EvaluatingAnimationin thePeripheryasa Mechanismfor MaintainingAwareness

D. Scott McCrickard Richard Catrambone John T. StaskoDepartmentof ComputerScience Schoolof Psychology Collegeof Computing

VirginiaPolytechnicInstitute andGVU Center andGVU CenterandStateUniversity Georgia Instituteof Technology Georgia Instituteof Technology

Blacksburg VA 24061-0106USA AtlantaGA 30332-0170USA AtlantaGA 30332-0280USAmccricks@cs.vt.edu rc7@prism.gatech.edu stasko@cc.gatech.edu

Abstract: Small animateddisplayssuchastickersandfadersareincreasinglybeingusedto convey informationon computerscreens.Relatively little is understood,however, abouttheir useasperipheraldisplays,that is, toolsfor communicatinglower-priority awarenessinformation to people. This article describestwo experimentsthatexaminethe tradeoff of communicationcapability versusdistractionin peripheraldisplays. We found that thepresenceof animatedtextual peripheraldisplaysdid not distractpeoplefrom a centralinformationbrowsingtask,andwe identifiedparticularanimationanddisplaycharacteristicsthatfacilitatedifferentinformation-centrictasks.

Keywords: peripheraldisplays,awareness,monitoring,animation,empiricalevaluation,dual-taskevaluation

1 Intr oductionPeoplenaturallywish to staycontinually informedofongoing events of interest. For instance,an officeworker may want to stay appraisedof the weatheroutside, the traffic situation for the ride home, howcertainstocksare performing,or how well a favoriteteamis playing. While peoplemay want to maintainawarenessof suchinformation,orperhapsevenmonitorit intermittently, such awarenessideally should notdistractthemfrom their primarywork or task.

A variety of information communicationdeviceshave beendevelopedto help peoplemaintaina senseof casualawarenessof interestinginformation. Theclassicexamplesof thesetypes of devices are emailalerts,loadmonitors,andstocktickers. More recently,similar displays use visual and audio presentationmethods to show news, weather, sports, personaldata, and other information in a small portion ofthe desktop(Greenberg, 1996; McCrickard, 1999;Zhao & Stasko, 2000). Also becoming prevalentare off-the-desktopinterfacesthat use objects in theenvironment and changesin lighting or backgroundnoiseto communicateanything from network traffic totraffic in the hallways (Ishii & Ulmer, 1997; Heineret al., 1999).

Our focus in this article is a set of peripheralcommunicationtechniquesusedon computerdisplaysthat we call peripheral displays. Typically, peripheraldisplaysusevery little screenreal estate,but they stillattemptto convey a fairly largeamountof information.Often, this translatesinto someuse of animation tocycle throughitemsof interestvia scrolling or fadingtechniques.

While animationhas beenshown to be a strongperceptive attention draw that consequently maydistract peoplefrom their primary task (Ware et al.,1992), it has also proven to be an effective way toshow large amountsof information in a small space

(Robertsonet al., 1993). Researchershave speculatedthatsmoothanimationswould not beoverly distracting(Fitzpatrick et al., 2001), and organizations likeYahoo,ESPN,andAOL provide tickering and fadingdesktopdisplaysthatshow continuouslyupdatednewsheadlines,stockquotes,sportsscores,weatherreports,and the computeractivity of friends. Thereare eventoolkits thathelpenableprogrammersto includetheseandsimilar techniquesinto their interfaces(Fitzpatricket al., 1998; McCrickard& Zhao,2000).

While numerousstudieshave examinedpeople’swillingness to use peripheraldisplaysin maintainingawareness(for example (Parsowith et al., 1998;McCrickardet al., 1999)),relatively little researchhasbeenconductedto understandbetter the informationcommunicationversusdistractiontradeoff for differenttechniquesof peripheralcommunication.Our goal isto explore the balancebetweendistraction, reaction,and comprehensionfor different animatedperipheraldisplays via empirical evaluations of realistic butcontrolledsituations.Thispaperdescribesseveralsuchevaluationsthat asked participantsto searchhypertextspacesfor answersto a series of questionswhilecompletingactivitiesandansweringquestionsbasedoninformationin peripheraldisplays.

2 RelatedworkSomeof theearliestevaluationsof constantlychangingdisplaysexaminedtheperceptibilityandreadabilityofrapid serial visual presentations(RSVPs) of letters,strings,andwords.Fosterfoundthatparticipantscouldcorrectly identify about four out of six words in asentencewhen rapidly presenteda word at a time inasinglevisuallocation(Foster, 1970).Juolaalsofoundthat comprehensionof information was comparablewhenpresentedasRSVPsandin multi-line paragraphformat (Juola et al., 1982). In some of the firststudiesof smootheranimatedeffects, Duchnicky andKolersperformedaseriesof experimentsexaminingthe

readabilityof text scrolledon visual displayterminalsas a function of window size (Duchnicky & Kolers,1983). They foundthat largerdisplaystypically led tofasterperformanceon readingtasks. A study led byGranaasfound that in scrolleddisplays,larger jumps(four to tencharacters)ledto bettercomprehensionthansmallerjumps(oneto two characters)(Granaaset al.,1984).KangandMuter, in comparingatickeringeffectto a non-animatedRSVP effect, found no differencein comprehensionfor a readingtask (Kang & Muter,1989). Theseexperimentsaddressedmany importantfactors that we explore further in our research,includingdifferentinformationaltasks(recognitionandcomprehension),differentsizeddisplays,anddifferentwaysto changethedisplay.

