewprovided the best fit. Most interviews appeared to have beenconducted after the days nightly network newscasts, further sup-porting our decision to model priming effects as a function ofnews content that was broadcast on the date of interview. In addi-tion, the high degree of similarity between nightly news contentand the agenda of other national news sources appearing on thesame day provides a strong theoretical rationale for structuring themodel in this way.



8/18

Krosnick and Brannon 1993) incorrectly attributePresident Bushs widely noted jump in popularity tothe successful prosecution of the ground war in lateFebruary, we see in this figure that the largest jump inBushs approval rating is an immediate and direct con-sequence of the start of the air war in mid-January. Apopulation that had been deeply divided about thewisdom of going to war became, immediately follow-ing its onset, a public galvanized for action.

Although the opinion trends in Figure 1 followrelatively stable declines until they are simultaneously

raised and reset by the start of the air war, the attitudestructures underlying the presidents approval ratingsare quite volatile during this time period. Figure 2shows daily changes in the size of priming effectsinfluencing the presidents job approval rating. (Coef-ficients from the complete regression models are pro-vided in a separate appendix available from this

journals web site.) The size of each priming effect wascalculated as the absolute change in the predictedprobability of approving the presidents handling ofhis job when the value of the priming variable is

FIGURE 1 Support for the President and the Use of Force during the Persian Gulf Crisis

0

10

20

30

40

50

60

70

80

90

100

5-Jul

19-Jul

2-Aug

16-Aug

30-Aug

13-Sep

27-Sep

11-Oct

25-Oct

8-Nov

22-Nov

6-Dec

20-Dec

3-Jan

17-Jan

31-Jan

14-Feb

28-Feb

14-Mar

28-Mar

%Support

Job ApprovalSupport for Troop DeploymentSupport for Offensive ForceGulf Crisis Approval

Iraqi Invasion

of Kuwait

Start

of Air

War

Start of

Ground

War

FIGURE 2 Daily Priming Effects on Job Approval During the Persian Gulf Crisis

.00

.20

.40

.60

.80

1.00

5-Jul

19-Jul

2-Aug

16-Aug

30-Aug

13-Sep

27-Sep

11-Oct

25-Oct

8-Nov

22-Nov

6-Dec

20-Dec

3-Jan

17-Jan

31-Jan

14-Feb

28-Feb

14-Mar

28-Mar

SizeofPrimingEffect

Effect of Gulf Crisis Approval

Effect of Support for Troop Deployment

Effect of Support for Offensive Force

Iraqi Invasion

of Kuwait

Start of

Air War

Start of

Ground War

Note: Points within each trend indicate the size of daily priming effects, defined as the change in the probability of approving thepresidents job when the support variable is changed from oppose to support, holding all other variables constant. Trend lines representa moving average of the current days effect, the effect from the previous point in the trend line, and the effect from the next point in thetrend line.



9/18

changed from opposition to support while holding thevalue of all other variables constant. The larger thechange in the predicted probability of approvingthe presidents handling of his job, the greater theweight that the priming variable has in the attitudestructure for job approval.8 For instance, black tri-

angles in Figure 2 represent changes in the probabilitythat a person would approve of how the president ishandling his job when Gulf Crisis approval changesfrom disapprove to approve. On August 9, the first dayof the series, switching from disapproving to approv-ing President Bushs handling of the Gulf Crisisincreases the probability of approving his overall jobas president by .45. Recalling that probabilities rangefrom 0 to 1, this is a substantial impact.

To clarify the trends in these daily priming effects,Figure 2 includes trend lines showing the size of onedays priming effect averaged with the size of the effect

immediately preceding and following it. The largestpriming effects tended to come from approval of howthe president was handling the Gulf Crisis, followed inrelative magnitude by support for the deployment ofAmerican military forces to Saudi Arabia, and supportfor using those military forces to drive Iraqi forces outof Kuwait.9 An exception to this general pattern camearound the start of the air war, when support foroffensive force briefly became a more important com-ponent of President Bushs job approval than supportfor troop deployment.

Figure 2 sheds light on two lingering questionsin the priming literature. First, on the question ofwhether priming effects are of short or long duration,Figure 2 shows that the magnitude of daily primingeffects is highly variable within each of the threetrends. This pattern is consistent with the typical find-

ings from experimental studies, which suggest thatrecent exposure to relevant stimuli tends to produceimmediate priming effects of relatively short duration.At first glance, this pattern is more difficult to squarewith the findings of previous survey-based primingstudies that report lingering effects lasting severalmonths after exposure to the news coverage of inter-est. If news priming effects were so durable, we mightexpect to see far less day-to-day variation in effectsizes. Some of this day-to-day variance is surelyrandom, a consequence in part of the sometimessmallish numbers of respondents surveyed on a par-

ticular day. But even though the size of priming effectscan change dramatically from day to day, these effectsare not merelyrandom: there are clear central tenden-cies in the effect trends that evolve in meaningful waysover time. The general pattern is for priming effectsto increase in size over time for all three variables,although each trend also seems to be influenced byidiosyncratic factors.

