Novel Method to Measure Inference Affordance in Static ...sara/pubs/fabrikant_etal_caj08.pdf ·...

Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

R E F E R E E D P A P E R

Novel Method to Measure Inference Affordance in StaticSmall-Multiple Map Displays Representing Dynamic Processes

Sara Irina Fabrikant1 Stacy Rebich-Hespanha2 Natalia Andrienko3 GennadyAndrienko3 and Daniel R Montello2

1Department of Geography University of Zurich Zurich Switzerland sarageouzhch 2Department of Geography

University of California Santa Barbara Santa Barbara CA USA 3Fraunhofer Institute for Intelligent Analysis and

Information Systems (IAIS) Schloss Birlinghoven 53754 Sankt Augustin Germany

Supported by eye-movement data collected during a controlled experiment on small-multiple map displays a new concept

coined inference affordance aimed at overcoming drawbacks of traditional empirical lsquosuccessrsquo measures when evaluating

static visual analytics displays and interactive visual analytics tools is proposed Then a novel visual analytics research

methodology is outlined to quantify inference affordance taking advantage of the well-known sequence alignment analyses

techniques borrowed from bioinformatics The presented visual analytics approach focuses on information reduction of

large amounts of fine-grained eye-movement sequence data including sequence categorisation and summarisation

INTRODUCTION

Cognitive scientists have attempted to tackle the funda-mental research question of how externalised visualrepresentations (eg statistical graphs organisationalcharts maps animations etc) interact with peoplersquosinternal visualisation capabilities and can facilitate inferenceand decision making (Scaife and Rogers 1996 Simon andLarkin 1987) Experimental research in psychology sug-gests that static graphics can facilitate comprehensionlearning memorisation communication of complex phe-nomena and inference from the depiction of dynamicprocesses (Hegarty 1992 Hegarty and Sims 1994)

The need to better understand the cognitive processesinvolved in using dynamic displays has become moreimportant recently paralleling the exponential growth ofanimation and dynamic graphics to which people are beingexposed in their everyday life (eg virtual-globe viewersgame controllers and weather animations on TV news) Aswith most rapid developments of new technologies thetheory and understanding of novel graphics technology andapplications has lagged behind

As real-time three-dimensional landscape fly-throughsand interactive map animations become ubiquitous withdissemination over the Internet an important question thatremains is how effective the potential increase of informa-tion density in these highly interactive visual forms really isfor (spatial) knowledge construction and decision-makingWe still know very little about how effective novelinteractive graphical data depictions and visual analytics

tools are for knowledge discovery learning and sense-making of dynamic multidimensional processes (Harrowerand Fabrikant 2008) Today a pervasive theme underlyingmany current (geo)visualisation research challenges isthe difficulty of effectively evaluating highly interactivevisualisation tools and complex displays and of identifyingtheir potentially positive influence on exploratory dataanalysis knowledge extraction and learning (Harrower2007)

STATIC AND DYNAMIC DEPICTIONS OF PROCESSES

AND EVENTS

Visual analytics is based on the intuition that highlyinteractive and dynamic depictions of complex and multi-variate databases amplify human capabilities for inferenceand decision making as they facilitate cognitive tasks suchas pattern recognition imagination association andanalytical reasoning (Andrienko and Andrienko 2007Thomas and Cook 2005)

This contention is supported by the congruence principlesuggested by Tversky et al (2002) This principle statesthat well designed external representations such as graphicdisplays show a natural cognitive correspondence instructure and content with the desired structure andcontent of the internal (mental) representation (ie theappropriate analytical inference) For example animationscongruently depict the concept of time and change with

The Cartographic Journal Vol 45 No 3 pp 201ndash215 Geovisualisation Special Issue August 2008 The British Cartographic Society 2008

DOI 101179000870408X311396

Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

changing displays over time so it seems obviousthat humans will have less difficulty comprehendingcomplex dynamic processes through well-designed dynamicdisplays

However in a series of publications surveying thecognitive research literature on animated graphics (thatdid not include map animations) Tversky and colleaguesclaim they failed to find benefits to animations forconveying dynamic processes (Betrancourt et al 2000Betrancourt and Tversky 2000 Morrison et al 2000Morrison and Tversky 2001 Tversky et al 2002) Thesecognitive scientists argue that studies reporting a super-iority of animations over static displays had experimentaldesign flaws For example additional interactivity for theanimated sequences violated the information equivalencebetween static and animated graphics However as Krygieret al (1997) suggest interactivity has different intensitiesor modalities with static graphics having the lowestinteraction intensity mdash but not zero Static graphics doafford mental (internal) interactivity A sequence ofstatic graphics (eg small multiples) can be seen as(mentally) interactive in the sense that people canproactively control with their eyes the viewing order ofthe static sequence they can always go back to thebeginning of the sequence and they can choose to studythe sequence at their own pace and in any order they wishGenerally if a static map is presented on a piece of paper (ora small display) it can be rotated andor folded astravellers commonly do with maps for instance Ananimated sequence being slightly more (externally) inter-active when featuring a start stop and rewind button is less(internally) interactive in the sense that the sequence mustbe passively viewed in a pre-defined order This reduction in(internal) interactivity may add cognitive load onto aviewerrsquos working memory thus limiting the animationrsquospotential for facilitating learning (Sweller 1994) It seemsthat this particular hypothesis has not been adequatelyinvestigated in animation studies especially not on dynamicmap displays

Results on (comparative) cartographic experiments areinconclusive partly because it depends on how lsquobetterrsquo isdefined and measured In some experiments that comparemap animations with static small-multiple displays partici-pants answer more quickly but not more accurately withanimations (Koussoulakou and Kraak 1992) In otherexperiments they take longer and answer fewer questionsmore accurately (Cutler 1998) or the time it takes toanswer the question does not relate to accuracy at all(Griffin et al 2004) In a mostly qualitative map animationstudy Slocum et al (2004) found that map animations andsmall multiples are best used for different tasks The formerare more useful for inspecting the overall trend in time-series data the latter for comparisons of various stages atdifferent time steps

We argue in the next sections that the question ofwhether animations are superior to static maps is not onlyan ill-posed question but also an unanswerable one Asgeovisualisation designers we should instead be interestedin finding out how highly interactive visual analytic displayswork identifying when they are successful and why(Fabrikant 2005)