All the previously mentioned evaluationsconsideredthe readingof small animateddisplaysasthesoletaskof theparticipant.However, in thecaseofperipheraldisplays,participantswould be performingsome main task with attention to a small animateddisplaypartof a secondarytask. Oneexperimentwiththis type of dual-taskscenariowas conductedwithOwnTime,a peripheraltimespacemanagementsystemthatalertspeoplewhenvisitorsarewaitingto meetwiththem(Rodensteinet al., 1999). The study found thatOwnTime visitor interactionswere lessintrusive thandirect engagementfor participantsperforming recallandcomprehensiontasks. The researchof Bartrametal consideredthe effectivenessof using motion cuesto draw attention(Bartramet al., 2001). They foundthatmotion cuesoutperformstaticrepresentationsandthat certaintypesof motionsaremoredistractingandirritating thanothers.

In other work, Maglio and Campbellperformedaseriesof dual-taskexperimentsin which participantsperformeddocumentediting taskswhile a peripheraldisplay showed news headlineslater usedto answerquestions(Maglio & Campbell,2000). Theperipheraldisplays included a continually scrolling displaythat jumped five pixels per step, a start-and-stopscrollingdisplaythatbriefly pausedwheneachheadlineappearedon the screen, and a fading display thatincreasedthe brightnessof the text to make it visible.They found no difference in the communicationabilities of different peripheraldisplays(as measuredby how well information is remembered). Also, allof the animatedperipheraldisplayswere found to bedistractingto themaintaskof documentediting,thoughthestart-and-stopdisplaywastheleastdistracting.

Researchon the effectsof InstantMessaging(IM)notificationson desktopcomputertasksfound that IMtypically was disruptive to primary tasks,particularlyso for fast,stimulus-driven searchtaskssimilar to theonesin the Maglio experiments(Cutrell et al., 2001).However, IM does not use smooth animation in itsupdates,which may have excaberatedthe distraction.Our studyexaminedwhethera slower, semantic-basedsearchtask is affected by various smoothperipheraldisplays, and whether the peripheral displays caneffectively communicateinformationto users.

3 ExperimentsTo examine whether animated displays impactinformation acquisitionwhen maintainingawareness,

two empiricalevaluationswereconducted.Participantswere asked to completea seriesof browsing taskswhile simultaneouslykeepingabreastof a peripheraldisplay showing constantlychangingnews, weather,stock, and sports information. We utilized threeperipheraldisplaysin theseexperiments: a tickeringdisplaythathorizontallymovesinformationacrossthescreen,a fading display that graduallyfadesbetweenpiecesof information, and a RSVP-style“blast” thatswitchesbetweenitemsin the displaywithout smoothanimation. For the tickering effect, we employed asmoothanimationthatrepeatedlymovesthetext apixelat a time in an attemptto minimize distraction. Theearly previously-describedstudies typically tickereda display by several charactersat a time (Duchnicky& Kolers, 1983; Granaaset al., 1984; Kang &Muter, 1989), and even Maglio and Campbell’s 5-pixel jump when scrolling createsa jerky effect thatmayhaveresultedin unnecessarydistraction(Maglio &Campbell,2000).Priorwork hasnotedthatpeopletendto performbetteroncertaindecision-makingtaskswithsmootheranimations(Gonzalez,1996). We suspectthatsmoothanimationsmayproveto belessdistractingthantheonesusedin prior work.

Figure1: Layoutof theexperimentalenvironmentexperiencedby participants.At thecenteris thebrowser

usedby theparticipantsin theexperiment.At thetopof thescreenis thefadeperipheraldisplaythatcyclically showed

thestateof severaltypesof information.At thebottomis theareausedfor monitoringactivities. After eachround,the

screenclearedexceptfor a questionareawheretheawarenessquestionswerepresented.

Participantsusedthe information presentedin theperipheraldisplay to completeshort-termmonitoring-style awarenessactivities (monitoring activities) andto answer longer-term knowledge-gain questions(awarenessquestions). The experimentsconsistedofseveral rounds (six in the first experiment, eight inthe second),eachconsistingof four browsing tasks,two monitoring activities, and up to five awarenessquestions. The layout of the information on thecomputerscreenis in Figure 1. Motivationsfor ourexperimentalchoicesfollow.

3.1 Browsing tasksIn performing the browsing tasks, participantsuseda simple browser and hypertext pages. The browserconsistedof a textual information area containinga number of condensedpages from World WideWeb sites. The text-only information areacontainedhighlighted, underlined links that pulled up otherpageswhenclicked with the mouse. The participantsnavigated the information spaceby clicking on thelinks andby usingthe forwardandbackbuttons. Thebrowsing taskswere non-trivial: the participantshadto readandnavigatethrougha hypertext spaceto findcertain information in the pages,enter it into a boxconnectedwith the browser, and press a button tocontinue.

To minimize the typing required,all solutionstobrowsing taskquestionswerenumerical(for example,“In what year was Mount Rushmorecarved?”) If anincorrectanswerwasentered,theinterfacebeepedandtheparticipanthadto continueworking on theproblemuntil thecorrectanswerwasentered.Whenthecorrectanswerwasentered,theparticipantcouldproceedto thenext browsingtask.Theorderin which browsingtaskswerepresentedwasheldconstantfor all participants.