These trends shed light on the question of whetherpriming effects derive more from increased accessibil-ity or increased applicability. If accessibility alone weredriving changes in the size of these priming effects, we

would expect news coverage about the Gulf Crisis toinfluence these three closely related support variablesin similar ways: the size of the priming effects shouldbe fairly similar for each of the three trends. Further-more, any movement in the size of priming effectsshould occur in parallel if such movement were beingproduced merely by differences in the accessibility ofthese constructs, presuming for the momentas pre-vious work has donethat accessibility is a functionof the overall amount of recent news coverage aboutthe Gulf Crisis. Yet the size of effects differs across thethree variablesgenerally largest with Gulf Crisis

approval and smallest with support for offensiveforcedespite the fact that the accessibility of all ofthem should be stimulated in similar ways by GulfCrisis news coverage. Moreover, the three trends donot always move in tandem: the correlation in the sizeof priming effects between Gulf Crisis approval andsupport for offensive force is a healthy .60 (p .01,n = 27), while that between support for offensive forceand support for troop deployment is a more modest.43 (p .05, n = 23), and that between Gulf crisisapproval and support for troop deployment is just .25

8Because of the complexity of the regression model used to esti-mate daily priming effects, and because the key independent vari-ables are not always available in the same surveys, separateregression models are run for each of the three independent vari-ables. This means that the effects shown in Figure 2 do not controlfor the simultaneous impact of all three variables. To test whetherincluding all three variables simultaneously might change the pat-terns shown in Figure 2, the authors conducted separate regressionanalyses for each survey in which more than one of these variableswere available. A total of 18 surveys contained the Gulf Crisisapproval variable along with the support for troop deploymentvariable, and six surveys contained both of those plus the supportfor offensive force variable. The priming effect trends produced bythis analysis followed the same patterns as Figure 2, except that thesize of priming effects was relatively smaller for the troop deploy-

ment and offensive force variables. This is to be expected, becausethese variables simultaneously influence job approval and GulfCrisisapproval, and entering all three into a single model estimatesonly their direct effects on the dependent variable. Since thesevariables also have significant indirect effects on job approvalthrough Gulf Crisis approval, the trends shown in Figure 2 shouldbe understood as presenting the total effects of each variable on

job approval.

9Gulf crisis approval influenced the probability of approving thepresidents job performance by an average of .60 (s.d. = .10,min = .38, max= .86), while support for troop deployment had anaverage daily effect of .45 (s.d. = .12, min = .23, max = .77) andsupport for offensive force had an average daily effect of .39(s.d. = .17, min = .12, max = .74).



10/18

(p .05, n = 66). In addition, the movement of effectsacross the three trends varies considerably from day today, which suggests that some of these variables arebeing primed differently than others. All of this pointsto the conclusion, consistent with recent experimentalwork on priming effects (e.g., Miller and Krosnick2000; although see Valentino, Hutchings, and White2002), that accessibility alone is unlikely to be the

main engine driving these effects.The tendency for priming effects in all three

trends to increase in size over time likely represents acumulative effect of exposure to Gulf Crisis news. Thelonger the crisis continued, the greater the cumulativeexposure to news about the crisis, and the larger thepriming effects tended to become regardless ofwhether news of the day was giving much currentattention to the crisis. An alternative explanation isthat the central tendencies of the Figure 3 trendsreflect little more than recency effects, produced by theebb and flow of daily news coverage of the Gulf Crisis.

If so, we should find that news coverage about theGulf Crisis rose steadily in volume over the period ofinterest.

However, Figure 3 shows a different pattern intwo big message indicators capturing the dailyamounts of relevant news coverage over the crisisperiod. The first trend (diamonds with a black linetracing a five-day moving average) shows the meannumber of Gulf Crisis stories broadcast on nightlynews programs for each day. This is a fairly broadmeasure of news attention, since it includes stories

that have little or no policy-relevant content (e.g.,descriptive stories about troop readiness or life in theSaudi desert). The other trend (squares with a grayline) shows the average number of discourse ele-ments about the Bush administrations Gulf Crisispolicies appearing each day in nightly news pro-grams. This more refined big message indicator ispositively correlated with the first trend (r= .66), but

they clearly run different courses. The averagenumber of Gulf Crisis stories per broadcast peaks inthe month following the Iraqi invasion of Kuwait,then declines rapidly into a low-coverage periodfrom October through December. In January thenumber of daily stories increases rapidly with thenearing of the January 15th United Nations deadlinefor Iraqi withdrawal, then again reaches August 1990levels in a second peak after the start of the air war.In contrast, the average number of policy discourseelements in a broadcast took a relatively smallerdecline before rebounding to August levels in late

November and early December. Toward the end ofDecember policy discourse increases steadily to peakright before the start of the air war, after which itdeclines rapidly before rebounding again in theweeks leading up to the start of the ground war.