INFORMATION EQUIVALENCE VS INFERENCE

AFFORDANCE

There is a fundamental problem with these kinds ofcomparative studies One the one hand it seems obviousthat well-designed animations need to be compared to well-designed static displays Very often however the stimuliare not prepared by design experts thus differences that arefound might be attributed simply to bad design choices Toprecisely identify differences in the measures of interest thedesign of the animations and small multiples to becompared requires tight experimental control to the extentthat it might make a comparison meaningless Tversky andcolleagues (citations above) argue that experimental studiesreporting advantages of animation over static displayslacked equivalence between animated and static graphicsin content or experimental procedures For example theyargue that animations show more information than staticgraphics because only the coarse segments are portrayed instatic graphics whereas animations portray both the coarseand fine segments of change This presupposes the notionthat animations and static displays can be informationallyequivalent a term coined by Simon and Larkin (1987) toexpress the idea that all information encoded in onerepresentation is also inferable from the other and viceversa In other words can all information available in a mapanimation be used to build an informationally equivalentsmall-multiple map display (SMMD) We argue that well-designed animations are inherently different from well-designed small multiples and should afford different kinds ofinformation extraction specifically amenable to the desiredinference modes and to the specific knowledge constructiontasks Making an animation equivalent in informationcontent to a small-multiple display in order to achievegood experimental control for comparisons may actuallymean degrading its potential power for certain tasksAnimations are not simply a sequence of static smallmultiples (Harrower 2003) Effective static displays depictconfigural information (ie states) or static snapshots(freeze frames) of events and processes For example in astatic time series of choropleth maps a seven-class solutionmight effectively display a complex pattern of change In ananimation using the same data however it might not bewise to portray maps with seven classes as viewers wouldnot be able to apprehend that much detail and keep it activein working memory when viewing a non-interactiveanimation If the goal of the animation is to depict changethen good design should focus particularly on emphasisingchange most effectively for example by applying smoothtransitions between display frames to avoid potentialchange blindness (Rensink et al 1997) something thatstatic displays are inherently unable to achieve (Fabrikantand Goldsberry 2005)

With their term computational equivalence Simon andLarkin (1987) suggest a much more useful concept forassessing the effectiveness of graphic representationsespecially when comparing different visual-analytics displaysthat typically afford different modes of interactions forinference making Two representations are said to becomputationally equivalent when any inference that is easilyand quickly drawn from the encoded information in onedisplay can be easily and quickly drawn from the other

202 The Cartographic Journal

Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

(informationally equivalent) display and vice versa Simonand Larkin (1987) do not specify what easily and quicklymean They suggest that the advantages of graphics overtext in general are computational not because they containmore information but because the presentation of theinformation can support extremely useful and efficient(computational) inference-making processes This suggeststhat computational equivalence might be the more usefulconcept to use for comparison of complex graphics andinteractive visual tools with varying degrees and differingkinds of interaction affordances We argue that computa-tional equivalence is inherently linked to informationequivalence and cannot be easily disentangled

When comparing displays that afford different interactionmodes there seems to be a trade-off between informationalequivalence and computational equivalence To compare anon-interactive choropleth map animation in a fair way to asmall-multiples display (eg with seven classes) theinformational equivalence of the two displays has to beviolated (ie the choropleth map classes must be reducedfor the animation) because the limited interaction possi-bilities afforded by the animation leads to greater cognitiveload which affects its computational performance

To better capture the effectiveness of a highly interactiveand dynamic visual analytics display we instead propose touse the concept of inference affordance that integrates bothinformational equivalence (amount and quality of content)and computational equivalence (quality and efficiency ofinferences based on design) Effective visual analytics is notonly about successfully extracting the content of the

encoded data but also about supporting different kindsof knowledge construction and inference-making processesthrough various cognitively adequate inference affordances

What this discussion has not touched on so far is thecomplex issue of individual differences including priorknowledge and training for visual-inference makingElsewhere it has been suggested that bottom-up (egperceptual) and top-down (ie cognitive) processes areinterlinked (Kriz and Hegarty 2007) In other words itdoes not just suffice to provide well designed graphics andvisual tools and hope for success but users also need tohave an established base capacity for recognising anddeciding which tool to select when how and for whataim and purpose (Lowe 1999)

EYE-MOVEMENT ANALYSIS AND INFERENCE

AFFORDANCE IN VISUAL ANALYTICS DISPLAYS

For over a century psychologists and other researchers haverecorded human eye movements mostly on static displaysto learn how people read texts and view various staticgraphic displays such as advertisements works of artcharts diagrams etc (Wade and Tatler 2005) Peoplemove their eyes so that the fovea (the vision centre withhighest acuity) is directed toward what they wish to attendto mdash to visually process at the highest possible detail(Rayner 1992) Continual ongoing eye movements arecalled saccades Saccades are interrupted by eye fixationsphases where our eyes are relatively static focusing on andattending to an object of interest

Figure 1 A gaze plot including eye fixations and saccades overlain onto a small-multiple map stimulus

Measuring Inference Affordance in Static Small-Multiple Map Displays 203

Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

The assumption that peoplersquos centre of visual attention istightly linked with where they look during scene viewinghas been recognised by a number of cognitive scientistswho utilise eye-movement records to infer knowledgeabout the cognitive processes involved in various visualcognition tasks (Rayner 1998) An advantage of eye-movement recordings compared with traditional empiricaldata collection (eg questionnaires interviews etc) is thatthey provide relatively unobtrusive real-time measures of(overt) visual and cognitive information processing beha-viour (Henderson and Hollingworth 1998)

Cartographers have utilised eye-movement recording asearly as the 1970s to investigate how people look at staticmaps (Steinke 1987) Cartographers were particularlyinterested in improving the design of their map productsbased on eye-movement research thereby creating betterand more user-friendly products (Montello 2002) Afterincreased interest in eye-movement studies with mapsduring the 1970s and early 1980s the collection of eye-movement data in academic cartography has almostdisappeared Montello (2002) suggested that one of thefactors might have been that eye-movement analysis tendedto be very expensive and notoriously difficult to performand analyze Other critical voices argued that this kind ofdata collection did not tell mapmakers anything they didnot already know and thus did not warrant the extra effortand expense Another reason for the limited success of eye-movement studies in cartography may have been thatresearchers tended to focus their studies on where peoplelooked without getting at the how and why of map readingthat generated the viewing pattern for particular map tasks(Brodersen et al 2002)

However especially when evaluating visual analyticstools where classic evaluation measures such as accuracyof response and time to respond might fall short (because ofthe entanglement of computational and informationalequivalence) eye-movement behaviour analysis shouldprovide additional insight into assessing the hard-to-measure concept of inference affordance proposed earlier

