Visual Attention to Repeated Print Advertising

54 downloads 913 Views 3MB Size Report
Dec 11, 2015 - Wedel is Professor of Marketing Research, Department of Economics,. University ...... Further research might examine the effect of media plan-.
Visual Attention to Repeated Print Advertising: A Test of Scanpath Theory Author(s): Rik Pieters, Edward Rosbergen and Michel Wedel Source: Journal of Marketing Research, Vol. 36, No. 4 (Nov., 1999), pp. 424-438 Published by: American Marketing Association Stable URL: http://www.jstor.org/stable/3151998 Accessed: 11-12-2015 09:09 UTC REFERENCES Linked references are available on JSTOR for this article: http://www.jstor.org/stable/3151998?seq=1&cid=pdf-reference#references_tab_contents You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/ info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

American Marketing Association is collaborating with JSTOR to digitize, preserve and extend access to Journal of Marketing Research.

http://www.jstor.org

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

RIKPIETERS,EDWARDROSBERGEN,and MICHEL WEDEL* The authors examine consumers' visual attentionduringrepeated exposures to printadvertisementsusing eye-trackingmethodology.The authors propose a statistical model comprising submodels for two key measures of visual attentionto elements of the advertisement:attention durationand inter-and intraelementsaccades. These measures are vital to understandingattentionwear-outand the impact of repetitionon advertising effectiveness but have not been considered in previous research. The results of two studies show that (1) attention duration decreases significantly across advertising repetitions and (2) attentional scanpaths, as measured through saccades, remain constant across advertising repetitions and across experimentally induced and naturally occurring conditions. Scanpaths obey a stationary, reversible, first-order Markov process. The authors offer implications for attention theory and advertising processing.

Visual

Attention

Advertising: A

Repeated

to

Test

Scanpath

of

The effectiveness of repeated print advertisementsmay decrease because of attention, learning, and acceptance wear-out (Rossiter and Percy 1997). Most researchto date has emphasized learning and acceptance wear-out and examined the relationshipsamong repeatedadvertisingexposure, cognitive processes, and affect toward the ad (cf. Anand and Sternthal1990; Haugtvedtet al. 1994; Rethans, Swasy, and Marks 1986). This researchhas revealed a nonmonotonic inverted-U relationship between exposure and affect toward the ad and has identified several factors that moderatethe relationship.To date, few studies have examined attention wear-out. This is surprisingbecause of the centralrole assigned to attentionin advertisingcommunication processes (Rossiter and Percy 1983, 1997; Shanteau 1983), particularly in advertising wear-in and wear-out across repetitions (Batra and Ray 1986; Craig, Sternthal, and Leavitt 1976; Pechmannand Stewart 1989). Calderand Sternthal(1980) arguethatinattentionis a majorcause of re*Rik Pieters is Professor of Marketing, Department of Business Administration, Tilburg University, The Netherlands (e-mail: F.G.M. [email protected]).Edward Rosbergenpresentlyis employed at MuConsult, Amersfoort,The Netherlands(e-mail: ). Duringthe researchon which this article is based, he was a doctoralcandidatein MarketingResearch.Michel Wedel is Professor of Marketing Research, Departmentof Economics, University of Groningen, The Netherlands (e-mail: [email protected]). This research was sponsored by Verify Nederland, Rotterdam,and the Economic Research Foundation, which is part of the Netherlands Organizationfor Scientific Research (NWO). The authors gratefully acknowledge the support of Dominique Claessens of Verify Nederland throughoutthe research.To interactwith colleagues on specific articles in this issue, see "Feedback"on the JMR Web site at www.ama.org/pubs/jmr.

Journal of MarketingResearch Vol. XXXVI (November 1999), 424-438

Print Theory

peated advertising'swear-out;that is, with increasedrepetition, consumers pay progressively less attention to advertisements.This contentionreceives indirectsupportin prior research on learning wear-out, which shows a decline in brandname recall when advertisingrepetitionincreases beyond some threshold (Craig, Sternthal,and Leavitt 1976). However, empirical supportusing direct measuresof attention wear-out is not available (cf. Unnava and Burnkrant 1991). This study addressesattentionwear-outand is the first to analyze visual attention to repeated advertising using eye tracking.We propose a methodology that enables us to examine closely specific patternsof visual attentionto key elements of advertising.The methodology builds on scanpath theory (Noton and Stark 1971; Stark 1994) that postulates relatively stable patternsof eye fixations. These patternsfacilitate identification of the advertisements'content in repeated exposures. In addition, we investigate if and how these scanpaths change across advertising repetitions and whetherotherfactorsaffect attentionwear-out,in particular, ad design and consumer motivation. Research on ad repetition typically has used television commercials (Axelrod 1980; Batra and Ray 1986; Belch 1982; Calder and Sternthal 1980; Haugtvedt et al. 1994; Rethans, Swasy, and Marks 1986), which are characterized by fixed, externally controlled exposure duration.In contrast, consumers control the exposure durationto print advertisingthemselves. This offers them ways to adaptto repetition by reducingor increasingthe durationof exposure to advertisements.The few studies on repetition of print advertising typically either have examined advertising under

424

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

425

Scanpath Theory fixed exposure duration(Bergerand Mitchell 1989; Unnava and Burnkrant1991) or have not measuredexposure duration at all (Cox and Cox 1988). As a consequence, little is known about the effect of repetitionon attentionto advertising if the exposure duration is under consumer control, such as for print, outdoor, Internet,and yellow page advertisements. In particular,when exposure duration is under consumer control, the investigation of attentionwear-outis importantbecause it may preclude the occurrenceof learning or acceptance wear-out.That is, attentionwear-outmay cause consumersto terminateexposureto advertisingbefore the other types of wear-outset in. Rossiterand Percy (1997, p. 602) state that "Diminishedattentionis particularlylikely to affect campaigns in print media where the easy response for the consumer or customeris to turnthe page." In this study, we examine the effect of repeatedexposure on consumers' attention to print advertising, using eye recordingsas measures of visual attention.Here, consumers control the duration of the exposure to the print advertisements. We examine measures of attention to the key elements of an advertisement:headline, pictorial, body text, and packshot.We conduct two studies to comparethe effects of repetition on attention in different set-ups. In the first study, consumers' motivation to attendto a printadvertisement for shampoo and the quality of argumentsin the advertisementare manipulatedsystematicallyto establish the generalizability of repetition effects across those relevant consumer and stimulus conditions (Batraand Ray 1986). In the second study, attentionto three advertisementsfor three different personal care productsis examined with naturally occurring differences in consumers' motivation to explore the generalizability of repetition effects across different study and ad designs. We proposea stochasticmodel of consumers' visual attention to advertisingthat enables detailed analyses of the durationof attentionto key elements and of the eye movements within and between those elements. In this way, insight is obtained into patternsof visual attention that have not been examined previously but that seem crucial to understandthe processing and effectiveness of repeated printadvertising. VISUALATTENTIONTOADVERTISING Because eyes go where attention is allocated (Rayner 1995), eye movements provide reliable informationabout consumers' visual attention to advertisements(Krugmanet al. 1994; Rosbergen, Pieters, and Wedel 1997). Apart from several smaller, corrective eye movements, eye movement data are composed of saccades and fixations. Saccades are the quick jumps of the eye from location to location, during which vision essentially is suppressed. Fixations are the pauses between saccades, duringwhich the eye is relatively immobile. They are an importantaspect of visual attention, and the evaluationof an advertisement'spotentialto gain attention should be based on the duration,position, and pattern of these fixations. In basic research on reading (Rayner 1995) and visual scanning (Viviani 1990), fixations typically are analyzed at the most desaggregatelevel, at which the exact position and durationof each individual fixation is retained.In studying visual attentionto print advertising,there are advantagesto assigning fixations to areas instead of retainingtheir exact positions. First, in advertisingdevelopmentand subsequent