The evaluative tone of this policy discourse alsoevolves over time. Figure 4 shows the number of dis-course elements in an average nightly newscast thatwere favorable (diamonds and a black line showinga five-day moving average) and unfavorable (squaresand a gray line) toward the Bush administrations

FIGURE 3 Big Message Indicators

0

10

20

30

40

50

60

5-Jul

19-Jul

2-Aug

16-Aug

30-Aug

13-Sep

27-Sep

11-Oct

25-Oct

8-Nov

22-Nov

6-Dec

20-Dec

3-Jan

17-Jan

31-Jan

14-Feb

28-Feb

14-Mar

28-Mar

AmountofPolicyDiscou

rse

0

2

4

6

8

10

12

14

16

18

20

NumberofStories

Amount of Policy Discourse

Number of Stories

Start of

Air War

Iraqi Invasion

of Kuwait

Start of

Ground

War

Note: Points within each trend indicate the average daily number of stories or discourse elements per broadcast. Trend lines represent afive-day moving average across these points.



11/18

policies.10 A moderate positive correlation betweenthese trends (p = .59) comes from the tendency forboth types of discourse to increase as the overallamount of news coverage goes up, but the balance offavorable to unfavorable discourse changes consider-ably over time. The flow of favorable discourse followsgenerally the big message trends from Figure 3, withthe highest levels of favorable coverage immediately

after the Iraqi invasion, during the period from lateNovember to early December, and around the start ofthe air war. Unfavorable discourse is muted throughmost of the crisis until the second week of January,when a divided Congress and growing public opposi-tion to the impending war produce news coverage thatis relatively balanced between positive and negativeperspectives. After the start of the air war, unfavorablediscourse tapers off at a faster rate than favorablediscourse, before leveling out over the month ofFebruary.

Comparing changes in the size of priming effects

over time (Figure 2) to the news discourse trends (Fig-ures 3 and 4) reveals a more nuanced relationshipbetween stimulus material and priming effects thanhas been suggested by previous research. News atten-tion to the Gulf Crisis varied over the period in whichpriming effects tended to grow in size, suggesting thatrecent exposure to the news was unlikely to be respon-sible for the steadily increasing effect sizes. Yet the

day-to-day variance in the effect trends is quite pro-nounced, suggesting that daily changes in news cover-age might be driving much of the observed variance.Moreover, the effect trends run different courseswhich suggest that news coverage was stimulatingthem in different ways, or stimulating some trendsmore than others.

To clarify which elements of news content might

be influencing the priming effects in Figure 2, weregressed each of the daily impact trends fromFigure 2 on the four measures of news content in Fig-ures 3 and 4 along with additional measures capturingthe cumulative count of stories and discourseelements that appeared between August 2 and the dateof interview. Comparing the size and significance ofbivariate regression coefficients for different typesof news content reveals clear patterns of primingeffects in this complex information environment.Several patterns stand out in Table 1.

First, nearly all of the news content variables are

positively related at conventional significance levels tothe size of priming effects for Gulf Crisis approval andsupport for offensive force, but only one of them hadeven a marginally significant relationship with the sizeof priming effects attributed to support for troopdeployment. If the priming effects in Figure 2 werebeing driven merely by the accessibility of relevantattitudes, then we would expect to find significant andpositive coefficients in the support for troop deploy-ment series as well. The dearth of such relationshipssuggests that accessibility alone is unlikely to be the

10Neutral discourse elements are omitted from this figure butincluded in the measure of total discourse shown in Figure 3.

FIGURE 4 Tone of Policy Discourse Indicators

0

5

10

15

20

25

5-Jul

19-Jul

2-Aug

16-Aug

30-Aug

13-Sep

27-Sep

11-Oct

25-Oct

8-Nov

22-Nov

6-Dec

20-Dec

3-Jan

17-Jan

31-Jan

14-Feb

28-Feb

14-Mar

28-Mar

AmountofPolicyDiscou

rse

Favorable Discourse

Unfavorable Discourse

Start of

Air War

Iraqi Invasion

of Kuwait

Start of

Ground

War

Note: Points within each trend indicate the average daily number of discourse elements per broadcast. Trend lines represent a five-daymoving average across these points.



12/18

main mechanism driving the observed priming effectsin Figure 2.11

Second, the recency of stimulus exposure seems tomatter more for some trends than for others. The toptwo rows of Table 1 show that daily changes in newsattention and news discourse about the Persian GulfCrisis increased the size of priming effects fromsupport for offensive force more than from Gulf Crisisapproval. To properly interpret the coefficients inTable 1, recall that we define a priming effect as thechange in the weight assigned to an independent vari-able among the factors predicting a dependent vari-able, where impact is measured as the absolutedifference in predicted probabilities when moving

from opposition to support in the independent vari-able. Each news story on the Gulf Crisis in an averagenewscast increased by three points the daily probabil-ity that support for offensive force predicts approval ofhow President Bush was handling his job, but onlyincreased the predictive power of Gulf Crisis approvalby three-quarters of a point. A similar pattern holdsfor daily changes in the amount of policy-relevantdiscourse about the Gulf Crisis. In contrast, recentnews exposure has no significant effects on the weightof support for troop deployment in the attitude struc-ture of presidential job approval.