SMALL-MULTIPLE DISPLAYS

Small-multiple displays (SMD) a graphic display typenamed and popularised by Tufte (1983) had gainedpublic attention for their potential to uncover complexdynamic processes at least since Muybridge introducedstop-action photography to study galloping horses in thelate 19th century (Encyclopedia Britannica 2008) Earlyon cartographers achieved a high level of sophistication inrepresenting complex dynamic spatio-temporal realitythrough the power of abstraction in the form of a seriesof static two-dimensional maps which Bertin (1967) callsthe lsquocollection of maps with one map characteristicrsquo Morerecently small multiples have resurfaced in highly inter-active and dynamic visual analytics displays allowing theuser to reorder brush and otherwise manipulate thedepicted spatio-temporal data on the fly (MacEachrenet al 2003) The informational effectiveness of a staticsmall-multiple display compared with an animation dependson using the appropriate number of small multiples and

choosing the key events that is it depends on how manyand which of the key events (macro steps) are selected todiscretely represent the continuous and dynamic process (ofmicro steps) Well-designed small-multiple displays depictthe most thematically relevant (pre-selected) key eventsand unlike non-interactive animations allow viewers toinspect the display at their own pace and viewing order Theinference affordance is directly related to the arrangementof the small multiples in the display which in turn might bedetermined by the inference tasks the display shouldsupport

EXPERIMENT

Utilising the eye-movement data collection method to trackpeoplersquos viewing behaviour we investigated the role ofinference affordance in static small multiple map displays(SMMD) The hypothesis at the outset is that if SMMDsand map animations are informationally equivalent onewould expect to find that viewersrsquo knowledge gained fromSMMDs would emphasise information about macro stepsand the configurational aspects of the display (ie its visuo-spatial properties) more than on change (ie micro steps)as claimed by cognitive scientists in the work cited aboveMoreover in terms of computational equivalence peoplersquosgazes would have to move sequentially from one map to thenext in the SMMD matching the sequential viewing orderusers are locked into in non-interactive animations regard-less of the knowledge-construction or inference-makingtasks

In this paper we report on experimental results that werecollected on SMMDs in isolation without comparing theresults to a map animation condition As argued earlier webelieve the comparative lsquowhat-is-betterrsquo question ofSMMDs vs animations to be unanswerable directly bymeans of a controlled experiment The results reported inlater sections will mostly focus on the computationalaspects (inference events and process) of the inferenceaffordance measure proposed earlier and specifically presenta novel analysis approach to assess eye-movement behaviourfor this purpose We first present inference making patternsof individuals (exemplars) and then discuss methods foraggregation and summarisation While we chose smallmultiple map displays as one typical static depiction methodfor representing a spatio-temporal process the presentedevaluation methodology is generic enough to be applicableto any spatial display (static or interactive) that may beproduced to support spatio-temporal inference making

In a controlled experiment we first asked noviceparticipants (n534) tested individually to study a seriesof small-multiple maps showing monthly ice creamconsumption for an average year for different states in afictitious country and then answer a number of questionsabout these maps (Figure 2)

The test questions required participants to make infer-ences varying in type and complexity test questionconstituted a within-subject independent variable Formore complex inference questions we asked participantsto explain their answers Digital audio-recordings ofparticipantsrsquo verbal statements permit joint analyses with


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

the accuracy of their responses (inference quality measure)and their eye-movement recordings (inference processmeasure) all dependent variables

Figure 3 shows a test participantrsquos eye-movement pat-terns overlain on two identical SMMDs but during twodifferent inference-making tasks The graduated circlesshow eye fixation durations (the larger the circle thelonger the fixation) and the connecting lines representsaccades rapid eye movements between fixations Thepassage of time is represented in both panels withtransparency that is the more opaque the saccades andfixations are the more recent

In Figure 3a the task is to gain an overall impression ofthe SMMD and verbally describe the patterns that arediscovered during its visual exploration In contrast inFigure 3b the task was to specifically compare two mapswithin the SMMD When a map-use context requires a userto compare items in a time series (across time space orattribute) the non-interactive animations (locking a viewerinto a pre-defined sequence) will always add cognitive loadas the viewer will have to wait and remember the relevantinformation until the respective comparative displays comeinto view When animating the collected gaze tracks onecan clearly see that the viewer is not exploring the display inthe implied sequence of the small-multiple arrangementbut going back and forth between the maps several times orjumping between different rows of maps Ironically this isone of our first success stories of the power of visualanalytics The interactive animation of eye-movementbehaviour in the visual analytics tool we developed to

analyze eye movements turned out to be far superior foranalyzing our collected data than the static gaze plotdisplays To summarise The SMMD allows the user tofreely interact with the data in the viewing sequence theydeem necessary for the task This is one example of violatingthe computational equivalency of SMMDs and non-interactive animations in order to affect their informationalequivalence

Figure 4 depicts eye-movement behaviours during twomagnitude-comparison tasks involving two maps at twodifferent time steps in an SMMD In Figure 4a a user isasked to compare ice-cream consumption rates between themonths of May and August and in Figure 4b between themonths of January and February The gaze patterns revealsthat only the information contained in those specific twomaps is investigated to answer the test question Theremaining small maps are completely ignored This suggeststhat for this particular task non-interactive animationswould indeed not be informationally equivalent toSMMDs as they would force a user to see much moreinformation than is relevant for the inference task Tomaximise inference affordance one could reduce theoverload of presented information by offloading it ie bymaking the animation interactive

Moreover these data further reveal that the design of theSMMD is an integral part of the inference affordanceproblem which was not investigated in the cognitive workreviewed above The particular design of the SMMDstimulus shown in Figure 4 seems to be ideal for detectingdetailed change information between the adjacent months

Figure 2 Sample small-multiple test stimulus with a general pattern detection question


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

of January and February but not between May and Augustas these maps are far apart In this case adding interactivityto an SMMD might alleviate the reduced computationalpower produced by a suboptimal layout (eg by being ableto move maps) as the arrangement of the SMMDs cannotbe manipulated in the static version A predefined layoutmight make this kind of inference task particularly difficult

The significantly different viewing behaviours depictedsuggest that small-multiple displays cannot generally becomputationally or informationally equivalent to non-interactive animations the computational and informa-tional equivalence of displays certainly depends on the taskthe information extraction goal and the decision-makingpurpose

VISUAL ANALYTICS OF EYE-MOVEMENT PATTERNS

Eye-movement research typically yields a tremendousamount of fine-grained behavioural data both spatiallyand temporally at very high levels of detail For example a30-min recording will yield about 90 000 records at atemporal resolution of 50 Hz (50 gaze pointsseconds)Raw eye data are seldom used directly they need to be

filtered based on a duration threshold an empiricalconstruct designed to better separate lsquowhere people lookrsquofrom where people cognitively lsquoprocess seen informationrsquo