copy testing, the goal is to understandwhetherand how frequently consumers attend to ad elements, such as brand name, product,headline, and pictorial, ratherthan to know the exact coordinatesof consumers'fixations on, for example, the headline. Print recognition copy tests, such as Starch,also focus on those key ad elements. Second, at each eye fixation, the visual field covers more thanjust the exact fixation point because of foveal ("yellow spot") and parafoveal (the surroundingarea of the retina) vision. Because a largerpart of the advertisementis covered in each fixation than is suggested by exact fixation positions (Rayner 1995), predefinedareasof the advertisementare more appropriateas the unit of analysis than exact fixation points. Third, with the numberof fixations and saccades that occur during a single exposure to an advertisement,the dimensionality of the databecomes prohibitivelylargewhen based on exact fixation positions insteadof on a limited numberof areas. Aggregationof raw eye-trackingdata is thus desirable in studying attentionto advertisements,but the level of aggregation must be determined. Previous advertising research typically has used global informationsuch as the gaze duration, that is, the total time subjects spent on the advertisement or its elements, to examine visual attention(e.g., Celsi and Olson 1988; Janiszewski 1993; Rosbergen,Pieters, and Wedel 1997). In this study, we use informationabout gaze durationas well, but to examine attentionscanpathsto print advertisementsin detail, we supplant this with more finegrained informationabout inter-and intraelementsaccades. Whereas previously, eye fixations have been treateddeterministically,we offer a stochastic model to analyze element-level eye-trackingdata for printadvertisementsand to examine regularitiesin fixation patterns.The model is based on scanpaththeory (Noton and Stark 1971; Stark 1994), and it accounts for the impact of repetitionand exogenous variables on those patterns. Before presenting the model, we first introducerelevantattentionresearch,in particularscanpath theory,and formulatehypothesesabout visual attention to repeatedprintadvertising. HYPOTHESESABOUTVISUALATTENTIONTO REPEATEDADVERTISING AttentionDurationDecrementAcross Advertisement Repetitions Attention to a television commercial appearsto increase initially with repeatedexposures and to decrease after a satiation point is reached,which is often aftertwo or threeexposures (e.g., Belch 1982; Cacioppo and Petty 1979, 1980; Calder and Sternthal 1980). With externally paced media such as television and radio, one exposure to a commercial message may be insufficient to grasp its content fully, in which case additionalexposures are needed. It is not obvious that such an attention patternalso emerges for advertisements in internally paced media such as newspapers, magazines, and the Internet.With internally paced media, consumers can pay as much attentionas they desire during the first exposure and can move freely to the next page whenever they are ready. Several researchershave argued that, for printadvertising,one exposure may suffice to communicate its message (Calderand Sternthal1980; Krugman 1972). In that case, the amount of attentionpaid to the advertisementwill decline after the first exposure. Because a

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

426

JOURNAL OF MARKETINGRESEARCH, NOVEMBER 1999

glimpse of an advertisementsuffices to identify its familiar content,subjectswould spend less time on advertisementsin subsequent exposures (Krugman 1972). This leads to the following hypothesis: Hi: Gaze durationand fixationfrequencyto printadvertisementsdecreaseacrossrepeatedexposures. ScanpathRobustnessAcrossAdvertisementRepetitions Scanpath theory (Noton and Stark 1971; Stark 1994; Stark and Ellis 1981) conjectures that sequences of fixations, called "scanpaths,"that occur during the first exposure to a stimulus reoccur during subsequentexposures. A spatial model, considereda precognitive,perceptualmodel, is assumedto control the sequences of eye movements.As a case in point, researchindicatesthat when a subjectengages in visual imagery,looking at a blank screen and visualizing a previously seen figure, the scanpathis similar to when he or she viewed the figure (Stark 1994). Ellis and Smith (1985) proposethatscanpathscan be generated by completely random,stratifiedrandom,or statistically dependent stochastic processes, but they did not test those conjectures. A completely random process assumes that each element of an advertisementhas equal probability of being focused on duringeach fixation. If eye movements would follow a stratifiedrandom(or a zero-order)process, the probabilitiesof ad elements being fixated reflect the attentional attractivenessof those elements, but they do not depend on previous fixations. In view of the perceptual processes that are assumed to underlie eye movements (Rayner 1995; Starkand Ellis 1981), it is unlikely that saccades from one fixation point to anotherare generatedby either completely random or stratified random processes. A statistically dependent stochastic process, however, specifies thatthe position of a fixation depends on previous fixations. This would imply that the patternof fixations can be described by a Markov process (Ross 1996). Molnar and Ratsikas (1987) propose that each fixation's position only depends on the directly precedingfixation point. From this, it would follow thatthe dependencebetween successive fixations on ad elements follows a first-orderMarkovprocess. Ellis and Smith (1985) argue that saccades from, say, element A to B occur as often as saccades from B to A. This means that the Markov transition probabilitiesdescribing the scanpathare symmetrical,or that the first-orderMarkov process is (time-) reversible(Ross 1996). Scanpaththeorypredictsthata subjectscans a new stimulus duringthe first exposureand stores the sequence of fixations in memory as a spatialmodel, so that a scanpathis established.When the subjectis reexposedto the stimulus,the eyes tend to follow the same scanpath,which facilitatesstimulus recognition.The theorythus suggests thatscanpathsare stationary across exposures. Surprisingly,scanpaths have been treateddeterministicallyin empirical researchto date (e.g., Groner 1988; Gronerand Menz 1985), and the appropriatenessof the Markovassumptionsfor scanpathshas remaineduntested.We test the following hypothesisaboutconsumers'scanpathsacross repeatedadvertisements: H2:Fixationsto elementsof a printadvertisement (a) dependon thelocationof thepreviousfixationand(b) areindependent of the fixation orderof the elements, and (c) theirpatternremains constant across repetitions.

H2 is supportedif scanpathscan be described by a stationary,reversible,first-orderMarkovprocess. Supportfor both HI and H2 would imply thatthe durationof attentionto print advertisingdecreasesacross repetitionsbut that,at the same time, the scanpathto elements of print advertisingremains constant. ScanpathRobustnessAcross Consumerand Message Differences In a review of the attitudinaleffects of advertisingrepetition, Pechmannand Stewart (1989, p. 287) argue that such effects are contingent on "whether the ad persuades via emotional images or verbalarguments,whetherinitially it is a high or low scoringad (for example, whetherthe verbalargumentsare strongor weak), and whetheror not consumers are motivated and able to process the ad." We examine the robustness of attentional effects of advertising repetition, following the suggestions of Pechmannand Stewart(1989). Specifically, we test the robustnessof attentiondurationand scanpathsacross differences in consumermotivationand argument quality,as is explained subsequently. Kroeber-Riel (1993) reports that motivated consumers pay more attentionto textual elements than less motivated consumers do, and Celsi and Olson (1988) find that motivated consumersoverall spent more time processing advertisements. Research on the interactionbetween consumer motivation and advertising repetition on patternsof visual attention is absent. We test whether motivated consumers pay more attentionto printadvertisingand its elements and whether attention patterns are robust across motivation conditions. Many studies have found significant effects of argument quality on attitude formation and change (e.g., Petty, Cacioppo, and Schumann1983; Unnava and Burnkrant1991), but none has examined argumentquality effects on attention. Attitudeformationand change are conceptualprocesses that follow and build on the perceptualprocesses that are centralto visual attention(cf. Greenwaldand Leavitt 1984; Viviani 1990). Argumentquality is a message variablethat is the focus of conceptual processes during advertisingexposure.Previousresearchhas found thatphysical featuresof advertisements(size, location, color) affect visual attention (Rosbergen,Pieters, and Wedel 1997), but as long as differences in argumentqualitydo not translateinto differencesin the physical featuresof advertisements,argumentquality is unlikely to affect visual attention.This leads to the following hypothesis: H3: (a) Consumer motivation has a positive effect, and (b) argument quality has no effect on duration of attention to advertisements.

If HI is supported,attention durationdiminishes across successive exposures to printadvertising.If H3aand H3bare supported,consumershigh in motivationpay more attention to print advertising than consumers low in motivation do, but the quality of argumentshas no significant effect on attention duration. Finally, direct evidence regardingthe effects of motivation and argumentquality on consumers' scanpaths is absent. Molnar and Ratsikas (1987, p. 371) argue that "there are no differences between subjects in the statistical structure of visual exploration,"implying that scanpathsdo not

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

427

Scanpath Theory differ systematically between subjects. They observe that many sequences of three,four, and sometimes five fixations were nearly identicalfor several subjects. Similarly,Groner and Menz (1985) find supportthat scanpathsdo not differ across subjects. The observed stability across subjects suggests that neither motivation nor argument quality affect consumers'scanpaths,but these relationshipshave not been tested experimentally.We offer the following hypothesis: H4:Advertisement scanpathsarestableacrosslevelsof (a) consumermotivationand(b) argument quality. Supportfor all four hypotheses would provide strongevidence for scanpaththeory and detailed insight into processes of attention wear-out across repeated exposures to print advertising. It would imply that, though attentionduration decreases across ad repetitions(Hi) and though differences in consumer motivation lead to differences in attentionduration (H3), scanpaths exhibit a first-order, reversible Markov process (H2), which is stable across consumermotivation and argumentqualityconditions (H4) and across repeated exposures (H2). THE VISUALATTENTIONMODEL To examine the visual attentionprocess and test the hypotheses, we propose submodels for attentiondurationand attentionscanpaths.We describe the submodels, along with the nested model tests employed to test the hypotheses. At the end of this section, we summarizehow each of the hypotheses is tested specifically. GammaModelfor AttentionDuration Rosbergen, Pieters, and Wedel (1997) recently showed thatgaze duration,defined as the numberof seconds subject i attendsto ad elementj duringexposurer, or dijr,can be described adequatelyby a gammadistribution: (1)