Third, the cumulative frequency of stimulus expo-sure was generally a more important factor than recentexposure in determining the size of priming effects.Five of the six beta weights for the measures of cumu-lative news exposure are larger than those for recentnews exposure, and one of them reaches marginallevels of significance in raising the weight of supportfor troop deployment in the attitude structure ofpresidential approval.12 For every hundred news

stories focusing on the Gulf Crisis, the predictedpriming effect of support for offensive force on presi-dential approval rose by 7.5 points, while that of GulfCrisis approval rose by 1.5 points, and the effect ofsupport for troop deployment rose by a statisticallynegligible .8 points. Since the average nightly newsprogram had broadcast 993 Gulf Crisis stories prior to

11Another possibility is that the priming effects in this trend arebeing influenced by an artifact of data collection. Since surveyrespondents contacted on the first day of interviewing might besomehow different from respondents contacted in later days (if, forinstance, later interviews came from respondents who were harderto contact), we tested whether the size of priming effects variedsystematically by the day of interview in each of the 20 surveys. Wefound no evidence of systematic bias in priming effects for theoffensive force trend (F [3, 23] = .45, p = .72) or the Gulf Crisisapproval trend (F [3, 70] = .36, p = .78). However, there was asignificant difference in the troop deployment trend (F [3,62] = 3.22, p = .03) caused by a higher average effect size for thefourth day of interviewing (M = .53, compared to means ofbetween .42 and .44 for the first three days of interviewing). It isunlikely that this date of interview artifact is alone responsible fora lack of significant news coverage effects, since Figure 2 showsthat the dynamics of priming effects in this trend took a differentpath relative to the other two independent variables, quite apartfrom any systematic error associated with the final day of inter-viewing. Yet this artifact undoubtedly contributes to the lack ofsignificant relationships between news coverage and the size ofpriming effects for the troop deployment measure.

12In addition, the coefficient for cumulative stories approachedmarginal levels of significance (p = .105) in predicting the size ofpriming effects from support for troop deployment.

TABLE 1 The Bivariate Impact of the Information Environment on the Attitude Structure of PresidentialJob Approval

Effect of Support forTroop Deployment

Effect of Support forOffensive Force

Effect of GulfCrisis Approval

b beta b beta b beta

Recent News CoverageNumber of Stories on Day of Interview .39 .11 3.31** .70 .77* .25Amount of Discourse on Day of Interview .20 .13 .89* .48 .43** .32

Cumulative News CoverageCumulative Stories (100s) .81 .20 7.52** .68 1.45** .45Cumulative Discourse (100s) .49 .21 3.51** .69 .94** .49

Tone of Discourse VariablesFavorable Discourse on Day of Interview .13 .05 1.04 .33 .46 .21

Unfavorable Discourse on Day of Interview .61 .19 1.78** .53 1.03** .37N of Days = 66 27 74

p .10, *p .05, **p .01.Note: Cells contain bivariate OLS regression coefficients. Cumulative news measures are scored in hundreds of stories or discourse

elements.



13/18

the start of the air war, as the crisis unfolded the pre-dicted impact of cumulative news exposure soonbecame much larger than that of recent news expo-sure. For instance, at the start of the air war the pre-dicted impact of cumulative news stories was toincrease the priming effect of Gulf Crisis approval by

nearly 15 points. By comparison, the average numberof Gulf Crisis stories per newscast was 6.7 per day overthe entire crisis period, which predicts a 5.2 pointincrease in the size of that same priming effect as aresult of recent news exposure.

Fourth, the evaluative tone of Gulf Crisis newscoverage was an important factor in producingpriming effects. Unfavorable policy discourse signifi-cantly primed support for offensive force and Gulfcrisis approval,13 while favorable policy discourse wasonly marginally significant in priming Gulf crisisapproval. Support for offensive force and Gulf Crisis

approval had substantially greater weight in predicting job approval on days in which the news containedhigher levels of unfavorable policy discourse. More-over, in every case the impact of unfavorable discoursewas nearly twice as large as that for favorable dis-course. This is another piece of evidence inconsistentwith the idea that accessibility alone is the main enginedriving priming effects.

The bivariate coefficients in Table 1 challenge agood deal of conventional wisdom about primingeffects. Yet these coefficients are also ambiguousbecause they represent the total effects of each support

variable on job approval, rather than their uniqueeffects when simultaneously controlling for the influ-ence of the other support variables. Because theGallup surveys rarely included all three support mea-sures in their surveys at once, our ability to parcel outthe unique effects of each variable is quite limited (seenote 8). A second source of ambiguity is that multipledimensions of news coverage are likely to be shapingpriming effects at the same time, but up to this pointwe have considered each in isolation from the others.We therefore ran multiple regression models thatsimultaneously estimated the effects of recent and

cumulative news exposure, as well as the tone of GulfCrisis discourse. To minimize collinearity as well as thenumber of coefficients needing to be estimated fromsmall numbers of cases, this multivariate analysis usesa differenced measure of discourse tone (favorable dis-course minus unfavorable discourse appearing in eachdays average newscast).