3.2 Monitoring activitiesWhile performingthe browsing tasks,the participantsusedinformationin theperipheraldisplayto completea setof monitoringactivities andto answera seriesofawarenessquestions.Theperipheraldisplaycyclicallyshowed instancesof different types of information,suchas a sportsscore,a stock quote,and a weatherreport. Each instancewas updatedfrequently butirregularly as it often is in real life. Participantswere asked to pressa button when the informationin the peripheraldisplay matchedsomecriteria (forexample,“Whenthetemperaturedropsbelow 35,pressOK1.”) The informationthat wasselectedfor displaywas interestingbut rarely vital, and the informationaloccurrencesthat were selectedwere chosenbecausethey mightspurauserto performsomereal-lifeactivity,suchasbringing in a plant that is outdoorsor sellingastockthatis performingpoorly.

Eachroundincludedtwo suchmonitoringactivities.Theorderin whichmonitoringactivitieswerepresentedwas held constantfor all participants. If the buttonwas pressedat the correct time (that is, after theneededinformationwaspresented),it wasgreyed outto alerttheparticipantthatthetaskhadbeencompletedsuccessfully. If the button was pressedtoo soon,theinterfacebeepedandthebuttonremainedactive.

3.3 AwarenessquestionsAt the end of each round, the participants weregiven awarenessquestionsthat asked them to recallinformation that wasshown in the peripheraldisplay.The questionswere multiple-answermultiple-choicequestionsthat addressedboth content and temporalissues. Eachquestionhad four possibleanswers,allinitially unselected,andtherewasalwaysat leastonecorrectanswer.

The first questionin eachset listed four typesofinformation and asked the participant to choosetheonesthathadbeendisplayed.If they correctlyrecalled

seeinginformation,laterquestionsasked aboutdetailsof it, such as which news stories appeared,whichstockquotesconstantlyincreased,or whichsportsteamscoredthe most points. For example,if a participantcorrectlynotedthatnewsheadlineshadbeendisplayed,later questionswould presenta list of headlinesandasktheparticipantto selecttheonesthathadappeared.All of the information was fictional but realistic, andno attempt was made to intentionally deceive theparticipants with slightly different information (forexample,a stockquotethatalmostalwaysincreased).

3.4 Data collectionand evaluationTo compareperformanceamonggroups,thedependentvariableswere the times for all browsing tasksandmonitoring activities and the answersto the post-roundawarenessquestions.The resultswereanalyzedto determinewhetherdifferencesin certainmeasuresoccurred for participants in different conditions(participantsusingdifferenttypesof peripheraldisplaysin thefirst experiment,andparticipantsusingdifferentsizesandspeedsin thesecond).

The browsing time is the time from which thebrowsing task and browser information appearedonthe screento the time when the participanttyped inthe correctanswerand pressedthe OK button. Themonitoringtime is thetime from whentheinformationwas first entered into the cyclic display until theinformationwasacknowledgedasmatchingthecurrentcriterionvia abuttonpress.

For the awarenessquestions, the participants’responsesto eachof the four answerswerecollected.We consideredeach question as being worth fourpoints: correctlyor incorrectlyassessingeachpossibleresponseto a question.

A number of different methodscan be used todeterminea participant’s ability to recall information.The most obvious measureis to comparethe percentof correct responsesfor the awarenessquestionsindifferent situations. The percentof correctresponsesis referredto asthecorrectnessrate.

The correctness rate measure potentially canmisrepresenta participant’s awarenessof information,however. Notethataparticipantwhodid not rememberseeinganything in the peripheraldisplay and left allresponsesunchecked could have a correctnessrate ashigh or higher than a participant who rememberedseeingseveral itemsbut wasmistakenaboutwhat wasseenandcheckedthewrongbox(es).

An alternate measure for determiningresponsivenessis the hit rate, a term from signaldetection theory defined as the ratio of correctlyidentified stimuli to the total number of times thestimulus was presented. The hit rate is typicallyaccompaniedby the false alarm rate, the ratio ofincorrectstimuli responsesto thetotalnumberof timeswhenthestimuluswasnotpresent.Sincea typicalgoalwhenusinga peripheraldisplayis to proactively recallseeinginformation, it may be better for a participantto be mistaken about seeing information that wasnot displayedthan to be mistaken about not seeinginformation that in fact was displayed. Perhapsapersonwantsto rememberthatastoryoccurred,or thata tornadowatchis underway, or that a traffic bulletin

appeared. The hit rate would reflect the awarenesspotentialof ananimateddisplay.

In analyzing the results, analysesof variance(ANOVAs) were performed to check for statisticalsignificance among different conditions of theexperiments. If the ANOVA revealed a significantdifference, pairwise t-tests were performed todeterminewhichconditionsdiffered.

3.5 Experiment 1The first experiment comparedrelative performancewhenusingfading, tickering, andblastingdisplaysaswell aswhenno peripheraldisplaywaspresent.

3.5.1 MethodThis experiment focused on three factors: thepossibility for degradation in performance on abrowsing taskwhena peripheraldisplaywaspresent,the speed in identifying and reacting to changesin peripheraldisplays, and the ability to rememberinformationthatappearedin aperipheraldisplay.

Seventyundergraduatestudentsparticipatedin thisexperiment for class credit. The experiment wasconductedon identicalworkstations,eachconnectedtoa 15-inchmonitor with an optical mouse.Participantswererun in smallgroups,oneparticipantpercomputer.The experimentwasexplainedto eachgroupverballyandagainon thecomputerwith examples.