Data typically contained in an eye-movement record aredepicted in Figure 5 A numeric identifier (lsquoMaprsquo) links theeye record with a particular graphic stimulus As stimuli areoften randomised to avoid potential ordering biases asecond identifier (lsquoSlidersquo) indicates the order in which thestimuli have been seen X- and Y-locations of the eyefixations are stored in display (screen) coordinatesTemporal information includes a time stamp released by atrigger event (lsquoStartrsquo in seconds) and a fixation duration(lsquoDurationrsquo in milliseconds) Additionally investigators canidentify areas of interest (AOI) in a stimulus that getrecorded as lsquointeractionrsquo events as soon as the eyes haveentered that particular AOI zone (lsquoTop Zonesrsquo column)Other user interactions such as mouse or keyboardmanipulations can be recorded as well and linked to gazetracks Based on available theory (Irwin 2004 Henderson2007) only gaze points above 100 milliseconds have beenretained for further analysis of the SMMD

To analyze these large datasets cross-fertilisation withGISciencegeovisualisation seems appropriate on severallevels Eye-movement software and other related time-based observational data-analyses packages typically do not

Figure 3 Task dependent viewing behaviour of two identicalSMMD stimuli

Figure 4 Gaze plots for two different inference tasks affected bylayout design


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

include any spatial-analytical tools to analyze or summariselocation-based data Visual analytics methods are missingentirely Herein lies a great opportunity for the GISciencegeovis community to reach out to other disciplines and helpin the analysis of eye-movement recordings The amountand complexity of the collected eye-movement recordingsrequired us to think carefully about how to make sense ofthe empirical data sets For this reason we developed alightweight visual analytics interface (using Adobe Flash)that allows us quickly to visually explore the collected eye-movement data (play back filter visually summarise)gaining first insights on individual behaviours beforerunning any hypothesis-testing analyses Figure 6 belowdepicts the Flash-based graphical user interface of oureyeview software1 developed as a proof-of-concept tool anddescribed in Grossmann (2007)

The system allows one to load text-based eye-movementrecords as shown in Figure 5 above and filter data basedon time attribute or location including more advancedspatial analyses the subset can then be displayed overlain ona graphic stimulus The most useful feature of this systemfor this research simply turned out to be the play-back andsequencing function which created animations of the eye-movement sequences

SEQUENCE ANALYSIS (SA)

Visual analytics methods and data exploration tools for theeffective depiction and analysis of time-referenced spatial

data sets at high resolution have recently gained newattention (Laube and Purves 2006 Andrienko andAndrienko 2007) Location changes order of eventssmooth pursuits etc have become new foci of process-based research using spatio-temporal moving-objects data-bases of various kinds and at different scales (ie movinghumans over a year or moving eyeballs in milliseconds)(Laube et al 2007) Very large databases containingmoving object behaviours are generated in abundance as aresult of various tracking devices available today (ie LBSGPS-enabled cell phones eye trackers for market researchand in psychology)

Sequence analysis (SA) is one promising approach to theanalysis of process event and change rather than the moretraditional analysis of objects and their configurationsincluding location (Abbott 1990) Depending on theresearch question and the collected sequence data differentkinds of SA methods are available As for traditionalstatistical analysis it is important first to distinguishcontinuous from categorical sequence data Moreovernon-recurrent sequences of equal length (in which eventscannot repeat in the sequence) or recurrent sequences withunequal lengths (containing sub-sequences with eventrepetitions) require different SA methods One also needsto consider if states within a sequence are dependent oneach other or if whole sequences are dependent on eachother

For example well-known Markov-type sequence analysesaim at modelling a process that reproduces a certain pattern(Hacisalihzade et al 1992) Markov analyses focus oninternal sequence dependencies These are modelled as astochastic process by means of a lsquostep-by-steprsquo computa-tion based on a transition probability matrix There areseveral reasons why these kinds of models are not suitable

Figure 5 Extract of a processed eye-movement data set

1The software was developed at the Geographic Information Visualization and

Analysis (GIVA) Unit of the Department of Geography at the University of Zurich

Switzerland


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

for our work For one in exploratory work the process isoften unknown thus empirical data cannot easily becompared with an idealised (theoretical) model sequenceSecond Markov models assume that the likelihood of anevent occurring is conditional only on the immediatepredecessor event which is too limiting for modellinginference behaviour based on eye movements In our workwe do not know what the process is at the outset We needfirst to identify patterns hidden in the large eye-movementdata collections by summarising and comparing variousinference-making histories as a whole We are also inter-ested in identifying similarities across people tasks andmodalities that might tell us something about theunderlying process being affected by varying inferenceaffordances

Sequence alignment methods discussed in the nextsection seem particularly promising for our purposebecause they are good at identifying prototypical inferencepatterns by means of summarising and categorising eye-movement sequences (ie chains of attention events)across people and tasks

SEQUENCE ALIGNMENT ANALYSIS (SAA)

Sequence alignment analysis (SAA) another technique ofrelevance to us has been indispensable in bio-medicalresearch for uncovering patterns and similarities in vastDNA and protein databases Sequence alignment algo-rithms were developed in biology and computer science inthe 1980s (Sankoff and Kruskal 1983) and respectivesoftware packages became available soon thereafter (egClustalW) On a most general level SAA algorithms

identify similarities between character sequences based onthe frequency and positions of characters representingobjects or events and on character transitions that arenecessary for similarity assessment (Wilson 2006) SAA hasalso become popular in the social sciences (Abbott 1995)including geography (Joh et al 2002 Shoval and Isaacson2007) but has hardly been looked at by the cognitivecommunity working with eye-movement data (West et al2006)

SEQUENCE ALIGNMENT ANALYSIS OF EYE-MOVEMENT

RECORDINGS

We employed the ClustalG software (Wilson et al 1999)to systematically compare and summarise individual infer-ence-making histories collected through eye-movementdata analysis ClustalG is a generalisation of the variousClustal software packages widely used in the life sciences toanalyze gene sequences in DNA and proteins (representedby characters with a limited alphabet) ClustalG has beendeveloped specifically to deal with social-science data thatrequire more complex coding schemes (ie an extendedalphabet) for describing more complex event histories andsocial processes (Wilson et al 1999) The proposed SAAon collected eye-movement data includes a two-stepapproach (1) data reduction of overt inference behaviourby summarisation of collected eye-movement sequences(across people and inference tasks) and (2) categorisationof found behavioural patterns by aggregating similarsequences into groups through cluster analysis The stepscan be applied in any order In the discussion below weinverted the analysis step sequence exemplified for one

Figure 6 Visual analytics interface to depict inference-making behaviour through eye movements


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

inference task with the SMMD (sample data shown inFigure 1)