The multivariate results in Table 2 generallysupport the findings from the bivariate analysis inTable 1. Compared to recent exposure, cumulativenews exposure remains a stronger predictor of the sizeof priming effects in five of the six equations, generallyhaving nearly twice the impact of recent news expo-

sure. But these relationships are robust across modelsonly when predicting the size of priming effects fromGulf Crisis approval: in these models, recent andcumulative exposure were both significant predictorsof priming effects regardless of whether they wereoperationalized as discourse elements or discretenumbers of relevant news stories. By contrast, in theequations predicting the size of priming effects fromsupport for offensive force, cumulative exposuredrives priming effects when news exposure is mea-sured as the amount of policy-related discourse in thenews, but recent exposure drives those effects when

news exposure is measured as numbers of storiesabout the Gulf Crisis.14 This difference is likely anartifact of when the offensive force series began: theother two support variables were measured over alonger period with more variance in daily news cover-age, but the offensive force variable was measured onlylate in the crisis when both cumulative and dailyamounts of news coverage were rising in tandem. Thisunfortunate coincidence makes it difficult to sort outwhether priming effects for offensive force supportwere produced more by cumulative or recent newsexposure, but the R-squared values for both equations

show that news exposure of whatever origin explainedaround half of the variance in the size of primingeffects for this variable. As was the case with Table 1,none of the news variables turns out to be a significantpredictor of priming effects from the support fortroop deployment variable.

The main difference between the bivariate andmultivariate results is that tone of policy discoursefails to become a significant predictor of primingeffects in any of the six multiple regression equations.Nonetheless, the negative signs on these coefficientsare in the expected direction: the more negative the

news on a particular day, the larger the priming effectfrom the support variables even after controlling forthe effects of recent and cumulative news exposure.Although our analysis produces mixed results on theimpact of news tone on priming effects, we believethat these findings are sufficiently promising to merit

13In addition, the coefficient for unfavorable policy discourse inpredicting the size of priming effects for support for troop deploy-ment was significant at the p = .12 level.

14In the support for the offensive force model, the coefficient forcumulative stories nearly reaches conventional levels of signifi-cance (p = .108).



14/18

further research on the relationship between tone ofnews discourse and the size of priming effects.

Conclusion

Taken together, the findings in Tables 1 and 2 suggesta pattern of nuanced influences on priming effectsrather than a mechanistic and determinative relation-ship between construct accessibility and constructweight in relevant attitude structures. We confirm thatbig message accessibility effects can influencepriming in complex information environments, but inmore subtle ways than previous studies have identi-fied. Recent exposure to relevant news content cangenerate priming effects, but cumulative exposure to

relevant news tended to be a more important factorinfluencing the size of priming effects in the case con-sidered here. The greater the cumulative exposure torelevant stimuli, the greater the likelihood that meremention of relevant stimuli triggered priming ofapplicable attitudes, regardless of the amount ofrecent coverage in the news.

These findings help to clarify why experimentsand surveys have tended to draw such different con-clusions about the duration of priming effects: natu-rally occurring priming effects seem to have both

short-term and long-term components. Experimentsare likely to have captured a mix of short-term effectsstimulated by recent exposure in the laboratory pluswhatever cumulative exposure subjects brought with

them to the lab. In contrast, survey-based primingstudies are likely to have captured long-term effectsfrom the cumulative frequency of exposure purged ofrecency effects. Our study, the first to document theseparate effects of recent and cumulative news expo-sure, suggests that apparently contradictory conclu-sions from experimental and survey-based research onthe duration of priming are both correct. The appar-ent contradiction arises because they are capturingdifferent components of priming effects.

Our findings also suggest that priming in complexinformation environments seems to be a function of

applicability effects as well as construct accessibility.Contrary to the claim that priming is driven by acces-sibility alone, no measured aspect of the informationenvironment was significantly and consistently relatedto changes in the size of priming effects across theboard. To the contrary, the pattern of findings makesmore sense when we consider how construct accessi-bility appears to be moderated by applicability effects.The impact of factors other than construct accessibil-ity took five different forms in our findings: in thedifferent dynamic courses taken by priming effects

TABLE 2 The Multivariate Impact of the Information Environment on the Attitude Structure ofPresidential Job Approval

Effect of Support forTroop Deployment

Effect of Supportfor Offensive Force

Effect of GulfCrisis Approval

b beta b beta b beta

Amount of DiscourseAmount of Discourse on Day of Interview .16 .10 .35 .19 .32* .23Cumulative Discourse (100s) .37 .16 2.97** .58 .79** .41Discourse Tone (Differenced) on Day of

Interview-.41 -.14 -.58 -.17 -.34 -.13

Constant 40.74** -4.42 50.40**

Adj. R2 = .02 .46 .26

N of Days = 66 27 74

Number of News StoriesNumber of Stories on Day of Interview .45 .13 2.12* .45 .88** .29Cumulative Stories (100s) .70 .17 3.81 .34 1.41** .44Discourse Tone (Differenced) on Day of

Interview

-.41 -.14 -.57 -.17 -.35 -.13

Constant 38.24** -8.93 45.91**Adj. R2 = .02 .52 .25

N of Days = 66 27 74

p .10, *p .05, **p .01.Note: Cells contain OLS multiple regression coefficients. Cumulative news measures are scored in hundreds of stories or discourseelements.