f(dijr) =

-(3f-d"ij >ir

exp

where Tcrepresentsthe expected numberof seconds subjects attendto ad elementj duringexposure r, and X representsa dispersion parameter.The gamma distribution is derived from the assumptionthat an individual fixation has a constant terminationhazard,so that the durationof individual fixations is exponential.The attentiondurationfor a particular element is the sum of several fixations, each exponentially distributed,which leads to a gamma distribution(cf. Rosbergen, Pieters, and Wedel 1997). We use the gamma model to investigatedifferences in attentiondurationacross exposures. In addition,the expected attentionduration,Tr,is modeled as a function of motivation or argumentquality. More specifically, we model ln(Tj)using a linear formulation [ERMQ],where E standsfor the elements in the advertisement, R for repetitionsof the advertisement,M for motivation of the consumer,and Q for quality of the arguments (for the use of the log-linear model notation, see Agresti 1990), and we estimate specific nested versions of the full model [ERMQ].For example, the model [E,R] involves only main effects of elements and repetitions. Repeated exposures.The gamma model is used in particular to test H1, that subsequent repetitions result in a decreaseddurationof attention.This hypothesisis tested by in-

vestigating whetheradding the effect of repetitions,R, to a model that only accounts for differences in advertisement elements, E, denoted by [E,R], leads to a significant improvementin fit in terms of the log-likelihood. Motivationand argumentquality.In addition,the gamma model is used to test the effects of motivationand argument quality on attentiondurationby examining the contribution of M (H3a)and Q (H3b)to model fit. Specifically, the models [E,R,M] and [E,R,Q] are compared with [E,R] using likelihood ratio tests. HeterogeneousMarkovModelfor AttentionScanpaths We representthe consumer'sattentionscanpathfor an advertisementas a Markovchain on a directedgraph,in which vertices representeye fixations on ad elements and edges representsaccades between those elements. We describethe attention scanpath by a stochastic model that involves the probabilitiesthat the consumer's eyes move from ad element j to ad element k, Pklj. We assume that scanpaths across ad elements can be describedby a first-orderMarkov process. That is, each eye fixation only depends on the previous one, and for each subject i and each exposure r, prkIj, the conditional probabilitythat the consumer's eyes move from ad element j to ad element k is as follows: (2)

jk

--

pr

K

Pklj

k=l

where mjkdenotes the expected numberof saccades fromj to k during exposure r. Equation 2 is consistent with a Poisson process that produces the inter- and intraelement saccades with expectation m[k(Lindsey 1995). Rosbergen, Pieters, and Wedel (1997) demonstratethat research on visual attention should take heterogeneity between subjectsinto account.In this study,we accountfor intersubjectvariability in the expected number of saccades, m5k,by making the standardassumptionthat m~k is a random variablethat follows a gamma distributionwith mean Ark and dispersion parameterv (McCullagh and Nelder 1989). In that case, the observed saccade frequencies,yijkr, follow a negative binomial distribution(MarkovNBD): 0)

(3)

f(Yijk,) =

f(, -"'I

^

r(V)

*

+

)v

V

'~Yijkr

+

The model assumes that differencesbetween consumersare randomand can be representedby a gammadistribution,but we examine the presence of structuraldifferencesacross the motivationor argumentquality conditions as well. To test the hypotheses, we formulatethe log-expectation of the Markov NBD, ln(grjk), as nested versions of the linear model [ODRMQ],where 0 denotes the origin (rows, ad elements) in the transition matrix, D the destination (columns, ad elements) in the transitionmatrix, and R, M, and Q are defined as previously.The hypotheses regarding the patternof fixations and saccades are tested by comparing models thatimpose differentnested structures,as we explain subsequently. Repeated exposures. Similar to the gamma model, Hi, which predictsan attentiondecrementacrossrepeatedexpo-

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

428

JOURNAL OF MARKETINGRESEARCH, NOVEMBER 1999

sures, is tested by comparing[O,D,R] with [O,D]. This tests whether the decrease in the number of fixations that consumers devote to the advertisementand its elements across repetitionsis significant. First-orderdependence. To test the three possible stochastic processes that may underlie visual scanning of an advertisement,we define three differentmodels: [-] for the completely randomprocess, [O,D] for the stratifiedrandom process, and [OD] for the statistically dependent process. We test H2afor first-orderdependence of the scanpathby examining the relativefit of [OD] versus [O,D] and use [-], the randomprocess, as the benchmarkmodel. Reversibility. H2b postulates that the scanpath is reversible; that is, the numberof saccades from ad element j to k, mjk= rkIjPj,is equal to the numberof saccades from k to j, mj = prj Ikpk(Ross 1996), where pJis the proportion of eye fixationsdirectedat ad elementj. Reversibilityof the scanpathcan be tested by comparingthe reversibleMarkov process characterized by a quasi-symmetric structure, [O,D,S], where S representssymmetry,with an unrestricted Markovprocess, [OD]. Here, [O,D,S] is a model that, apart from the distortingeffects of the marginal proportionsdefined by 0 and D, restricts the expectation of ad element (j,k) in the transitionmatrixto be equal to thatof ad element (k, j) for all k andj. The term S has as many levels as there are pairedcombinations(jk) and (kj) wherej is unequalto k and specifies each symmetric pair to have the same level (Lindsey 1995). Accounting for the effect of ad repetitions, H2b is supportedif the difference in the log-likelihood of [O,D,S,R] and [OD,R] is insignificant. Stationarity.H2cpredicts that the scanpathof consumers who are exposed repeatedlyto the same advertisementremains constant across repetitions. That is, the hypothesis predictsthat the Markov chain is stationaryand that transition probabilitiesare constantacross repetitions.Such a stationary Markov process implies that the interactioneffects OR and ODR are simultaneously zero (Lindsey 1995). Thus, we test H2cfor quasi-symmetryby comparingthe loglikelihood of the model [OR,D,SR] and that of [O,D,S,R]. Consumer motivation and argument quality. Similar to the gamma model, the Markovmodel is used to test the effects of consumermotivationand argumentquality on fixation frequency(H3), by comparingthe log-likelihood of the model [O,D,S,R,M] with, respectively, [O,D,S,R,Q] and [O,D,S,R]. In addition, H4, which predicts that individuals essentially exhibit the same statistical structureof visual scanning and thus have identical scanpaths,is investigated by testing the robustnessof the scanpathacross conditions in which levels of consumermotivationor argumentquality vary.For the scanpathto be stationaryacross consumermotivation and argumentquality conditions, the interactiveeffect of M and Q with S should not be significant.Thus, we test H4 for heterogeneity across advertisement and consumer conditions by comparing the log-likelihood of models [OM,D,SM,R]and [OQ,D,SQ,R]with thatof [O,D,S,R]. To summarize, we test Hl to H4 by comparing the following nested models: Hi: [E,R] versus [E] for gaze duration,and [O,D,R]versus [O,D]forscanpaths; H2a:[OD,R]versus[O,D,R]versus[-] forscanpaths; H2b:[O,D,S,R]versus[OD,R]for scanpaths; H2c:[OR,D,SR]versus[O,D,S,R]forscanpaths;

H3a:[E,R,M]versus[E,R]for gaze duration,and[O,D,S,R,M] versus[O,D,S,R]for scanpaths; H3b:[E,R,Q]versus[E,R]for gaze duration,and [O,D,S,R,Q] versus[O,D,S,R]forscanpaths; versus[O,D,S,R]for scanpaths; and H4a:[OM,D,SM,R] versus[O,D,S,R]for scanpaths. H4b:[OQ,D,SQ,R] Model Estimationand Selection All models are estimated with the method of maximum likelihood using computer programs written in Gauss (Aptech 1992). The optimal model specifications for the gamma and the Markov NBD models are determined by comparing nested models on the basis of the consistent Akaike informationcriterion(CAIC; Bozdogan 1987), and likelihood ratio (LR) tests. Because of the relatively large numberof tests performedacross the two studies, we report results at a significance level of p < .01. DESIGNOF THESTUDIES Two studieswere conducted.In the first study,consumers' motivationto attendto a printadvertisementfor shampooand the quality of argumentsin the advertisementwere manipulated systematicallyto test the hypotheses and establish the generalizabilityof repetition effects across consumer and stimulusconditions.Subjectswere presentedwith the advertisementson slides projectedon a screen in a predetermined order.Although Study I yields high internalvalidity and is well suited to test our hypotheses, the experimentaldesign and the particularprintadvertisementused may limit the externalvalidity of the results. One specific advertisementfor one brandof shampoo was used, subjects were not exposed to the advertisementin its naturalcontext (i.e., in a magazine), only a single sequenceof advertisementswas used, and differencesin consumermotivationwere manipulated.A second study was performedto extend and complementthe first study in several importantways. Study 2 uses threedifferent advertisementsfor grooming and personal care products,a largersample of subjects,and a more naturalexposure context (an electronicmagazine)with controlledorderof ad exposure and naturallyoccurringdifferences in consumermotivation,as we explain subsequently. Study I Sample, stimuli, and procedure.Sixty-eight randomlyselected consumers, 19 to 52 years of age, who wore neither glasses nor contact lenses, were invited to come to the office of the marketresearchcompany that conductedthe experiment. The experiment lasted approximately half an hour, and subjects were paid the equivalent of $20 for their participation.