CATEGORISATION OF EYE-MOVEMENT BEHAVIOUR

As mentioned earlier aside from raw X- Y-coordinates wealso collected fixation sequences based on pre-defined areasof interest (AOI) one area for each map in the SMMD Wepost-processed the AOI data for each test participant andstored categorical character sequences into one ASCII textfile (for one exploratory inference task see Figure 2)Sequences vary considerably in length from about 300words to over 1100 words where a word includes 3-character abbreviations for the months in the depictedSMMD time series (ie lsquoJanrsquo lsquoFebrsquo etc)

The loaded sequences are colour-coded based on themonths of the year One row represents a viewing sequencefor one participant The viewing sequence begins on the lefthand side of Figure 7 at starting position lsquo1rsquo found on thebottom row (x-axis) labelled lsquorulerrsquo One can immediatelysee the winter months cluster at the beginning in coldcolours (blue to purple) followed by the summer months inwarm colours (yellow to brown) Next a multiple align-ment process is carried out based on recommended inputvalues by the ClustalG developers (Wilson et al 1999)The first alignment phase includes a global pairwise-alignment procedure to identify similarities between wholesequences The result is a resemblance matrix that is inputto an unrooted phylogenetic-tree model (Saitou and Nei1987) This tree model (not depicted) represents branchlengths proportional to the estimated sequence uniquenessalong each branch and is subsequently applied to guide themultiple alignment phase Phase two multiple alignment isin essence a series of pairwise alignments following thebranching order of the previously computed tree model

Figure 8 portrays an extract of aligned sequences Onecan see that the JanndashFeb pattern (in blue) is well aligned

followed by gaps where sequences do not align (indicatedin Figure 8 with dashes) and aligned portions of a NovndashDecpattern This pattern suggests that a significant group ofpeople may have treated the temporally adjacent wintermonths as an inference unit but not at the same momentduring the exploration Perhaps this is due to JanndashFeb andNovndashDec months being spatially far away from each otheron the SMMD and people seem to have employed varyingviewing strategies and orders to compare them

The uniqueness information contained in the clusteringtree can be further analyzed to categorise alignedsequences Based on the dendrogram we identified threeclusters One cluster (containing three participants) can becharacterised by viewing behaviour with considerable noisedue to significant eye-tracking signal loss as shown inFigure 9 (most and longest fixations outside the viewingarea in the upper left corner)

The other two clusters are more difficult to analyze bysimply playing back the viewing behaviour or by visuallycomparing the groups of gaze plots For this reason wedecided to employ a powerful geovisual analytics toolkitspecifically targeted for the analysis of movement data(Andrienko et al 2007) Details of the software andprovided analysis routines can be found in Andrienko et al(2007)

SUMMARISATION OF EYE-MOVEMENT BEHAVIOUR

Trying to make sense of gaze data for one single testparticipant on one inference task is already difficult enoughdue to extensive overplotting (as shown in the figuresabove) Trajectory data from Figure 1 shown earlier hasbeen processed with a summarisation method fromAndrienko et al (2007) and the aggregated eye-movementpath for that same participant is visualised in Figure 10

The summarisation analysis depicted in Figure 10bincludes directional information for the trajectories in the

Figure 7 Participantsrsquo eye-movement sequences loaded into ClustalG


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

gaze plot (blue lines with arrows) Thicker lines indicatemore movements The depicted pattern suggests thatthis participant did not divide hisher attention equallyover all maps The first row was investigated morefrequently in both directions and in various spatial intervals

(eg onetwo steps forward onetwo steps backwardsetc) Short vertical lines between rows suggests that theparticipant also chose a spatial viewing strategy that isviewing nearby displays irrespective of the suggested tem-poral sequence Longer trajectories (missing arrowheads)

Figure 8 Subset of aligned sequences

Figure 9 Outlier eye movement sequence due to eye tracking recording problems


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

mean that information below the line was looked at lsquoinpassingrsquo if at all For example the last row includingOctober November and December has comparatively fewfixation locations (see next Figure 11) and were looked atin reverse order from the suggested viewing sequence Tovalidate the summarisation procedure it also helps just tolook at fixation patterns as visualised in Figure 11

The overplotting problem gets exacerbated when tryingto inspect trajectories across all test subjects as shown inFigure 12 below

As Figure 12 shows severe overplotting does not allowone visually to discover anything To identify potentialviewing strategies on a single inference task we summarisedall participant data based on cluster membership discussedearlier identified during phase two of the sequencealignment procedure As mentioned earlier participantsare clustered based on similarities in viewing behaviour (ieviewing sequences) The results of the three summarisationsby participant clusters are displayed in Figure 13

In other words the following discussion of results andconclusions are based on summarisations across all partici-pants Generally the spatial trajectory patterns can bedescribed in terms of completed distances (ie long orshort moves) andor movement headings (ie vertical

horizontal and diagonal moves) The horizontal trajectoriesat the bottom of each panel in Figure 13 are generallyrelated to reading the test question even if the lines are notdisplayed exactly over the respective text portion in theabove displays This visual mismatch is dependent on theaggregation algorithm used Horizontal trajectories withina row of maps suggest that participants are moving theireyes in the suggested temporal sequence Sequentialviewing behaviour is also indicated when horizontaltrajectories are connected by diagonals from the end ofone row of maps to the beginning of the next row belowWhen playing back eye movement behaviours one can seethat diagonal moves are always performed in the forwarddirection while horizontal moves can be both performedforwards and backwards Vertical moves across map rowssuggest two things Firstly longer vertical moves (startingor ending from the question) are performed whenparticipants initially read the test question and then startinspecting the maps or when eyes are returning to the testquestion during the map exploration task Second shortervertical moves within and across map rows indicate spatialexploration behaviours for example when nearby maps areinspected instead of following the suggested temporalarrangement

Visual pattern inspection suggests a couple of distin-guishing features across behavioural clusters lsquoSpatialsearchrsquo behaviour is depicted noticeably in the star-liketrajectory pattern shown in Cluster 1 in Figure 13a(representing 30 of the participants) The centre of thestar is the second map from the left in the centre row Asimilar star pattern is visible in Cluster 3 (8 of theparticipants) and its centre at the same location (ie theJune map) as in Cluster 1 Cluster 2 shown in Figure 13bincludes the largest proportion of participants (62) andfeatures dominantly horizontal trajectories By animatingthe eye movement behaviours for this cluster one can detectthat the horizontal trajectories include forward moves andbacktracking within map rows A participantrsquos summarisedtrajectory exhibiting this kind behaviour is shown inFigure 1 Interestingly the horizontal moves within therows are not only connected with diagonals in Cluster 2but also with vertical lines at respective row ends Wheninspecting these eye movements again by animation one cansee that people combine temporal and spatial searchstrategies The map sequences are looked at in reversetemporal order in the middle row perhaps to increasespatial search efficiency