Theparticipantsperformedsix roundsof browsingtasks,monitoring activities, and awarenessquestions.In eachround, participantscompletedfour browsingtaskswhile performingtwo monitoringactivities usingeither a fade, ticker, or a blast animation. Thespeed with which the information was displayedcorrespondedto the mean speedsfor each deviceselected by the participants in a previous study(McCrickard, 2000). While this resultedin differentratesof informationdisplayfor theanimations,we feltit wasa morerealisticandecologicallyvalid measureof how peoplewould usethem. Theticker continuallyshiftedonepixel every 50 milliseconds,while thefadeandblastupdatedtheir entirecontentsevery2 seconds.The fade required500 millisecondsto fade betweenitems,while theblastupdatedinstantaneously.

At the end of each round, participantsansweredawareness questions about the information thatappearedin the animateddisplay. The first questionasked which types of information appearedin thedisplay. For each correctly-identified instance ofinformation appearing, two questions about theinformationwereaskedup to a totalof fivequestions.

As a basecase,onegroupof participants(n � 15)did not have any animationspresentat any time andassuchperformedonly thebrowsingtasks.For theothergroups,all participantsexperiencedall animations,withordersbasedon a latin squaredesign(blast - fade -ticker (n � 17), fade- ticker - blast(n � 17),or ticker -blast- fade(n � 21)). A differentanimationwasusedfor eachof thefirst threeroundswith theorderrepeatedon thelastthree.

3.5.2 ResultsFor the time requiredto carry out the browsing tasks,therewasnotasignificantimpactdueto thepresenceofaperipheralanimateddisplay(F

�3 � 58� � 0 � 60,MSE �

46277� 71,p � � 62). Furthermore,thetypeof animationdid not affect the browsing times (F

�2 � 46� � 0 � 62,

MSE � 25411� 63, p � � 54), (seeFigure2).

1 2 3 4 5 60

50

100

150

200

250

300

350

400

450

Round

Bro

wsi

ng ti

me

(sec

)

No animationBlast Fade Ticker

Figure2: Averagecompletiontimesfor browsingtasksforeachroundbasedon thetypeof animationthatwaspresent.Participantsperformedaboutthesameon thebrowsingtasksregardlessof thetypeor eventhepresenceof animation.Byshowing theroundsindividually, onecanseethatthereis notevena trendto suggestthatparticipantsperformedbetterin

certaincases.

In the monitoringactivities, participantspressedabutton when they noticed specifiedinformation in aperipheraldisplay. Since a round endedwhen thebrowsing taskswere completed,it was possiblethatnot all of the monitoring activities were completed,meaningit waspossiblethat sometypesof peripheraldisplaysled to more activities completedthanothers.However, the numberof activities completed(out of 4possible,blast � 3 � 83,fade � 3 � 80,ticker � 3 � 71)doesnot dependon the typeof animation,F

�2 � 96� � 0 � 77,

MSE � 0 � 25, p � � 46. That is, it doesnot appearthata participantis morelikely to identify (or miss)a pieceof informationwhenusingonetypeof animateddisplaythananother.

While the identification rate when monitoringinformation was not affected by device type, thetime to react to it was. The times to completethemonitoring activities differed significantly dependingon the type of animationsused,with blast requiringan averageof 88.85 secondsin each round for thetwo activities, fade117.41seconds,andticker 192.93seconds(F

�2 � 52� � 17� 24, MSE � 4528� 75, p �

� 001). Pairwise comparisonsrevealedthat the blastand fade animationsresulted in significantly fastermonitoringtimesthantheticker(p ��� 001andp � � 01,respectively) andtherewasa trendtoward fasterblasttimesthanfade(p ��� 09).

For the awarenessquestions,Figure3 summarizesthe resultsusingthe threemetricsdescribedearlierformeasuringperformanceon the awarenessquestions:correctnessrate,hit rate,andfalsealarmrate.

While the correctnessrate for ticker was slightlyhigher than that for fadeor blast, suggestingthat thetickermaybebetter, thedifferencewasonly marginallysignificant,F

�2 � 96� � 2 � 62, MSE � 0 � 02, p � � 08. In

turningto thehit rate,therewasasignificantdifferenceamonghit rates.Thehit ratefor ticker washigherthanthat for fadeandblast,F

�2 � 96� � 3 � 87, MSE � 0 � 03,

p ��� 03.

Correct Rate Hit Rate False Alarm Rate0

10

20

30

40

50

60

70

Rat

e (p

erce

nt)

Blast Fade Ticker

Figure3: Cumulative correctnessrate,hit rate,andfalsealarmratefor theawarenessquestions.Theparticipantshad

a significantlyhigherhit ratewhenusingtheticker.