15/18

from the three support variables; in the tendency forGulf Crisis approval effects to be consistently largerthan effects of support for troop deployment, whichin turn were consistently larger than the effects ofsupport for offensive force; in the tendency for newsdiscourse to have larger effects on the size of priming

effects from Gulf Crisis approval than on those fromsupport for offensive force; in the tendency for unfa-vorable discourse to produce larger priming effectsthan favorable discourse; and in the finding that newsdiscourse had no consistently significant effects on theweight of support for troop deployment in the attitudestructure of presidential job approval. Data limitationsin the Gallup surveys preclude us from testing specificpsychological mechanisms that might be contributingto these observed patterns, but the important point isthat these patterns cannot be explained by accessibilityeffects alone.

These findings, coming as they do from a casestudy of a unique event, are unlikely to be universal ordefinitive. Moreover, each of the observed relation-ships between news primes and observed effects, whileplausible, would be difficult to predict in advancegiven the current state of the literature. Thus, the valueof this analysis lies less in the light that it sheds on thechanging structure of job approval attitudes than inthe challenge it presents for the conventional wisdomregarding the causes and cognitive mechanismsunderlying priming effects stimulated by news cover-age. This conventional wisdom has developed largely

without reference to a broad range of empiricalresearch conducted by social psychologists, researchthat is entirely consistent with the findings reportedhere.

Our approach naturally has its share of limita-tions relative to the standard methods of assessingpriming effects outside the laboratory. While theo-retical considerations guided the particular choiceswe made in operationalizing assumptions aboutpriming effects from the social psychological litera-ture, the limited availability of survey data allowed usto test only rudimentary attitude structures. We are

also fully aware that our choice to analyze primingeffects within the fairly unique context of the GulfCrisis reduces our confidence in generalizing to othercases from the patterns revealed in our analysis.However, since so much of the research on news-induced priming effects has been conducted on theGulf Crisis, our choice of cases seems appropriate fora study aiming to reassess the standard approaches tostudying priming effects.

While we observe patterns that are consistent withexpectations about applicability effects, our study

cannot provide strong evidence of applicability effectsbecause the Gallup data used in this analysis containno direct means of measuring construct applicability.The presumed applicability effects that we observecould be influenced by the nature of the news stimu-lus, by the ways that individual motivations and pre-

dispositions affect judgmental processes, or by acombination of both. Future research on newspriming will need to grapple with the challenge ofsorting out which of various kinds of applicabilityeffects might be moderating priming processes.

Finally, our approach to modeling the dailydynamics of priming effects using survey data invitesany number of possible confounds or spurious rela-tionships to obscure the precise mechanisms thatproduce observed effects. In particular, replacing thenotion of a big message effect with one that allowsfor the dynamic flow of multiple dimensions to news

discourse makes it difficult to pin down preciselywhich elements of news content might be responsiblefor observed changes in the structure of attitudes. Wemodeled three such dynamicsrecency, cumulativefrequency, and evaluative toneand found effects foreach, but our study might have produced differentresults had we chosen to model different dimensionsof news coverage.

We believe that the methodological approachdeveloped in this paper, suitably refined, holdspromise for future work in priming effects. We arehopeful that replications of this method in other cases

and other contexts will clarify the ways that accessibil-ity and applicability moderate the impact of primingeffects produced by exposure to news content, as wellas the features of news content most likely to moderatethese effects.

Acknowledgments

The authors thank Michael Cobb for coding theevening news data and the Research Board at the Uni-

versity of Illinois at Urbana-Champaign for fundingthe collection of these data. We are indebted to Kath-leen Frankovic, Nehemiah Geva, Philip Habel, JenniferJerit, James Kuklinski, Kathleen McGraw, ThomasNelson, Paul Quirk, Thomas Rudolph, Mira Sotirovic,David Tewksbury, and members of the UIUC CitizenCompetence Group for their assistance and helpfulcomments on various aspects of this project.

Manuscript submitted 7 January 2005Manuscript accepted for publication 17 February 2006



16/18

References

Althaus, Scott L. 2002. American News Consumption duringTimes of National Crisis. PS: Political Science & Politics 35 (3):51721.

Althaus, Scott L. 2003. When News Norms Collide, Follow theLead: New Evidence for Press Independence. Political Commu-

nication 20 (4): 381414.Anderson, Craig A. 1997. Effects of Violent Movies and Trait

Hostility on Hostile Feelings and Aggressive Thoughts.Aggres-sive Behavior23: 16178.