These empirical findings on static small multiple displayssuggest the following design principles for providingcomputationally equivalent animations Animations shouldnot only provide a play lsquoforwardrsquo button andor lsquoforwardrsquosequencing interactivity but also include backwards anima-tion and reverse sequencing options to provide at leastequally efficient inference affordances compared with smallmultiples Making SMMDs interactive so that users canrearrange the map sequence according to the spatialtemporal or spatio-temporal inference making tasks andrespective knowledge extraction goals can alleviate layoutproblems in static SMMDs

In terms of methodology this research proposes acombined geovisualisation and visual geoanalytics

Figure 10 Effect of data reduction (a) original and (b) sum-marised eye movements


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

approach to better quantify peoplersquos inference makingprocesses from and with visuo-spatial displays Consideringthat eye-movement recordings are location-based they canbe easily imported into an off-the-shelf GIS or as in ourcase a specifically developed visual geoanalytics tool Eyemovements can be displayed and analyzed in more detailwith powerful spatial analytical tools in a similar fashion tothe display and analysis geographic movement dataGeovisualisation methods are helpful for getting firstinsights on inference behaviours of individuals for exampleby simply being able to display gaze plots andor play back

peoplersquos gaze trails over the explored graphic stimuliHighly interactive visual geoanalytics toolkits such asproposed by Andrienko et al (2007) provide an additionalexcellent framework to more efficiently handling massivefine grained spatio-temporal movement data by summaris-ing and categorising groups of behaviours Empirical resultsbased on the methods described earlier can be additionallylinked to the more traditional success measures such as taskcompletion time and accuracy of response For example infuture work we will be exploring the potential relationshipbetween viewing strategies based on identified clustermembership with the quality and speed of response

CONCLUSIONS

A new concept coined inference affordance is proposed toovercome drawbacks of traditional empirical lsquosuccessrsquomeasures when evaluating static visual analytics displaysand interactive tools In doing so we hope to respond tothe ICA Commission on Geovisualisationrsquos third researchchallenge on cognitive issues and usability in geovisualisa-tion namely to develop a theoretical framework based oncognitive principles to support and assess usability methodsof geovisualisation that take advantage of advances indynamic (animated and highly interactive) displays(MacEachren and Kraak 2001) Furthermore a novelresearch methodology is outlined to quantify inferenceaffordance integrating visual geoanalytics approaches withsequence alignment analyses techniques borrowed frombioinformatics The presented visual analytics approach

Figure 11 Fixation pattern of same participant as in Figure 10

Figure 12 Gaze plots for several test participants


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

focuses on information reduction of large amounts of fine-grained eye-movement sequence data including sequencecategorisation and summarisation

Presented inference-making behaviours extracted fromeye movement records provide first support to thecontention that small-multiple displays cannot generally

be computationally or informationally equivalent to non-interactive animations (in contrast to claims by cognitivescientists cited above) the computational and informationalequivalence of displays do depend on the task the informa-tion extraction goal and the decision-making context

By applying the outlined framework to collectedempirical evidence on static small multiple displays wehope to provide a better understanding of how people usestatic small-multiple displays to explore dynamic geographicphenomena and how people make inferences from staticvisualisations of dynamic processes for knowledge con-struction in a geographical context

BIOGRAPHICAL NOTES

Sara Irina Fabrikant is anassociate professor of geo-graphy and head of theGeographic Visualisationand Analysis Unit in theDepartment of Geo-graphy at the Universityof Zurich SwitzerlandHer research interests arein geographic informationvisualisation GIScienceand cognition graphicaluser interface design anddynamic cartography Sheearned a PhD in geogra-

phy from the University of Colorado-Boulder (USA) andan MS in geography from the University of Zurich(Switzerland)

ACKNOWLEDGMENTS

This material is based upon work supported by the USNational Science Foundation under Grant No 0350910and the Swiss National Science Fund No 200021-113745This work would not have happened without the help of anumber of people we would like to thank Scott Prindle andSusanna Hooper for their assistance with data collectiontranscription and coding Maral Tashjian for the stimulidesigns Adeline Dougherty for database design and config-uration and the UCSB students who were willing toparticipate in our research We are indebted to JoaoHespanha for the development of the eyeMAT Matlabtoolbox allowing us to handle complex data calibrationerrors and preprocessing of the raw eye movement data toThomas Grossmann for the development of the eyeviewtool and to Georg Paternoster for his help on sequence datapost-processing Last but not least we are also grateful forMary Hegartyrsquos continued insightful input discussion andbrainstorming since the inception of this project

REFERENCES

Abbott A (1990) A Primer on Sequence Methods OrganisationScience 1(4) 375ndash392

Abbott A (1995) Sequence Analysis New Methods for Old IdeasAnnual Review of Sociology 21 93ndash113

Figure 13 Summarised eye movements across participant clustersbased on viewing behaviour (a) movement cluster 1 (b) movementcluster 2 (c) movement cluster 3


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

Andrienko G Andrienko N and Wrobel S (2007) Visual AnalyticsTools for Analysis of Movement Data ACM SIGKDDExplorations 9(2) 38ndash46

Andrienko N and Andrienko G (2007) Designing Visual AnalyticsMethods for Massive Collections of Movement Data Cartgraphica42(2) 117ndash138

Bertin J (1967) Semiologie Graphique Les Diagrammes ndash lesReseaux ndash les Cartes Mouton Paris

Betrancourt M and Tversky B (2000) Effect of ComputerAnimation on Usersrsquo Performance A Review Le travail Humain63(4) 311ndash330

Betrancourt M Morrison Bauer J and Tversky B (2000) LesAnimations Sont-Elles Vraiment Plus Efficaces RevueDrsquoIntelligence Artificielle 14 149ndash166

Brodersen L Andersen H H K and Weber S (2002) ApplyingEye-Movement Tracking for the Study of Map Perception andMap Design Kort and Matrikelstyrelsen National Survey andCadastre Denmark Copenhangen Denmark

Cutler M E (1998) The Effects of Prior knowledge on ChildrenrsquosAbility to Read Static and Animated Maps Unpublished MSthesis Department of Geography University of South CarolinaColumbia SC

Duchowski (2007) Eye Tracking Methodology Springer BerlinGermany

Encyclopaeligdia Britannica (2008) Muybridge Eadweard (httpwwwbritannicacomebarticle-9054508Eadweard-MuybridgeJan 8 2008)

Fabrikant S I (2005) Towards an Understanding of GeovisualisationWith Dynamic Displays Issues and Prospects ProceedingsAmerican Association for Artificial Intelligence (AAAI) 2005Spring Symposium Series Reasoning with Mental and ExternalDiagrams Computational Modeling and Spatial AssistanceStanford University Stanford CA Mar 21ndash23 2005 6ndash11