Onedrawbackof many techniquesthatachievehighhit ratesis thatthey oftenresultin highfalsealarmratesas well. However, this was not true in this situationas therewas not a significantdifferenceamongfalsealarmratesfor blast,fade,andticker, F

�2 � 96� � 0 � 55,

MSE � 0 � 03, p � � 58.3.5.3 DiscussionOnemight suspectthat participantswho arenot facedwith ananimationandnotburdenedwith theadditionalmonitoring activities and awarenessquestionswouldperformsignificantlyfasteron the browsing tasks,butthe resultssuggestthat this is not the case. In fact,the resultsdid not even indicatea trend toward lowertimes in the caseswhen an animateddisplay is notpresent(seeFigure2). While this seemsto contradicttheresultsfoundin prior studies(Maglio & Campbell,2000),recallthattheprimarytaskin thosestudieswereediting tasksthat requiredparticipantsto perform in-depth readingsand make corrections. The browsingtasks of this experiment were less demanding,butbasedon previous studies(Fitzpatrick et al., 1998;McCrickard,2000)they alsomaybettermatchthetypeof primarytasksthatauserwouldbedoingwhile usingperipheraldisplays.

The times to complete the monitoring activitiesdiffered significantly depending on the type ofanimations used. The blast and fade animationsresultedin significantly fastermonitoring times thanthe ticker. This result seemsto follow from previousresults that indicate that moving text is read moreslowly than non-moving text (Sekey & Tietz, 1982;Granaaset al., 1984).As thetickeringdisplayreliesonmotionto cycle betweenitemswhile thefadeandblastdo not, it seemsreasonablethat theticker would resultin slower performance,particularly if the participantswere readingthe displaysto identify information thatthey weremonitoring.

In analyzing the responsesto the awarenessquestions,the correctnessand hit ratesfor the threeanimation types suggestedthat the ticker may bebetter. This does not contradict the results notedpreviously indicating that moving text was moredifficult to readthannon-movingtext. Thenatureof themonitoring activities and the awarenessquestionsarequite different. In fact, otherstudieshave shown thatcomprehensibility, unlikereadingspeed,is not affectedby motion (Kang & Muter, 1989), so it is reasonableto expect the results to differ. The implication oftheseresultson the developmentanduseof animateddisplaysis clear:if thegoalis to identify itemsquickly,an in-place display like a fade or blast should beused,while if the goal is to increasecomprehensionandmemorability, a motion-baseddisplaylike a tickershouldbeused.

3.6 Experiment 2The previous experiment suggestedthat there aredifferencesin performancewhen using the fade andticker displays.In a follow-up experiment,we wantedto explore whether certain factors, namely displaysize and animation speed,impactedperformanceinany way. Perhapsmaking the display area largerwould result in faster recognition times and allowthe awarenessquestionsto be answeredwith greateraccuracy, or perhapsa slower speedwould be lessdistracting,resulting in lower times on the browsingtasks.

3.6.1 MethodNinety-oneundergraduatestudentsparticipatedin thisexperiment for class credit. The materials andprocedureweresimilar to theonesusedin thepreviousexperimentwith thedifferencesdescribedhere.

A between-subjectssize and speedcondition wasadded.Theparticipantswerepresentedwith a displayhaving oneof threecharacteristics:normaldisplaysizeand animationspeed,normal size but slow speed,orsmallsizebut normalspeed.

The normal displayswereusedas the comparisonpoint for thesmallandslow displays.Normaldisplaysusedlarge display areasand fast speeds,thoughbothwell within the rangesof sizesandspeedsselectedbyparticipantsin a previous study (McCrickard, 2000).Both the fadeand ticker had a width of 1180 pixels(about160 characters)with a heightof oneline. Thissizewaschosenbecauseit fits nicely alongthe top orbottom of the screenand becauseit is large enoughto hold long streamsof information (such as newsheadlinesandweatherbulletins) in their entirety. Theticker speedwas at the upper rangeof the possiblespeedsfor theplatform,onepixel per20 milliseconds.The fade cycle step had a 100 millisecond delaybetweeneachof five stepswith a three-seconddelaybeforethenext fade.

Thesmalldisplayuseda smallerareabut thesamespeedas the normal display. The fade and tickerwidth was more than halved to 840 pixels (about70characters),smallenoughto fit abovea singleterminalwindow. Thisreductionin sizemeantthatmoststreamsof informationcouldnot beshown in their entirety.

The slow displaywas the samesizeas the normaldisplay, but slower. Thespeedwaschosento beat the

slow end of the rangeselectedby participantsin thepreviousstudy. Thetickerupdatedatarateof onepixelevery 140 milliseconds. The fadeupdatedone shadeevery150millisecondswith adelayof 9 secondsbeforethenext fade.Thesizefor thewidgetswasthesameasin thenormaldisplay.

We focusedon resultsfrom pairs of displaysthatdiffered in eithersizeor speedbut not both. That is,we did not examinethe differencesbetweenthe slowand the small displaysbecausethey differed in bothspeedandsize.Insteadwefocusedonthenormal-smallversusnormal-slow pairingsso that any differencesinperformancecould be attributed more confidently toonefactor.

Theawarenessquestionsalsodifferedfrom thefirstexperiment. In the first experiment,the first questionasked participantsto select the types of informationthat were displayed, then for each case where theparticipantstatedcorrectly that a type of informationwas displayed, two additional questionswere askedaboutthat information,thefirst relatingto contentandthe secondrelating to order. In this experiment,eachparticipantansweredall five of the questions. Thischangeseemedreasonablesincea cuesuchasa wordor phrasein aquestioncanaid retrieval from memory.