On entering the experimentalroom, subjects received a booklet containing instructions regarding the experiment. They were instructedto watch carefully a series of slides of "draft"versionsof printadvertisements.The advertisements promoted eight different products. Half the subjects were told that the study's purposewas to gain insight into the impact of information within advertisements on judgments aboutthe productsadvertisedand were promiseda choice of shampoo at the end of the session (high motivation condition; for similar instructions, see Petty, Cacioppo, and Schumann 1983). The other half were told that the purpose was to develop a new method for testing "draft"versions of

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

429

Scanpath Theory advertisementsand that they were to evaluate the advertisements (low motivation condition). These instructionswere aimed to stimulate differences in situational involvement (cf. Miniard et al. 1991; Rosbergen, Pieters, and Wedel 1997), or what Celsi and Olson (1988) call situational sources of personalrelevance. The study's target advertisement,an advertisementfor a new brandof shampoo, Aquavital,appearedin the second, fourth,andninthpositions.Two versionsof the advertisement were designed specially by an advertisingagency. Both versions containedfour elements:a headline,a pictorial,a packshot, and body text with five textualargumentsin favorof the product.The argumentswere eitherstrong(e.g., "Thesea extracts in Aquavital provide naturalnutrientsessential to the strengthand vitalityof your hair")or weak (e.g., "Itis suited to everyone's hair"),and the contentof the headlinewas adjusted to the type of argumentsused.Argumentselection was basedon the resultsof a pilot study.The combinationof headline and argumentswas tested on its persuasive force. We show the targetadvertisementfor the weak argumentcondition at the top left cornerof Figure I (advertisementA). Subjects were seated in front of a screen, on which the slides were projectedfrom the back, and they were instructed to place their chin on a small chin rest. Eye positions were recorded50 times per second by an infraredcameralocated at the subject's left side, such as not to interferewith normal viewing behavior (Young and Sheena 1975). The camera was trainedon the subject's right eye, and subjects performeda calibrationtask. Before they were exposed to the 13 advertisements,subjects were instructedto press a button in front of them to go throughthe advertisementsat their own pace, but advertisementswere shown to the subjects for 20 seconds at most. After attendingto the advertisements,subjectscompleted a questionnairecontaining questions about their motivation to process the advertisementand the perceivedqualityof the arguments in it. Motivation to process the advertisement was measured by asking subjects to rate, on a seven-point scale ("completelyagree"/"completelydisagree"),theirmotivation to evaluate the arguments listed in the advertisement. Argument quality was measured by having subjects rate the arguments on three seven-point items. All items ranged from +3 (highest) to -3 (lowest), and scores on the three items were averaged (coefficient alpha = .91). Manipulationand reliabilitychecks.Analysis of variance shows that subjects in the high motivation condition score, on average,approximatelyone point higheron motivationto evaluate the argumentslisted in the shampoo advertisement than subjects in the low motivationcondition (F164 = 4.22, p < .05). Manipulationof argumentquality was successful as well, because strong argumentswere perceived as stronger than weak arguments(.59 versus-.32; F1,64= 7.56, p < .05). Because of factors such as excessive blinking of the eye or tear fluid in the eye, reliabilityof the eye movementdata of 16 subjects was insufficient to include them in further analyses. The remaining 52 subjects were divided among the four conditions in approximatelyequal numbers. Study 2 Sample, stimuli, and procedure. One hundred eighteen randomly selected consumers participated in the second study.Twenty-threepercentwas between 16 and 25 years of

age, 26% was between 26 and 35 years, 30% was between 36 and 45 years, and 21% was 46 years of age or older. Fifty-five men and 63 women participated.The purpose of this study was to provide a more naturalexposure situation than in the first experiment, using, in addition, more advanced presentationand eye-tracking instrumentation,examining multiple advertisements,and examining naturally occurringdifferences in consumermotivation. On entering,subjectsfirstcompleteda brief questionnaire assessing sociodemographics and various opinions. The questionnairecontained items about the relevance to consumers of 28 productand service categories. Two items assessed the relevance to consumersof grooming and personal care products: "To me personal care products and cosmetics are..." and "To me clothing and textile goods are...,"both accompaniedby five-point responsescales from "very important"to "very unimportant."Scores on the two items were averaged (coefficient alpha = .70). Four consumer motivation groups ("very low" to "very high") were constructedby categorizingthe scores into four quantiles,so that mean motivation scores differed strongly among the four groups. This procedureaims to tap naturaldifferences in enduringinvolvement (cf. Rosbergen,Pieters, and Wedel 1997), or what Celsi and Olson (1988) call intrinsicsources of personalrelevance. Next, subjects participated in the eye-tracking study. They were seated in front of a 20-inch, touch-sensitive liquid crystal display monitor,on which a computerizedmagazine appeared.The monitor was built into a table, slightly tilted, to mimic the position in which consumers regularly read magazines. Subjectscould page throughthe magazine at theirown pace by touchingthe lower right(page forward) or left (page backward)of the screen. The magazine contained 50 double pages, both editorialmaterialand new and familiaradvertisements.Three advertisementsfor grooming and personalcare productswere insertedthree times. They pertainedto new advertisementsfor a new brandof toothpaste ("Fluocaril"),an existing brandof personalcare products ("Nivea"), and an existing brand of soap products ("Sunlight").All advertisementscontaineda headline, body text, packshot, and pictorial, as did the advertisement in Study 1. The advertisementswere inserted three times, always with differenteditorialmaterialfacing it. Orderof the advertisementswas randomizedacross subjects,with the restrictions that the targetadvertisementsnever came in first or last place (to rule out primacy and recency effects) and that two consecutive exposures to the target advertisement were interspersedby at least three unrelateddouble pages. Subjects were instructedto page freely through the magazines at their own pace, as if at home. The three target advertisementsin Study 2 are presentedin the top rightcorner and at the bottom of Figure I (advertisementsB-D). While subjects paged through the magazines, their eye movements were recorded with infraredscleral reflection eye-tracking methodology, similar to but more advanced than that used in Study 1. This particular eye-tracking methodology allows subjects to move their heads freely (within a virtualbox of approximately30 centimeters)while paging throughthe electronicmagazineas camerastrackthe position of both the head and the eye and continuously match the information.The position of the fovea (yellow spot) is measuredat 50 hertz.Subjectswith glasses received

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

JOURNALOF MARKETING RESEARCH,NOVEMBER1999

430

Figure1 ANDTHEPREDICTED ATTENTION DISTRIBUTION ACROSSELEMENTS THEFOURPRINTADVERTISEMENTS

A: Aquavital

:A A a-leis

J.9

B:Fluocaril

XXAfl :JAIA .X.., I d. oet woor j e Iaaar ............

// /o///111 /////7 ////41/ %% / ^,/-

IL

24 W ... - -/ 10 ^Illil

17 I

//[/llii---I

I

26

48 I*11111I D: Sunliaht

C: Nivea

.1 ....

''//y/ ////

20

-111,11-1.

_

///////1. .

.-_-.

---

----

111111

Q8

47 .I

25

L/I

2

--. -

/7/ 7ll[l

-

. . . . _I_

20

11

47

//////11111 c"~~~'''''

Bodytext

10

16

Headline

.

Packshot

.

,

.

,,

.

.

.

3

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

.

.

..

'

Y

Y-

Y

Y

Pictorial

_

22

L

431

Scanpath Theory special lightweight replacementglasses that allow valid infraredeye tracking.Subjects indicatedin the debriefingthat they generallyfelt at ease (94%) and relaxed(97.5%) during the study. RESULTS Study 1 Attention duration. The average attention duration in Study 1 was 1.45 seconds for the headline, 1.25 seconds for the pictorial,4.25 seconds for the text, and 1.04 seconds for the packshot.In Table 1, we presentthe results of the model tests, and in Table 2, we presentthe parameterestimates for the best models for attentionduration. Inspection of Table 1 shows that, for Study 1, model [E,R] provides the best representationof the data because it yields the lowest value of the CAIC statistic. This implies that attentiondurationdiffers significantly across elements of the advertisementand across repeatedexposures.Table 1 shows that neitherconsumermotivationnor argumentquality affects the amount of attentionpaid to the ad elements, because neither model [E,R,M] nor model [E,R,Q] represents the data betterthan model [E,R] does. Thus, H3ais not supported,but H3bis. Parameterestimates for the best model [E,R] in Study 1 are displayed in Table 2. They reveal that attentionduration is longest for the body text, followed by the headline, and shortest for the pictorial and the packshot. The percentage difference indicates the difference in gaze durationbetween the indicatedad element and the packshot(the reference ad element). For example, the attention durationto the body text is three times longer than the attentiondurationto the packshot. In addition, a progressivedecrease in the expected attentiondurationis observed across exposures, as indicated by the estimates of R(1) and R(2) in Table 1. Attention duration during the first exposure to the advertisement is close to one and a half times longer thanduringthe thirdexposure. This supportsH1. Scanpaths.The average numberof fixations on the ad elements was 5.76 for the headline, 3.79 for the pictorial, 15.74 for the body text, and 5.62 for the packshot.In Table 3, we present the results of the model tests, and in Table 4, we presentthe parameterestimates for attentionscanpaths. The left-hand part of Table 3 presents the results of the model selection procedurefor the scanpathmodels in Study 1. The significant improvementin fit from the first model [-] to the second [O,D] indicates that the selection of fixation points cannot be regarded as completely random. Adding a main effect for repetitionto the model significantly improves the model's fit ([O,D,R] versus [O,D]). This supports Hi, in that attentiondurationdecreases across repeated exposures. As indicated by the highly significant LR test for the model [OD,R] versus the model [O,D,R] and the associated drop in CAIC, there is strong dependence between successive fixations on ad elements. The overwhelming evidence for a first-orderMarkov process supportsH2a.Reversibility of this process, in turn, requiresthat the model fit does not decrease significantly when we impose a quasi-symmetric (the model [O,D,S,R]) instead of the unrestrictedMarkov structurein the model [OD,R] on the interactionof rows and columns of the transitionmatrices.In Table 1, the compari-