Fabrikant S I and Goldsberry K (2005) Thematic Relevance andPerceptual Salience of Dynamic Geovisualisation DisplaysProceedings 22th ICAACI International CartographicConference A Coruna Spain Jul 9ndash16 (CDROM)

Griffin A L MacEachren A M Hardisty F Steiner E and Li B(2004) A Comparison of Animated Maps with Static Small-Multiple Maps for Visually Identifying Space-Time ClustersAnnals of the Association of American Geographers 96(4)740ndash753

Grossmann T (2007) Ansatz zur Untersuchung der Wahrnehmungbei geographischen Darstellungen Ein Werkzeug zur visuellenExploration von Blickregistrierungsdaten Unpublished MasterThesis UNIGIS Program Salzburg

Hacisalihzade S S Stark L W and Allen J S (1992) VisualPerception and Sequences of Eye Movement Fixations AAtochastic Modeling Approach IEEE Transactions on SystemsMan and Cybernetics 22(3) 474ndash481

Harrower M (2003) Designing Effective Animated MapsCartographic Perspectives 44 63ndash65

Harrower M (2007) The Cognitive Limits of Animated MapsCartographica 42(4) 349ndash357

Harrower M and Fabrikant S I (in press) The Role of MapAnimation in Geographic Visualisation In Dodge M Turner Mand McDerby M (eds) Geographic Visualisation ConceptsTools and Applications Wiley Chichester UK pp 49ndash65

Hegarty M (1992) Mental Animation Inferring Motion from StaticDisplays of Mechanical Systems Journal of ExperimentalPsychology Learning Memory and Cognition 18(5) 1084ndash1102

Hegarty M and Sims V K (1994) Individual Differences in MentalAnimation During Mechanical Reasoning Memory andCognition 22 411ndash430

Henderson J M (2007) Regarding Scenes Current Directions inPsychological Science 16 219ndash222

Henderson J M and Hollingworth A (1998) Eye MovementsDuring Scene Viewing An Overview In Underwood G (ed)Eye Guidance in Reading and Scene Perception Eye Guidancewhile Reading and While Watching Dynamic Scenes ElsevierOxford UK 269ndash293

Irwin E (2004) Fixation Location and Fixation Duration as Indicesof Cognitive Processing In Henderson J M and Ferreira F(eds) The Integration of Language Vision and Action Eye

Movements and the Visual World Psychology Press New YorkNY 105ndash134

Joh C-H Arentze T Hofman F and Timmermans H (2002)Activity Pattern Similarity A Multidimensional SequenceAlignment Method Transportation Research Part B 36 385ndash403

Koussoulakou A and Kraak M J (1992) Spatio-temporal Maps andCartographic Communication The Cartographic Journal 29101ndash108

Kriz S and Hegarty M (2007) Top-down and Bottom-upInfluences on Learning from Animations International Journalof Human-Computer Studies 65 911ndash930

Krygier J B Reeves C DiBiase D and J Cupp J (1997)Multimedia in Geographic Education Design Implementationand Evaluation Journal of Geography in Higher Education21(1) 17ndash39

Laube P and Purves R (2006) An Approach to Evaluating MotionPattern Detection Techniques in Spatio-Temporal DataComputers Environment and Urban Systems 30(3) 347ndash374

Laube P Dennis T Forer P and Walker M (2007) MovementBeyond the Snapshot ndash Dynamic Analysis of Geospatial LifelinesComputers Environment and Urban Systems 31(5) 481ndash501

Lowe R K (1999) Extracting Information from an Animationduring Complex Visual Learning European Journal ofPsychology of Education 14(2) 225ndash244

MacEachren A M and Kraak M-J (2001) Research Challenges inGeovisualisation Cartography and Geographic InformationScience 28(1) 13ndash28

MacEachren A M Dai X Hardisty F Guo D and D L (2003)Exploring High-D Spaces with Multiform Matrices and SmallMultiples Proceedings IEEE Symposium on InformationVisualisation Seattle WA Oct 19ndash24 2005 (CDROM)

Montello D R (2002) Cognitive Map-Design Research in the 20thCentury Theoretical and Empirical Approaches Cartography andGeographic Information Science Special Issue on The Historyof Cartography in the 20th Century 29(3) 283ndash304

Morrison J B and Tversky B (2001) The (in)effectiveness ofAnimation in Instruction Proceedings Jacko J and Sears A(eds) Extended Abstracts of the ACM Conference on HumanFactors in Computing Systems Seattle WA 377ndash378

Morrison J B Betrancourt M and Tverksy B (2000) AnimationDoes it Facilitate Learning Proceedings Papers from the 2000AAAI Spring Symposium Smart Graphics 53ndash60

Rayner K (ed) (1992) Eye Movements and Visual CognitionScene Perception and Reading Springer Verlag New York NY

Rayner K (1998) Eye Movements in Reading and InformationProcessing 20 Years of Research Psychological Bulletin 124(3)372ndash422

Rensink R A OrsquoRegan J K and Clark J J (1997) To See or Notto See The Need for Attention to Perceive Changes in ScenesPsychological Science 8 368ndash373

Saitou N and Nei M (1987) The Neighbor-Joining Method ANew Method for Reconstructing Phylogenetic Trees MolecularBiology and Evolution 4 406ndash425

Sankoff D and Kruskal J (1983) Time Warps String Edits andMacromolecules The Theory and Practice of SequenceComparision Addison-Wesley Reading MA

Scaife M and Rogers Y (1996) External Cognition How DoGraphical Representations Work International Journal ofHuman-Computer Studies 45 185ndash213

Shoval N and Isaacson M (2007) Sequence Alignment as a Methodfor Human Activity Analysis in Space and Time Annals of theAssociation of American Geographers 92(2) 282ndash297

Simon H A and Larkin J H (1987) Why a diagram is (sometimes)worth ten thousand words Cognitive Science 11 65ndash100

Slocum T A Sluter R S Kessler F C and Yoder S C (2004) AQualitative Evaluation of MapTime A Program for ExploringSpatiotemporal Point Data Cartographica 39(3) 43ndash68

Steinke T R (1987) Eye Movement Studies in Cartography andRelated Fields Cartographica 24(2) 40ndash73

Sweller J (1994) Cognitive Load Theory Learning Difficulty andInstructional Design Learning and Instruction 4 295ndash312

Thomas J J and Cook K A (2005) Illuminating the Path Researchand Development Agenda for Visual Analytics IEEE PressRichland WA


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

Tufte E (1983) The Visual Display of Quantitative InformationGraphics Press Cheshire Connecticut