Anderson, John R. 1983. The Architecture of Cognition. Cambridge,MA: Harvard University Press.

Danielian, Lucig H.,and Stephen D. Reese.1989.A Closer Look atIntermedia Influences on Agenda Setting: The Cocaine Issue of1986. In Communication Campaigns about Drugs: Government

Media and the Public, ed. P. J. Shoemaker. Hillsdale, NJ:Lawrence Erlbaum.

Domke, David, Dhavan V. Shah, and Daniel B. Wackman. 1998.Media Priming Effects: Accessibility, Association, and Activa-tion. International Journal of Public Opinion Research 10 (1):

5174.Dorman, William A., and Steven Livingston. 1994. News and

Historical Content: The Establishing Phase of the Persian GulfPolicy Debate. In Taken by Storm: The Media, Public Opinion,and U.S. Foreign Policy in the Gulf War, eds. W. L. Bennett andD. L. Paletz. Chicago: University of Chicago Press.

Druckman, James N. 2004. Political Preference Formation: Com-petition, Deliberation, and the (Ir)Relevance of FramingEffects. American Political Science Review 98 (4): 67187.

Druckman, James N., and Kjersten R. Nelson. 2003. Framing andDeliberation: How Citizens Conversations Limit Elite Influ-ence. American Journal of Political Science 47 (4): 72945.

Entman, Robert M. 2003. Projections of Power: Framing News,Public Opinion and U.S. Foreign Policy. Chicago: University of

Chicago Press.Fan, David P. 1988. Predictions of Public Opinion from the Mass

Media. Westport, CT: Greenwood Press.

Fein, S., and S. J. Spencer. 1997. Prejudice as Self-Image Mainte-nance: Affirming the Self Through Derogating Others.Journalof Personality and Social Psychology 73: 3144.

Gamson, William A., and Andre Modigliani. 1989. Media Dis-course and Public Opinion on Nuclear Power: A Construction-ist Approach. American Journal of Sociology95 (1): 137.

Gans, Herbert. 1979. Deciding Whats News: A Study of CBSEvening News, NBC Nightly News, Newsweek, and Time . NewYork: Pantheon.

Higgins, E. Tory. 1996.Knowledge Activation: Accessibility, Appli-cability, and Salience. In Social Psychology: Handbook of BasicPrinciples, eds. E. T. Higgins and A. W. Kruglanski. New York:Guilford.

Higgins, E. Tory, and Gillian A. King. 1981. Accessibility of SocialConstructs: Information-Processing Consequences of Indi-vidual and Contextual Variability. In Personality, Cognition,and Social Interaction, eds. N. Cantor and J. Kihlstrom. Hills-dale, NJ: Erlbaum.

Iyengar, Shanto, and Donald Kinder. 1987. News that Matters:Television and American Opinion. Chicago: University ofChicago Press.

Iyengar, Shanto, Donald R. Kinder, Mark D. Peters, and Jon A.Krosnick. 1984. The Evening News and Presidential Evalua-tions. Journal of Personality and Social Psychology46: 77887.

Iyengar, Shanto, Mark D. Peters, and Donald R. Kinder. 1982.Experimental Demonstrations of the Not-So-Minimal Con-sequences of Television News Programs. American PoliticalScience Review 76 (4): 84858.

Iyengar, Shanto, and Adam Simon. 1994. News Coverage of theGulf Crisis and Public Opinion: A Study of Agenda-Setting,Priming, and Framing. In Taken by Storm: The Media, PublicOpinion, and U.S. Foreign Policy in the Gulf War, eds. W. L.

Bennett and D. L. Paletz. Chicago: University of Chicago Press.Kim, Young Mie. 2005. Use and Disuse of Contextual Primes in

Dynamic News Environments. Journal of Communication 55(4): 73755.

Kirton, John. 1993. National Mythology and Media Coverage:Mobilizing Consent for Canadas War in the Gulf. PoliticalCommunication 10: 42541.

Klapper, Joseph. 1960. The Effects of Mass Communication.Glencoe, IL: Free Press.

Krosnick, Jon A., and Laura A. Brannon. 1993. The Impact of theGulf War on the Ingredients of Presidential Evaluations: Mul-tidimensional Effects of Political Involvement. AmericanPolitical Science Review 87 (4): 96375.

Krosnick, Jon A., and Donald R. Kinder. 1990.Altering the Foun-dations of Support for the President Through Priming.Ameri-can Political Science Review 84: 497512.

Kruglanski, Arie W. 1989. Lay Epistemics and Human Knowledge:Cognitive and Motivational Bases. New York: Plenum.

Lang, Gladys Engel, and Kurt Lang. 1994. The Press as Prologue:Media Coverage of Saddams Iraq, 19791990. In Taken byStorm: The Media, Public Opinion, and U.S. Foreign Policy in theGulf War, eds. L. Bennett and D. Paletz. Chicago: University ofChicago Press.

Lau, Richard R. 1989. Construct Accessibility and ElectoralChoice. Political Behavior11 (1): 532.