The numberof roundswas increasedfrom six toeight. It wasdeterminedthatparticipantswouldstill beable to completethe experimentwithin the requestedhour even with the additionalrounds. Also, the typeand order of the animationswas changed. Sincetheblastdisplayresultedin performancesimilar to thefadedisplayandwasconsistentlyratedasthe leastfavoritedisplaybyparticipantsin thefirst experiment,it wasnotusedin the secondexperiment.Participantsalternatedbetweenusingfadeandticker in eachround,with onegroup startingwith fade,then twice using ticker thenfadeagainandrepeating(fade,ticker, ticker, fade),andtheothergroupswappingtheorderstartingwith ticker.

In summary, thereweresix groupsof participantsdifferentiatedbyanimationclassification(normal,slow,or small)andstartinganimation(fadeor ticker). Eachgroup had 15 participantsexcept the slow fade-firstgroupwith 16 participants.

3.6.2 ResultsFor the cumulative time required to complete thebrowsing tasks,changesto the size and speedof theanimateddisplaydid not leadto differencesfor eitherthe fadeor the ticker, (for fade: normal � 2999� 03,slow � 2890� 87, small � 3131� 10 with F � 2 � 88� �0 � 58, MSE � 765992� 08, p � � 56; for ticker: normal� 3036� 83, slow � 3036� 93, small � 3079� 45 withF�2 � 88� � 0 � 01,MSE � 997900� 50, p � � 99).For monitoring activities, the resultsshowed that

changesto thenatureof theperipheraldisplayaffectedperformancewhen using the ticker (F

�2 � 29� � 5 � 23,

MSE � 40792� 90,p � � 01)but not thefade(F � 2 � 66� �1 � 62, MSE � 12712� 58, p � � 21). We consideredpairwise t-tests to determinewhere the significancelay. Did the size of the display affect performanceon monitoring activities? Figure 4 suggeststhat itdoes. Whenusingthe ticker, the time to completethemonitoring activities wassignificantly different basedon thesizeof theanimateddisplay, p � � 02.

Fade Ticker0

20

40

60

80

100

120

Mea

n m

onito

ring

time

(sec

)

LargeSmall

Figure4: Meancompletiontimesfor eachmonitoringactivity whenusinglarge(normal)andsmalldisplays.Fortheticker, smallerdisplaysresultedin significantlylower

timesthanlarger.

Did thespeedof thedisplayalsoaffectperformanceon monitoringactivities whenusing ticker? For eachdisplaytype themonitoringtimesappearto have beensimilar regardlessof speed(fade41.1and45.7seconds,ticker 102.3 and 100.7 secondsfor the fast and slowdisplays,respectively). Theanalysisverifiesthat therewasno significantdifference:for the ticker, the t-testresultedin p � � 79,for thefade,p � � 87.

For theawarenessquestions,thecumulativehit ratewhenusingthe fadewasvirtually identical regardlessof display size or speed(with normal 67.4%, slow66.8%,small68.3%andF

�2 � 88� � 0 � 09,MSE � 0 � 02,

p � � 92). However, when using the ticker, the hitratesdid differ significantly(with normal68.2%,slow61.3%, small 68.3% and F

�2 � 88� � 3 � 26, MSE �

0 � 02, p � � 04). A t-test revealedthat the differencebetweenthe normalandslow displayswassignificant(p � � 03),while thedifferencebetweenthenormalandsmall displayswasnot (p � � 98). As in the previousexperiment,thefalsealarmratedid notdiffer regardlessof display size or speedfor either the fade (normal18.4%, slow 16.8%, small 20.8%; F

�2 � 88� � 2 � 24,

MSE � 0 � 01,p � � 17)or theticker(normal19.3%,slow18.8%, small 19.7%; F

�2 � 88� � 0 � 10, MSE � 0 � 01,

p � � 91).3.6.3 DiscussionAs wasnotedin thepreviousexperiment,for thetypeorpresenceof ananimateddisplay, neitherthedisplaysizenor the animationspeedseemedto negatively impactthetime requiredto completebrowsingtasks.

Users did not complete monitoring tasks morequickly with a fast animation than a slow one.One possibleexplanationfor this result is that eventhough the increasein speedgives participantsmoreopportunitiesto seethe information,eachopportunityis shorter becausethe information disappearsmorequickly. Thesetwo factorscould balanceout to resultin similarmonitoringtimes.

Althoughchangesin thespeedof thedisplaydonotseemto affect monitoring times, changesin the sizedo. A smaller display results in significantly fastermonitoring times when using the ticker, and there is

a trend toward fastertimes when using the fade(seeFigure4). While thisresultis notimmediatelyintuitive,it does seemto correspondto the model that mostpeopleuseduringperipheralmonitoringactivities: theyare focused on a primary task while occasionallyglancingat theperipheraldisplay. As notedby Rayner(Rayner,1978),only alimited numberof characters(upto 20) canbe processedin a quick glanceat a display.The greaternumberof charactersin a larger displaymay make it more difficult for a personto find thedesiredinformationwith aquick glance.

While the size of the display seemedto impactperformanceon the monitoring activities, it doesnot seem to impact performanceon the awarenessquestions. The participantsperformedequally wellon the questionswhetherusing the larger or smallerdisplay. One reason the display size may makea difference in answeringthe questionsbut not inperformingthe monitoringactivities lies in the natureof the two tasks. Whereasthe monitoring activitiesrequire only a quick glance, the ability to answerquestionsrequiresa morecarefulreadingof the entireinformation entry. In fact, it is somewhat surprisingthat the larger display did not show better resultsonthe awarenessquestionsthan the smallerone, thoughthis maybebecausethe smallerdisplaywasstill largeenoughto containmost or all of the information formany of theinformationentries.