son of these two models shows that the quasi-symmetric model fits the data as well as the unrestrictedmodel and that CAIC is lowest for the quasi-symmetricmodel. This implies that the scanpathis reversible, which supportsH2b. The inclusion of an interactioneffect of ad repetitionsand the row and the quasi-symmetryterm [OR,D,SR] does not result in a significantly better fit to the data than the benchmarkmodel [O,D,S,R]. This means thatthe Markovchain is stationaryand the stochastic scanpath is stable across repeated exposures of the print advertisement,so H2c is supported. Thus, attention scanpaths in Study 1 obey a firstorder,reversible,stationaryMarkovmodel. Finally, neither motivation ([O,D,S,R,M] versus [O,D,S,R]) nor argument quality ([O,D,S,R,Q] versus [O,D,S,R]) significantly affects the numberof fixations. In addition,the scanpathis stationaryacross those conditions, as is shown by the nonsignificantimprovementof the models [OM,D,SM,R]and [OQ,D,SQ,R].Thus, H4 is supported completely, but H3 is supportedonly for argumentquality. Parameterestimates for the selected scanpathmodel are presentedin Table 4. The estimates and t-values show that subjects pay significantly more attention to the advertisement duringthe first exposure thanduringsubsequentexposures. Furthermore,body text, headline, packshot, and pictorialreceive a decreasingamountof attention,in thatorder. The distributionof the estimated volume of fixations across the four ad elements and across the three exposures is indicated in the barchartbelow the targetadvertisementin Figure 1. Inspectionof advertisementA shows that 41 % of the estimatednumberof fixations goes to the body text, 24% to the headline, 18%to the packshot,and 17%to the pictorial. The results of the Markovmodel for scanpathscorroborate those of the gamma model for attentionduration. As hypothesized, scanpaths of print advertisementscan be described by a stationary,reversible, first-orderMarkov process. Inspectionof Table 4 shows that (1) the amountof attention paid to the advertisementdecreases by approximately 50% from the first exposure to the third;(2) the majority of saccades, approximately75%, occur within ad elements, in particularin the body text; (3) most interelement saccades startfrom or end at the packshot;(4) the expected transitionmatrices are quasi-symmetric;and (5) the transition probabilitiesremainconstantacross exposures and motivation and argumentquality conditions. The steady state probabilitiesof the Markovchain indicate that the probability thatthe subject'seyes will move to a specific ad element ultimatelywill equal .19 for the headline, .13 for the pictorial, .50 for the body text, and . 18 for the packshot.In Figure 2, we offer the estimated scanpathsuperimposedon the advertisement.Arrows indicate the directions of saccades. Self-reflecting arrowsindicate saccades within ad elements. Inspection of Figure 2 indicates, for example, the higher likelihood of scanning from the packshot to the body text (.21) thanthe otherway around(.07). The high likelihood of saccades within the body text (.89) and the headline(.74), as comparedwith the pictorial(.55) and the packshot(.50), indicates reading with its associated multiple saccades in a designated space. Contrary to our expectations, differences in consumer motivationdid not translateinto differences in attentionduration, as indicated by the nonsignificanteffect of motivation in the gamma and NBD models. We expect that this re-

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

Table 1 MODELTESTS FOR GAZEDURATIONS Study2

Study 1 Model [-] [El [E,R] [E,R,M] [E,R,Q]

In(L) -1050.00 -957.39 -922.16 -920.40 -920.65

CAIC 2114.87 1951.96 1896.36** 1900.29 1900.79

Nivea

Fluocaril

Aquavital LR (df) 185.23* (3) 70.46* (2) 3.52 (1) 3.40 (1)

In(L)

CAIC

-210.60 -18.02 15.82 38.36

437.19 77.32 26.14 5.83** -

-

LR (df)

ln(L)

385.16* (3) 67.68* (2) 45.08* (3)

415.79 598.69 630.65 648.50

-

-

CAIC -815.07 -1156.10 -1203.50 -1214.45** -

LR (df) 365.80* (3) 63.92* (2) 35.70* (3) -

lIn

551 663 676 705 -

*LR test significantat p < .01. **Denotes the minimumCAIC value for that advertisement. Notes: CAIC = consistentAkaike informationcriterion,LR = likelihood ratio, and df = degrees of freedom.

Table 2 PARAMETER ESTIMATES FORGAZEDURATIONS Study2

Study1 Aquavital Model Intercept E(headline) E(pictorial) E(body text) R(l) R(2) M(l) M(2) M(3) Lambda

Estimate

t-valte

-.409 .311 .201 1.378 .861 .382 -

-4.22 2.68 1.75 11.88 8.52 3.82 -

-

-.039

-

-.83

Nivea

Fluocaril Percentage Difference -

36% 22% 297% 137% 47% -

Estimate

t-value

-1.795 1.474 -.690 .419 .752 .485 .567 .459 .030 -.658

-16.37 14.05 -6.57 3.99 8.26 5.33 5.33 4.35 .30 -21.16

Percentage Difference -

337% -50% 52% 112% 62% 76% 58% 3%

Estimate

t-value

-3.024 2.165 1.494 .843 .740 .288 .614 .216 .235 -.608

-27.44 21.16 14.65 8.18 8.31 3.25 5.83 2.10 2.37 -19.45

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

Percentage Difference -

771% 345% 132% 110% 33% 85% 24% 26%

Esti

-2.3 1 . -.1 . . . . . -.7

Table 3 MODELTESTS FOR SCANPATHS Study 1

Study 2 Fluocaril

Aquavital Model [-] [O,D] [O,D,R] [OD,R] [O,D,S,R] [OR,D,SR] [O,D,S,R,M] [OM,D,SM,R] [O,D,S,R,Q] [OQ,D,SQ,R]

In(L) -786.53 -767.17 -760.72 -645.96 -647.67 -644.83 -647.57 -643.43 -647.13 -644.13

LR (dJ)

CAIC 1585.57 1565.64 1565.24 1392.03 1376.69** 1483.63 1382.75 1430.77 1381.87 1432.18

19.36* (3) 6.45* (2) 229.52* (9) 3.42 (3) 5.68 (18) .20 (1) 8.28 (9) 1.08 (1) 6.00 (9)

ln(L) -727.56 -675.25 -666.85 -588.53 -592.05 -585.46 -590.00 -578.08

Nivea

CAIC

LR (df)

1467.63 1381.79 1377.50 1277.19 1265.45** 1364.91 1280.11 1425.21

104.62* (3) 16.80* (2) 156.64* (9) 7.04 (3) 13.18 (18) 4.10 (3) 23.84 (27)

-

-652.41 -578.53 -566.80 -470.14 -472.01 -463.12 -455.43 -437.20

-

-

1317.34 1188.34 1177.41 1040.40 1025.37 1120.23 1010.98** 1143.46 -

--

--

LR (df)

CAIC

ln(L)

147.76* (3) 23.46* (2) 193.32* (9) 3.74 (3) 17.78 (18) 33.16* (3) 36.46 (27) -

Table 4 PARAMETER FORSCANPATHS ESTIMATES Study 2

Intercept E(headline) E(pictorial) E(body text) R(l) R(2) M(I) M(2) M(3) S(headline,pictorial) S(headline,body text) S(headline,packshot) S(pictorial,body text) S(pictorial,packshot) S(body text, packshot) nu

Estimate

t-value

Estimate

.640 .241 -.092 .802 .624 .283

2.75 1.59 -.61 5.28 4.27 1.91

2.376 1.105 -.525 .487 .673 .495

-2.004 -2.192 -2.146 -1.319 -.673 -1.429 .611

.-

-1.637 -4.406 -1.487 -2.425 -1.137 -1.764 -1.953

Nivea

Fluocaril

Aquavital Model

Estimate

t-value

9.72 7.12 -3.15 3.12 4.48 3.27

.139 1.775 1.141 .965 .778 .268

-8.14 -9.71 -9.18 -5.41 -2.76 -6.14 5.67

.366 -1.213 -1.678 -1.244 -1.853 -1.274 -1.279 2.566

.61 15.75 9.86 8.26 8.78 2.88 5.42 1.82 3.46 -11.36 -13.95 -7.58 -12.25 -6.59 -6.26 9.84

t-value

--.562 -.193 --

-7.18 -15.35 -6.61 -10.50 -5.03 -7.98 -18.78

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

-639. -585. -579. -479. -484. -474. -471. -449. -

*LR test significantat p < .01. **Denotes minimumCAIC for that advertisement. Notes: CAIC = consistentAkaike informationcriterion,LR = likelihood ratio, and df= degrees of freedom.