Lodge, Milton, and Charles Taber. 2000. Three Steps toward aTheory of Motivated Political Reasoning. In Elements ofReason: Cognitions, Choice, and the Bounds of Rationality, eds.A. Lupia, M. McCubbins, and S. Popkin. New York: Cambridge

University Press.Martin, Leonard, and John Achee. 1992.Beyond Accessibility:The

Role of Processing Objectives in Judgment. In The Construc-tion of Social Judgments, eds. L. Martin and A. Tesser. Hillsdale,NJ: Erlbaum.

McCombs, Maxwell, and Donald Shaw. 1972.The Agenda-SettingFunction of Mass Media. Public Opinion Quarterly36: 17687.

Mendelberg, Tali. 2001.The Race Card: Campaign Strategy, ImplicitMessages, and the Norm of Equality. Princeton: Princeton Uni-versity Press.

Mermin, Jonathan. 1996. Conflict in the Sphere of Consensus?Critical Reporting on the Panama Invasion and the Gulf War.Political Communication 13 (2): 18194.

Mermin, Jonathan. 1999. Debating War and Peace: Media Coverage

of U.S. Intervention in the Post-Vietnam Era. Princeton: Princ-eton University Press.

Miller, Joanne M., and Jon A. Krosnick. 1996.News Media Impacton the Ingredients of Presidential Evaluations: A Program ofResearch on the Priming Hypothesis. In Political Persuasionand Attitude Change, eds. D. C. Mutz, P. M. Sniderman andR. A. Brody. Ann Arbor: University of Michigan Press.

Miller, Joanne M., and Jon A. Krosnick. 2000.News Media Impacton the Ingredients of Presidential Evaluations: PoliticallyKnowledgeable Citizens are Guided by a Trusted Source.

American Journal of Political Science 44 (2): 295309.

Nelson, Thomas E. 1999. Public Rhetoric, Policy Goals, andPolitical Attitudes. Paper presented at the annual meeting ofthe American Political Science Association, Atlanta.



17/18

Pan, Zhongdang, and Gerald M. Kosicki. 1993.Framing Analysis:An Approach to News Discourse. Political Communication 10:5575.

Pan, Zhongdang, and Gerald M. Kosicki. 1997. Priming andMedia Impact on the Evaluations of the Presidents Perfor-mance. Communication Research 24 (1): 330.

Power, J. Gerard, and Sheila T. Murphy. 1996. Priming Prejudice.Human Communication Research 23: 136.

Price, Vincent, and David Tewksbury. 1997. News Values andPublic Opinion: A Theoretical Account of Media Priming andFraming. In Progress in Communication Sciences, eds. G.Barnett and F. Boster. London: Ablex Publishing.

Roskos-Ewoldesen, David R., Beverly Roskos-Ewoldesen, andFrancesca R. Dillman Carpentier. 2002. Media Priming: ASynthesis. In Media Effects: Advances in Theory and Research,eds. J. Bryant and D. Zillmann. Mahwah, NJ: LawrenceErlbaum Associates.

Shrum, L. J., and Thomas C. OGuinn. 1993.Processes and Effectsin the Construction of Social Reality. Communication Research20: 43671.

Sniderman, Paul M., and Sean M. Theriault. 2004.The Dynamicsof Political Argument and the Logic of Issue Framing. In

Studies in Public Opinion: Gauging Attitudes, Nonattitudes, Measurement Error and Change, eds. W. E. Saris and P. M.Sniderman. Princeton: Princeton University Press.

Sparrow, Bartholomew H. 1999. Uncertain Guardians: The NewsMedia as a Political Institution. Baltimore: Johns Hopkins Uni-versity Press.

Thompson,Erik P., Robert J. Roman,Gordon B. Moskowitz,ShellyChaiken, and John A. Bargh. 1994. Accuracy MotivationAttenuates Covert Priming Effects: The Systematic Reprocess-ing of Social Information. Journal of Personality and SocialPsychology66: 47489.

Trope, Yaacov, and Zvi Ginossar. 1988. On the Use of Statisticaland Nonstatistical Knowledge: A Problem-Solving Approach.In The Social Psychology of Knowledge, eds. D. Bar-Tal and A. W.

Kruglanski. New York: Cambridge University Press.Valentino, Nicholas A. 1999. Crime News and the Priming of

Racial Attitudes during Evaluations of the President. PublicOpinion Quarterly63 (3): 293320.

Valentino, Nicholas A., Vincent L. Hutchings, and Ismail K. White.2002. Cues that Matter: How Political Ads Prime Racial Atti-tudes during Campaigns.American Political Science Review 96(1): 7590.

Zaller, John. 1992. The Nature and Origins of Mass Opinion. NewYork: Cambridge University Press.

Scott L. Althaus is associate professor of political

science and speech communication, University of Illi-nois at Urbana-Champaign, Urbana, IL 61801. YoungMie Kim is assistant professor of communication, TheOhio State University, Columbus, OH 43210.



18/18

Journal of Politics 2006 Althaus

Documents

Transcript of Journal of Politics 2006 Althaus