The speed of the tickering display seemedtoimpact performanceon awarenessquestions. Aslower animation resultedin poorer performanceonthe questionsthan a fasterone, perhapsbecausetheparticipants too often glanced up to see the sameinformation. This resultis consistentwith theGranaaswork thatshowedfastertickers(with largerjumpsin thescrolling)resultedin bettercomprehensionthanslowertickerswith smallerjumps(Granaaset al., 1984).

In comparing performance on the awarenessquestionsbetweenexperiments, the result from thepreviousexperimentwasnot replicated:the ticker didnot resultin improvedperformanceover thefade.Thismay be relatedto the amountof information on thescreen.Thetickersin this experimentwereaslargeorlargerthantheonesin thefirst experiment,resultinginfarmoreinformationonthescreenatany giventimeforthe ticker, especiallycomparedto the fade. The shapeof the displaymay alsobe a factor. The fadedisplayin the first experimentwasa three-linedisplay, whilein thesecondit wasa one-linedisplayidenticalin sizeand shapeto the ticker. It is possiblethat multi-linedisplaysaremoredifficult to processandcomprehendin a glancethan single-linedisplays. Futurework isnecessaryto testthathypothesis.

4 GeneraldiscussionThegoalof theempiricalevaluationswasto explorethebalancebetweendistraction,reaction,comprehension,and memorability when using peripheral animateddisplays. The first experiment showed that fading,tickering, and blasting peripheral displays did notsignificantly distract users from a primary task yetcould effectively communicateinformation. The typeof animation impacted performanceon monitoringactivities and awarenessquestions. The second

experimentshowedthatchangesin sizeandspeedalsocouldimpactperformanceon monitoringactivities andawarenessquestions.

The following recommendationscan be derivedfrom theresultsof theseexperiments:

Animated displays can be used in the peripherywith minimal negative impact on certain primarytasks. While other work (Maglio & Campbell,2000)seemsto suggestotherwise,bothexperimentssupportedthis claim. Thedifferencemayresultfromaprimarytaskthatis lesscognitively demandingandsmoother, sloweranimations.

In-place displayssuchasfadeand blast arebetterthan motion-baseddisplays lik e ticker for rapididentification of items. Participantswere able tocompletemonitoring activities more quickly whenusingthe fadeandblast thanwhenusingthe ticker.This seemsto extendprior resultsthat indicatedthatmoving text is moredifficult to readthanstatic text(Sekey & Tietz,1982; Granaaset al., 1984). Motion-based displays such as ticker are better

than in-place animations for comprehensionand memorability. While in-place displays aidrapid identification, on the awarenessquestionsparticipantswho used the ticker obtaineda betterhit rateanda marginally bettercorrectnessratethanthosewho usedthe blastand the fade. At the veryleast,this suggeststhat if it is essentialto rememberinformation at the risk of mis-rememberingit, amotion-baseddisplayshouldbeused.

Small displays result in faster identification ofchanging information. This may be relatedto theamountof information that a viewer can read in aglance. Larger displays may make it difficult toobtaindesiredinformation. Fast displaysare better than slow for establishing

comprehensionand memorability. This wasnotedin thesecondexperimentfor thetickeringanimationandmayberelatedto theamountof new informationthat is available in a glance. Slower displaysmaydiscouragepeoplefrom lookingat themoften.

5 Conclusionsand futur e workThe results of the experiments and the growingpopularity of peripheraldisplayssuggestthat peoplecan use them to help maintain awareness. Besidesproviding immediateinformation, smoothlyanimatedchangesto the display can be small, subtle, andpredictable,allowing theuserto adaptto thechangingdisplay to the point where it is less distracting. Itis necessaryto understandthe trade-off betweentheimportanceof theinformationbeingcommunicatedandtheresourcesnecessaryto displayandprocessit.

This researchhas developedan understandingofthenatureof theawarenessproblemandof how users’wantsandneedsdiffer in maintainingawareness.It isreasonableto concludethat the useof animationcanassistin maintainingawarenesswithout causingunduedistractionin particularsituations.As with mosttools,peripheralanimationscan be andhave beenmisused,

but whenusedproperly, this researchhasshown theyhave thepotentialto bebeneficial.

The population of participantsin the study wasundergraduatesat a technicalschool with significantexposure to computers and interfaces. Furtherstudiesare necessaryto determinethe effectivenessof peripheral displays for other segments of thepopulation,suchaspeoplewith morelimited skills andpeoplewith little computerexperience.Otherongoingwork is examiningwhat typesof informationarebestsuitedfor display in the peripheryand other primarytasksthat arenot negatively impactedby the presenceof peripheraldisplays. As was notedpreviously, theparticular focus of this work on controlledsituationsallows us to understandandcompareperformanceforcertainscenarios,but it is also importantto apply thelessonslearnedin this work to real-world applications.

AcknowledgmentsThis researchwas partially supportedby Air ForceOffice of Scientific ResearchGrant F49620-98-1-0362 to Richard Catrambone. Thanks to GregoryAbowd, Amy Bruckman,Mark Guzdial, Alex Zhao,and the anonymousreviewers for their assistanceandcommentson this research.

ReferencesBartram,L., Ware,C. & Calvert, T. (2001),Moving Icons:

DetectionandDistraction,in Proceedingsof INTERACT2001, Tokyo, Japan.