Study I

ln(

434

JOURNAL OF MARKETINGRESEARCH, NOVEMBER 1999 Figure 2 CHARACTERIZED BY CONDITIONAL PRINTADVERTISEMENT TRANSITION PROBABILITIES (A) ANDITSSCANPATH (B)

doet ?AQUAVITIAL alles voor j e har

DC fe,g,odc .xtr-ict . .AQUAVrTAL sb,-mpoo cv jc t bw cten kkc acts. fi;t is zowdXo ecn

2f .I aIlseen 300mL.verpaddng c vcrkijg'c. iH-t Nim is 4iwoxm wrcci gcbirkSi wrtt, icda? dag AQtXAVr#TAV^ lte Ww4 woo je bw a is cci sowm.pv dije hbaamct Verdiwi zo doct Wzc

ks- ^, ?t>

sbwnpowckc dtg elm voorjs

f hwes.

suit is, at least partly,due to the overall high levels of attention in Study 1. Note that the average attentiondurationin Study 1 is significantly longer than what is reportedin attention research conducted in more naturalconditions (cf. Kroeber-Riel 1993; Rosbergen, Pieters, and Wedel 1997; Rossiter and Percy 1997, p. 295). Study 2 was designed to examine this issue in more detail. It studies attentionto print advertising in more naturalconditions, using naturallyoccurringdifferences in consumermotivation. Study2 Attentionduration.The average attentiondurationto the advertisementsin Study 2 across the three repetitions was 2.53 seconds for Fluocaril, 1.52 seconds for Nivea, and 1.42 seconds for Sunlight, which is much shorterthan in Study 1. The issue is whether the hypotheses are supported in the more naturalconditions in this study. The model tests are found in Table 1, the parameterestimates of the selected model in Table 2. In Study 2, model [E,R,M] provides the best representation of the datafor all threeadvertisements;the CAIC of this model is lowest. This implies that, for all three advertisements, attention durationdiffers between ad elements (E), attentiondurationdecreases significantlyover time (R), and significantdifferences between consumermotivationgroups are observed (M). These results supportHI and H3a.The at-

tention decrementfrom the first to the thirdexposure to the advertisementsis substantial:for Fluocaril from 3.49 seconds to 1.58 seconds, for Nivea from 2.11 seconds to 1.04 seconds, and for Sunlight from 1.67 seconds to 1.16 seconds. On average, the printadvertisementsreceive approximately 55% of the attention during their third exposure comparedwith theirfirst exposure. The attentiondecrement is similar to that in Study 1. Inspection of Table 2 further shows that, in all threeadvertisements,the headlinereceives most attention-in the case of the Nivea advertisement,almost eight times more than the packshot.The pictorialsof the Nivea and Sunlightadvertisementreceive more attention than their respective packshots, but the pictorial of the Fluocariladvertisementreceives approximately50% less attention than its packshot. Table 2 also reveals significant differences between consumer motivationgroups in attentionduration.For all three advertisements,attention duration from the lowest to the highest consumer motivation group increases significantly. Compared with the lowest motivation group, the highest motivation group pays 76% more attentionto the Fluocaril advertisement,85% more attentionto the Nivea advertisement, and 110% more attention to the Sunlight advertisement. Finding a significant attention decrement across ad repetitions,significant differences between consumermotivation groups,and substantialheterogeneityin attentiondu-

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

435

Scanpath Theory rationacross the threeadvertisementswould make universal supportfor the scanpathhypotheses even more telling. That issue is addressednext. Scanpaths. In Table 3, we present the results of the tests for the scanpathmodels for Study 2. As in Study 1, there is a significant improvementin fit from the first model [-] to the second [O,D] for all threeprintadvertisements.This provides strong supportagainst a completely randomselection of fixation points. As indicated by the significant improvement of the model, including the repetitioneffect ([O,D,R] versus [O,D]), the numberof fixations on each of the three advertisementsdrops significantly across ad repetitions,as in Study 1. This supportsHI and corroboratesthe findings of the gamma model for attentionduration. As indicated by the highly significant LR test for the model [OD,R] versus the model [O,D,R] and the associated drop in CAIC, there is clear evidence for a first-order Markov process in attention scanpaths for all three advertisements.This supportsH2a.Reversibilityof the scanpathis supportedas well, because for all three advertisements,the model fit does not decrease significantly when we impose a quasi-symmetricinsteadof an unrestrictedMarkovstructure (model [O,D,S,R] versus model [OD,R]). The CAIC of the quasi-symmetricmodel is betterfor each of the threeadvertisements. This implies that the scanpath is reversible and that Study 2, as did Study 1, supportsH2b. The interactionof ad repetitions with the row effect and with the quasi-symmetric Markov structure (model [O,D,S,R] versus model [OR,D,SR]) does not lead to a better fit to the data for each of the three advertisements,as is indicatedby an increase in the CAIC and nonsignificantLR tests. This shows that the stochastic scanpath is stationary across exposures, so that, as in Study 1, H2cis supportedfor all three advertisements. Motivation significantly affects the fixation frequencies for two of the three advertisements (Nivea and Sunlight, [O,D,S,R,M] versus [O,D,S,R]), though this is not the case for Fluocaril, for which the CAIC is lowest for the model [O,D,S,R]. This diverges from the findings for gaze duration for that brand,but we conclude that there is some support for H3a. As predicted, the interactionbetween consumer motivation and the quasi-symmetricMarkov structureis insignificant for all three advertisements ([O,D,S,R,M] versus [OM,D,SM,R]), as is indicated by the CAIC, which is lowest for the formermodel. This shows that the attentionscanpath is stationaryacross consumermotivationgroups for all three advertisements,which supportsH4. The models with the lowest CAIC are selected for further inspection: model [O,D,S,R] for Fluocaril and model [O,D,S,R,M] for Nivea and Sunlight. Inspection of the parameterestimates in Table 4 shows that (1) the headline receives most attention, followed by the pictorial for two of the three advertisements(Nivea and Sunlight), and by the body text for the other (Fluocaril); (2) most interelement saccades startfrom or end at the headline;(3) the transition matricesare quasi-symmetric;and (4) the scanpathremains unchangedacross repeatedexposuresand consumermotivation groups. In both studies, there is strong evidence of a scanpaththatis stable across repeatedexposuresto the same advertisementand across degrees of consumerinvolvement. These results were obtained across advertisementswith dis-

tinct scanpathsand attentionduration,which supportstheir credibility.In Figure 1, advertisementsB-D, the distribution of estimated fixations across the ad elements is indicatedin the bar chartsbelow the targetadvertisements.For example, 47% of the estimated fixations on the Nivea advertisement go to the headline, 25% to the pictorial, 20% to the body text, and 8% to the packshot. DISCUSSION To our knowledge, this is the first investigation in which attentiondurationand attentionscanpathsto printadvertising are examinedjointly and in detail. We propose a gamma regression model for attentiondurationand a MarkovNBD model for attention scanpaths.The duration and scanpath models were applied in two studies to examine consumers' visual attentionto repeatedadvertisementsusing eye-tracking methodology. The results providestrongsupportfor predictionsderived from scanpath theory. In supportof H1, we found in both studies that repetitionreducedthe amount of attentionpaid to the advertisementand its elements by approximately50% from the first to the thirdexposure. In supportof H2, repeatedexposure to advertisementsdid not change the order in which ad elements were scanned in either Study I or Study 2. In particular,as predicted by scanpath theory, we established the existence of attention scanpathsthat can be describedadequatelyby a stationary, reversible, first-orderMarkov process. The stability of the stochastic scanpathacross repeatedexposures is particularly striking,because it indicatesthat the patternaccordingto which consumers scan print advertisementsis determined largely during the first exposure and is difficult to change duringsubsequentexposuresto the same advertisement.We do not claim that scanpaths are deterministically identical across consumers and ad repetitions. Our Markov NBD model for the scanpaths is a stochastic model that allows randomvariationsin the saccade patternacross subjectsand exposures. We allow for heterogeneity in the transition probabilitiesacross subjects through a gamma distribution of those probabilities.That distributiondeals with random variationof the scanpathacross consumers.Thus, both studies show that stable scanpathsexist across subsequentexposures to advertisements,though they take randomvariation of the scanpathsacross individualsinto account. In supportof H3, differencesin argumentquality in Study 1 did not lead to differences in attention duration.The results presentedfurtherindicate that the quality of the arguments in the advertisementdoes not postpone satiation, because argument quality does not moderate the effect of repetition on any aspect of visual attention. It appears that when consumers believe they "know"or have seen the advertisementbefore, subsequentexposures stimulateidentification responses, which lead to reductionsin the durationof attention, as is conjectured by Krugman(1972). Although variationsin advertisementexecutions may raise attentionto the levels attainedduringthe firstexposure (Haugtvedtet al. 1994), strong argumentsin our researchdid not keep attention at that level. Counterto H3,differences in consumermotivationhad no effect on attentiondurationor fixation frequencyin Study 1, but they did have strong effects on attentiondurationfor all three advertisementsin Study 2, though the effect on fixa-