Cutrell, E., Czerwinski, M. & Horvitz, E. (2001),Notification, Disruption, and Memory: Effects ofMessagingInterruptionson MemoryandPerformance,in Proceedingsof INTERACT2001, Tokyo, Japan.

Duchnicky, R.L. & Kolers,P. A. (1983),“Readabilityof TextScrolledon Visual Display Terminalsasa FunctionofWindow Size”,HumanFactors 25(6), 683–692.

Fitzpatrick, G., Kaplan, S., Mansfield, T., Arnold, D. &Segall, B. (2001), “SupportingPublic Availability andAccessibilitywith Elvin: ExperiencesandReflections”,ComputerSupportedCooperativeWork: TheJournalofCollaborativeComputing. to appear.

Fitzpatrick,G.,Parsowith, S.,Segall, B. & Kaplan,S.(1998),Tickertape:Awarenessin a SingleLine, in ConferenceCompanionof theACM Conferenceon HumanFactorsin ComputingSystems(CHI ’98), Los Angeles,CA,pp.281–282.

Foster, K. I. (1970),“V isual perceptionof rapidly presentedwordsequencesof varyingcomplexity”, PerceptionandPsychophysics8, 215–221.

Gonzalez, C. (1996), Does animation in user interfacesimprove decisionmaking?,in Proceedingsof theACMConferenceon HumanFactors in ComputingSystems(CHI ’96), Vancouver, BC, pp.27–34.

Granaas,M. M., McKay, T. D., Laham,R. D., Hurt, L. D.& Juola,J.F. (1984),“ReadingMoving Text on a CRTScreen”,HumanFactors 26(1), 97–104.

Greenberg, S. (1996), Peepholes:Low Cost AwarenessofOne’s Community, in ConferenceCompanionfor theACM Conference on Human Factors in ComputingSystems(CHI ’96), Vancouver, BC, pp.206–207.

Heiner, J. M., Hudson, S. E. & Tanaka,K. (1999), TheInformation Percolator:Ambient Information Display

in a Decorative Object, in Proceedingsof the ACMSymposiumon User InterfaceSoftware andTechnology(UIST’99), Asheville, NC, pp.141–148.

Ishii, H. & Ulmer, B. (1997), Tangible Bits: TowardsSeamlessInterfacesbetweenPeople,Bits, andAtoms,in Proceedingsof the ACM Conference on HumanFactors in ComputingSystems(CHI ’97), Atlanta,GA,pp.234–241.

Juola, J. F., Ward, N. J. & McNamara,T. (1982), “V isualSearch and Reading of Rapid Serial Presentationsof Letter Strings, Words, and Text”, Journal ofExperimentalPsychology: General 111(2), 208–227.

Kang, T. J. & Muter, P. (1989), “Reading DynamicallyDisplayed Text”, Behaviour and InformationTechnology 8(1), 33–42.

Maglio, P. P. & Campbell, C. S. (2000), Tradeoffs inDisplaying PeripheralInformation, in Proceedingsofthe ACM Conferenceon HumanFactors in ComputingSystems(CHI 2000).

McCrickard, D. S. (1999), Maintaining InformationAwarenesswith Irwin, in Proceedingsof the WorldConference on Educational Multimedia/Hypermediaand Educational Telecommunications(ED-MEDIA’99), Seattle,WA.

McCrickard, D. S. (2000), Maintaining InformationAwarenessin a Dynamic Environment: AssessingAnimation as a Communication Mechanism, PhDthesis, College of Computing, Georgia Institute ofTechnology, AtlantaGA.

McCrickard, D. S. & Zhao, Q. A. (2000), SupportingInformation Awarenessusing Animated Widgets, inProceedingsof the2000USENIXConferenceonTcl/Tk(Tcl2K), Austin,TX, pp.117–127.

McCrickard, D. S., Stasko, J. T. & Zhao, Q. A. (1999),Exploring Animation as a PresentationTechniqueforDynamicInformationSources,TechnicalReport99-47,GeorgiaTechGVU Center.

Parsowith, S.,Fitzpatrick,G.,Segall, B. & Kaplan,S.(1998),Tickertape:NotificationandCommunicationin aSingleLine, in AsiaPacificComputerHumanInteraction1998(APCHI ’98), Kanagawa,Japan.

Rayner, K. (1978), “Eye movements in reading andinformation processing”, Psychological Bulletin85, 618–660.

Robertson, G. G., Card, S. K. & Mackinlay, J. D.(1993),“InformationVisualizationusing3D InteractiveAnimation”,Communicationsof theACM 36(4),57–71.

Rodenstein,R., Abowd, G. D. & Catrambone,R. (1999),OwnTime: A System for TimespaceManagement,in Conference Companion of the ACM Conferenceon HumanFactors in ComputingSystems(CHI ’99),Pittsburgh,PA, pp.200–201.

Sekey, A. & Tietz, J. (1982), “Text display by saccadicscrolling”, Visible Language 16, 62–76.

Ware,C., Bonner, J.,Cater, R. & Knight, W. (1992),“SimpleAnimationasaHumanInterrupt”,InternationalJournalof Human-ComputerInteraction4(4), 341–348.

Zhao, Q. A. & Stasko, J. T. (2000), What’s Happening?TheCommunityAwarenessApplication,in ConferenceCompanionof theACM Conferenceon HumanFactorsin ComputingSystems(CHI 2000), pp.253–254.