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

436

JOURNAL OF MARKETINGRESEARCH, NOVEMBER 1999

tion frequency was significant for two of three advertisements only. We do not expect that the absence of a motivation effect on attentiondurationin Study I is due to an insufficiently strong motivation manipulation, because manipulationchecks were statisticallysignificant.It is more likely that the absence of such an effect in Study I is due to a ceiling effect on attentionduration.The average attention durationacross motivationconditions and ad repetitionsin Study 1 was almost 8 seconds, which is significantlyhigher than the 1.82 seconds for the three advertisementsin Study 2 and higher than what is usually reportedin attentionresearch in naturalconditions(cf. Kroeber-Riel1993). We expect that the data collection procedurein Study I and the fact that a new advertisementfor a new brand was used raised consumermotivation from high to very high instead of from low to high. Withsuch extremely high levels of motivation, differences in attentionduration apparentlycould not be identified.In Study 2, in more naturalexposure conditions, the hypothesizedmotivationeffect on attentiondurationconsistentlyemergedfor all threeadvertisements,and it emerged on fixation frequencyfor the two advertisements for existing brands(Nivea and Sunlight).In a setting akin to our Study 2, Rosbergen,Pieters, and Wedel (1997) find significantdifferencesin attentiondurationbetween consumers who differed in motivation for an existing brandof shampoo. In summary,our results show that naturallyoccurring differences in intrinsic sources of consumer motivation (Celsi and Olson 1988) affect the durationof visual attention to printadvertisements.The results suggest that differences in attentiondurationdue to consumermotivationmay be attenuated quickly by increased levels of overall attention, which are stimulatedby characteristicsof the exposuresituation and the stimuli used. Finally, in supportof H4, neither consumer motivation nor argumentquality affected attention scanpaths. Our results attest to the importanceof attentionin understanding the mechanisms underlyingadvertising effectiveness and wear-out. The finding that attention duration dropped by nearly 50% over three repeated exposures (while the scanpathitself remainedstable) indicates that, if consumers control exposure duration themselves, such as with print or electronic advertising,they may adapt exposure durationto limit satiation.The results show how quickly attentionwear-outfor printadvertisingcan take place. To date, most repetition research has focused on acceptance wear-out. Such researchtypically has been conducted with fixed and ratherlong exposure durationsto the advertisements. In such conditions, satiationmay set in rapidly,and consumers'attitudetowardthe advertisementmay drop significantly across only a few repetitions.Although we did not measure ad liking in this study and despite our finding that attentiondurationdrops consistently, even when the initial exposure durationis alreadyvery low (a few seconds only), we warn against concluding prematurelythat advertising repetitionshave fast and strongnegativeeffects on ad liking. The first instrumentthat consumers can use to manage repeated exposure to advertisementsis reducing the duration of their attention(see also Chatterjee,Hoffman, and Novak 1998). When consumers can control their exposure to advertising through paging, zapping, zipping, or clicking a mouse button, acceptance wear-outmay be more an exception and attentionwear-outthe rule.

A potential limitation of this research is the relatively short intervals (i.e., less than an hour) between ad exposures, which may have led to sharperattentiondurationdecrease and higher stabilityof the attentionscanpathsthan in real settings. However, we confirmedthe existence of stable scanpathsin two studies with differentdesign and different length of the intervalsbetweenexposures.Moreover,our results are directly relevantfor impact scheduling, that is, repeating the same advertisementmultiple times within the same print issue or commercial block, which is a common media strategy.In addition,our researchsetting is a typical pretest situation to examine the quality of a new advertisement and the impact of repeatingthe advertisementseveral times within a short time frame, and interexposureintervals are similar to those used in previous studies on advertising repetition(Cacioppoand Petty 1979, 1980; Ray and Sawyer 1971; Schumann,Petty, and Clemons 1990). Moreover,basic researchon scanpaththeoryfinds similarlystable effects across exposure durationsof up to a week (Stark 1994). In summary,strong evidence for attention wear-out was found, as well as significant differences between consumerswho differed in motivation.Strong supportalso was found for the existence of attention scanpathsthat are stable across advertising repetitions and consumer and message characteristics. FurtherResearch Replication of this study should use advertisementsfor differentproductsplaced in several advertisingmedia; television and electronic advertisementsare particularlyinteresting because of their inherentcomplexity. Several of our hypotheses may carry directly over to those media, but we leave it for additionalresearchto provide the theoreticalunderpinningsand empiricalsupport. Furtherresearchmight examine the effect of media planning tactics, such as bursting,and the effect of interexposure intervals on attention wear-out. Although multiple exposures within a short duration in different media vehicles (bursting)also may lead to substantialdrops in attention,attention graduallymay returnclose to its original level after the burst.Additional research that examines the impact of actual versus experienced exposure frequencyon visual attention to advertising appears particularlyworthwhile.Attention wear-out may be more due to self-reportedthan to actual prior exposure frequency,and perhapsthey have independenteffects. Another potential avenue for furtherresearchpertainsto the determinantsof attentionscanpaths.Attentionscanpaths are likely to be the joint outcome of (1) consumers' goals and the cognitive schemata they have about the dominant architectureof print advertisementsin general in a specific medium or for a specific product/brandand (2) specific ad characteristicsthatattractand guide attention,such as color, contrast,letter and picturesize, and so forth. Suggestive evidence for the former indicates that people almost instantaneously form spatialmodels of stimuli, consideredprimitive precognitive models, and that they use these mental representationsto scan new stimuli. When the mentalrepresentation of the stimuluschanges, the attentionscanpathchanges (Starkand Ellis 1981). The result that minimal differences in scanpaths were presentacross subjectcharacteristics,whereasthe ad layout

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

437

Scanpath Theory (headline, pictorial, packshot, body text) exerted strong effects, suggests that data- or stimulus-drivenprocesses are importantand perhapsdominantin the scanninginstructions in our research.To date, virtuallyno systematicresearchhas been performedon the interplaybetween advertisementand consumer characteristicsin attentionprocesses, despite the key importanceascribedto it in advertisingresearch(Batra, Myers, and Aaker 1996; Kroeber-Riel 1993; Rossiter and Percy 1997). In Study 1, the body text received the most attention, followed by the headline. In Study 2, the headline received the most attentionfor each of the three advertisements, and the body text received much less attention.As yet, we cannot determine whether the difference between studies in attention to the body text and headline is due to differences in ad layout, differences in exposure conditions, or other factors. However, compared with Study 1, the attention durationin Study 2 was substantiallyshorter,and recent researchsuggests thatwhen consumersspend less time on commercial stimuli, attentionto the textual information suffers most (Pieters and Warlop1999). The resultsof Study 2 appearmore typical and converge with printcopy principles that emphasize that "Since most people reading print ads never go beyond the headline, it is also extremely importantthat the headline and visual complementeach other so well and 'tell the story' so easily, that a readerwho only looks at the headline and main visual can 'get the message' without having to read a word of the body copy" (Batra, Myers, and Aaker 1996, p. 425). Furtherresearch might build on our findings and on the model of visual attention that we developed here. Research in which architecturesof advertisements(i.e., size and location of headline, pictorial, body text, and packshot;cf. Rosbergen, Pieters, and Wedel 1997) and conditions of consumersare varied systematically will provide insight into the relative importanceof person- versus stimulus-specific determinantsof attentiondurationand scanpaths. A final importantquestion is whether certain scanpaths contributemore to advertisingeffectiveness thanothers.Assuming equal attentionduration,does it matterhow, for example, a consumer scans two differentadvertisements?According to scanpath theory, the scanpath facilitates subsequentrecognition of advertisementsand thus of advertised brands. Perhaps some scanpathseven stimulate message learning and attitudeformationmore than others. Additional research that examines both regularities in eye movements and the effects of such regularitieson subjects' cognitive responses, memory,and attitudesappearsparticularly worthwhile. REFERENCES Agresti, Alan (1990), Categorical Data Analysis. New York:John Wiley & Sons. Anand, Punam and Brian Sternthal(1990), "Easeof Message Processing as a Moderatorof Repetition Effects in Advertising," Journal of MarketingResearch,27 (August), 345-53. Aptech (1992), Gauss ApplicationsManual:MaximumLikelihood. Maple Valley: Aptech, Systems Inc. Axelrod, Joel N. (1980), "AdvertisingWear-Out,"Journal of Advertising Research, 20 (October), 13-18. Batra,Rajeev, John G. Myers, and David A. Aaker (1996), Advertising Management, 5th ed. Upper Saddle River, NJ: Prentice Hall. and Michael L. Ray (1986), "SituationalEffects of Advertising Repetition: The Moderating Influence of Motivation,

Ability, and Opportunityto Respond,"Journalof ConsumerResearch, 12 (March),432-45. Belch, George E. (1982), "The Effects of Television Commercial Repetition on Cognitive Response and Message Acceptance," Journal of ConsumerResearch, 9 (June),56-65. Berger,Ida E. and AndrewA. Mitchell (1989), "The Effect of Advertisingon AttitudeAccessibility, AttitudeConfidence, and the Attitude-Behavior Relationship," Journal of Consumer Research, 16 (December), 269-79. Bozdogan, H. (1987), "Model Selection and Akaike's Information Criterion(AIC): The GeneralTheory and Its Analytical Extensions," Psychometrika,52, 345-70. Cacioppo, John T. and RichardE. Petty (1979), "Effects of Message Repetition and Position on Cognitive Responses, Recall, and Persuasion,"Journal of Personalityand Social Psychology, 37, 97-109. and (1980), "Persuasivenessof Communicationsis Affected by ExposureFrequencyand Message Quality:A TheoreticalandEmpiricalAnalysis of PersistingAttitudeChange,"in CurrentIssues and Researchin Advertising,James H. Leigh and Claude R. Martin,eds. Ann Arbor,MI: Universityof Michigan, 97-122. Calder,Bobby J. and BrianStemthal (1980), 'Television Commercial Wear-Out:An InformationProcessing View," Journal of MarketingResearch, 17 (May), 173-86. Celsi, Richard L. and Jerry C. Olson (1988), "The Role of Involvement in Attentionand ComprehensionProcesses,"Journal of ConsumerResearch, 15 (September),210-24. Chatterjee, Patrali, Donna L. Hoffman, and Thomas P. Novak (1998), "Modelingthe Clickstream:Implicationsfor Web-Based AdvertisingEfforts,"Project2000 Manuscripts,VanderbiltUniversity, Nashville (available at URL: http://www2000.ogsm. vanderbilt.edu/papers/clickstream/clickstream.html). Cox, Dena S. and Anthony D. Cox (1988), "WhatDoes Familiarity Breed? Complexity as a Moderatorof Repetition Effects in AdvertisementEvaluation,"Journal of ConsumerResearch, 15 (June), 111-16. Craig, C. Samuel, BrianStemthal, and ClarkLeavitt (1976), "Advertising Wear-Out:An Experimental Analysis," Journal of MarketingResearch, 13 (November), 365-72. Ellis, StephenR. and James DarrellSmith (1985), "Patternsof Statistical Dependencyin Visual Scanning,"in Eye Movementsand Human InformationProcessing, Rudolf Groneret al., eds. Amsterdam:Elsevier Science PublishersBV, 221-38. Greenwald,Anthony G. and Clark Leavitt (1984), "Audience Involvement in Advertising:Four Levels," Journal of Consumer Research, 11 (June), 581-92. Groner,Rudolf (1988), "Eye Movements,Attentionand Visual InformationProcessing:Some ExperimentalResults and Methodological Considerations,"in Eye MovementResearch: Physiological and Psychological Aspects, Gerd Liier et al., eds. Toronto:C.J. Hogrefe, 295-319. and ChristineMenz (1985), 'The Effect of StimulusCharacteristics, Task Requirements and Individual Differences on Scanning Patterns,"in Eye Movementsand Human Information Processing, Rudolf Groneret al., eds. Amsterdam:Elsevier Science PublishersBV, 239-50. Haugtvedt,CurtisP., David W. Schumann,WendyL. Schneier,and WendyL. Warren(1994), "AdvertisingRepetitionand Variation Strategies: Implications for UnderstandingAttitude Strength," Journal of ConsumerResearch,21 (June), 176-89. Janiszewski, Chris (1993), "PreattentiveMere Exposure Effects," Journal of ConsumerResearch, 20 (December), 376-92. Kroeber-Riel,Werner(1993), Bild Kommunikation:Imagerystrategie fur die Werbung.Munchen:FranzVahlenVerlag. Krugman,Dean M., RichardJ. Fox, James E. Fletcher, Paul M. Fischer, and Tina H. Rojas (1994), "Do Adolescents Attend to Warnings in Cigarette Advertising? An Eye-Tracking Ap-

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions

438

JOURNALOF MARKETING RESEARCH,NOVEMBER1999

proach,"Journal of Advertising Research, 34 (November/December), 39-52. Krugman, Herbert E.(1972), "Why Three Exposures May Be Enough,"Journal of AdvertisingResearch,4, 298-303. Lindsey, James K. (1995), Modeling Frequencyand Count Data. New York:Oxford UniversityPress. McCullagh,P. andJ.A. Nelder (1989), GeneralizedLinearModels, 2d ed. London:Chapman& Hall. Miniard,Paul W., Sunil Bhatla, KennethR. Lord, Peter R. Dickson, and H. Rao Unnava (1991), "Picture-BasedPersuasion Processes and the ModeratingRole of Involvement,"Journalof ConsumerResearch, 18 (June),92-107. Molnar,Francoisand Demitrios Ratsikas (1987), "Some Aesthetical Aspects of Visual Exploration,"in Eye Movements:From Physiology to Cognition, J. K. O'Regan and A. Levy-Schoen, eds. Amsterdam:North Holland, 363-74. Noton, D. and Lawrence Stark (1971), "Scanpathsin Eye Movements During PatternPerception,"Science, 171, 308-11. Pechmann, Cornelia and David W. Stewart (1989), "Advertising Repetition:A Critical Review of Wear-Inand Wear-Out,"Current Issues and Researchin Advertising, 11, 285-329. Petty, Richard E., John T. Cacioppo, and David W. Schumann (1983), "Centraland PeripheralRoutes to AdvertisingEffectiveness: The ModeratingRole of Involvement,"Journal of Consumer Research, 10 (September), 135-46. Pieters, Rik G.M. and Luk Warlop(1999), "VisualAttentionDuring Brand Choice," InternationalJournal of Research in Marketing, forthcoming. Ray, Michael L. and Alan G. Sawyer (1971), "Repetitionin Media Models: A LaboratoryTechnique,"Journal of MarketingResearch, 8 (February),20-29. Rayner, Keith (1995), "Eye Movements and Cognitive Processes in Reading, Visual Search,and Scene Perception,"in Eye Movement Research: Mechanisms,Processes and Applications,John M. Findlay et al., eds. Amsterdam:Elsevier Science Publishers BV, 3-22. Rethans,Arno J., John L. Swasy, and Lawrence J. Marks(1986), "Effectsof Television CommercialRepetition,Receiver Knowledge, and CommercialLength:A Test of the Two-FactorModel," Journal of MarketingResearch, 23 (February),50-61.

Rosbergen, Edward,Rik G.M. Pieters, and Michel Wedel (1997), "Visual Attention to Advertising:A Segment-Level Analysis," Journal of ConsumerResearch, 24 (3), 305-14. Ross, Sheldon (1996), Stochastic Processes, 2d ed. New York: John Wiley & Sons. Rossiter,John R. and LarryPercy (1983), "VisualCommunication in Advertising,"in InformationProcessing Research in Advertising, RichardJackson Harris,ed. Hillsdale, NJ: LawrenceErlbaum Associates, 83-125. and (1997), AdvertisingCommunicationsand Promotion Management,2d ed. New York:McGraw-Hill. Schumann, David W., Richard E. Petty, and D. Scott Clemons (1990), "Predictingthe Effectiveness of Different Strategiesof Advertising Variation:A Test of the Repetition-VariationHypothesis," Journal of Consumer Research, 17 (September), 192-202. Shanteau, James (1983), "Cognitive Psychology Looks at Advertising: Commentaryon a Hobbit's Adventure,"in Information Processing Research in Advertising, Richard Jackson Harris, ed. Hillsdale, NJ: Lawrence Erlbaum Associates, 153-66. Stark,LawrenceW. (1994), "Sequencesof Fixations and Saccades in Reading,"in Eye Movementsin Reading, Jan Ygge and Gunnar Lennerstrand,eds. Tarrytown,NY: Elsevier Science Inc, 151-63. and Henry C. Ellis (1981), "ScanpathsRevisited: Cognitive Models Direct Active Looking,"in Eye Movements:Cognition and VisualPerception,Dennis F. Fisheret al., eds. Hillsdale, NJ: LawrenceErlbaumAssociates, 193-226. Unnava, H. Rao and Robert E. Burnkrant(1991), "An ImageryProcessing View of the Role of Pictures in Print Advertisements,"Journal of MarketingResearch,28 (May), 226-31. Viviani, Paolo (1990), "Eye Movements in Visual Search:Cognitive, Perceptual and Motor Control Aspects," in Eye Movements and their Role in Visual and Cognitive Processes, E. Kowler, ed. Amsterdam: Elsevier Science Publishers BV, 353-93. Young, Laurence R. and David Sheena (1975), "Eye-Movement Measuring Techniques,"American Psychologist, 30 (March), 317-30.

This content downloaded from 129.93.176.95 on Fri, 11 Dec 2015 09:09:01 UTC All use subject to JSTOR Terms and Conditions