Tversky B Bauer Morrison J and Betrancourt M (2002)Animation Can it Facilitate International Journal of Human-Computer Studies 57 247ndash262

Wade N and Tatler B (2005) The Moving Tablet of the Eye Theorigins of modern eye movement research Oxford UniversityPress Oxford UK

West J Haake A R Rozanski E P and Karn K S (2006)eyePatterns Software for Identifying Patterns and Similarities

Across Fixation Sequences Proceedings 2006 Symposium onEye tracking Research amp Applications San Diego CA Mar 27ndash292006 149ndash154

Wilson C (2006) Reliability of Sequence Alignment Analysis of SocialProcesses Monte Carlo tests of ClustalG software Environmentand Planning A 38 187ndash204

Wilson C Harvey A and Thompson J (1999) ClustalG Softwarefor Analysis of Activities and Sequential Events ProceedingsLongitudinal Research in Social Sciences A Canadian FocusWindermere Manor London Ontario Canada Oct 25ndash27 1999


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

changing displays over time so it seems obviousthat humans will have less difficulty comprehendingcomplex dynamic processes through well-designed dynamicdisplays

However in a series of publications surveying thecognitive research literature on animated graphics (thatdid not include map animations) Tversky and colleaguesclaim they failed to find benefits to animations forconveying dynamic processes (Betrancourt et al 2000Betrancourt and Tversky 2000 Morrison et al 2000Morrison and Tversky 2001 Tversky et al 2002) Thesecognitive scientists argue that studies reporting a super-iority of animations over static displays had experimentaldesign flaws For example additional interactivity for theanimated sequences violated the information equivalencebetween static and animated graphics However as Krygieret al (1997) suggest interactivity has different intensitiesor modalities with static graphics having the lowestinteraction intensity mdash but not zero Static graphics doafford mental (internal) interactivity A sequence ofstatic graphics (eg small multiples) can be seen as(mentally) interactive in the sense that people canproactively control with their eyes the viewing order ofthe static sequence they can always go back to thebeginning of the sequence and they can choose to studythe sequence at their own pace and in any order they wishGenerally if a static map is presented on a piece of paper (ora small display) it can be rotated andor folded astravellers commonly do with maps for instance Ananimated sequence being slightly more (externally) inter-active when featuring a start stop and rewind button is less(internally) interactive in the sense that the sequence mustbe passively viewed in a pre-defined order This reduction in(internal) interactivity may add cognitive load onto aviewerrsquos working memory thus limiting the animationrsquospotential for facilitating learning (Sweller 1994) It seemsthat this particular hypothesis has not been adequatelyinvestigated in animation studies especially not on dynamicmap displays

Results on (comparative) cartographic experiments areinconclusive partly because it depends on how lsquobetterrsquo isdefined and measured In some experiments that comparemap animations with static small-multiple displays partici-pants answer more quickly but not more accurately withanimations (Koussoulakou and Kraak 1992) In otherexperiments they take longer and answer fewer questionsmore accurately (Cutler 1998) or the time it takes toanswer the question does not relate to accuracy at all(Griffin et al 2004) In a mostly qualitative map animationstudy Slocum et al (2004) found that map animations andsmall multiples are best used for different tasks The formerare more useful for inspecting the overall trend in time-series data the latter for comparisons of various stages atdifferent time steps

We argue in the next sections that the question ofwhether animations are superior to static maps is not onlyan ill-posed question but also an unanswerable one Asgeovisualisation designers we should instead be interestedin finding out how highly interactive visual analytic displayswork identifying when they are successful and why(Fabrikant 2005)

INFORMATION EQUIVALENCE VS INFERENCE

AFFORDANCE

There is a fundamental problem with these kinds ofcomparative studies One the one hand it seems obviousthat well-designed animations need to be compared to well-designed static displays Very often however the stimuliare not prepared by design experts thus differences that arefound might be attributed simply to bad design choices Toprecisely identify differences in the measures of interest thedesign of the animations and small multiples to becompared requires tight experimental control to the extentthat it might make a comparison meaningless Tversky andcolleagues (citations above) argue that experimental studiesreporting advantages of animation over static displayslacked equivalence between animated and static graphicsin content or experimental procedures For example theyargue that animations show more information than staticgraphics because only the coarse segments are portrayed instatic graphics whereas animations portray both the coarseand fine segments of change This presupposes the notionthat animations and static displays can be informationallyequivalent a term coined by Simon and Larkin (1987) toexpress the idea that all information encoded in onerepresentation is also inferable from the other and viceversa In other words can all information available in a mapanimation be used to build an informationally equivalentsmall-multiple map display (SMMD) We argue that well-designed animations are inherently different from well-designed small multiples and should afford different kinds ofinformation extraction specifically amenable to the desiredinference modes and to the specific knowledge constructiontasks Making an animation equivalent in informationcontent to a small-multiple display in order to achievegood experimental control for comparisons may actuallymean degrading its potential power for certain tasksAnimations are not simply a sequence of static smallmultiples (Harrower 2003) Effective static displays depictconfigural information (ie states) or static snapshots(freeze frames) of events and processes For example in astatic time series of choropleth maps a seven-class solutionmight effectively display a complex pattern of change In ananimation using the same data however it might not bewise to portray maps with seven classes as viewers wouldnot be able to apprehend that much detail and keep it activein working memory when viewing a non-interactiveanimation If the goal of the animation is to depict changethen good design should focus particularly on emphasisingchange most effectively for example by applying smoothtransitions between display frames to avoid potentialchange blindness (Rensink et al 1997) something thatstatic displays are inherently unable to achieve (Fabrikantand Goldsberry 2005)

With their term computational equivalence Simon andLarkin (1987) suggest a much more useful concept forassessing the effectiveness of graphic representationsespecially when comparing different visual-analytics displaysthat typically afford different modes of interactions forinference making Two representations are said to becomputationally equivalent when any inference that is easilyand quickly drawn from the encoded information in onedisplay can be easily and quickly drawn from the other


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty







EXPERIMENT






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Switzerland


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty







RECORDINGS




Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty














Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty



CONCLUSIONS





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty







EXPERIMENT






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Switzerland


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty







RECORDINGS




Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty














Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty



CONCLUSIONS





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty







EXPERIMENT






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Switzerland


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty







RECORDINGS




Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty














Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty



CONCLUSIONS





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Switzerland


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty







RECORDINGS




Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty














Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty



CONCLUSIONS





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Switzerland


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty







RECORDINGS




Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty














Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty



CONCLUSIONS





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Switzerland


Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty







RECORDINGS




Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty














Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty



CONCLUSIONS





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty







RECORDINGS




Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty














Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty



CONCLUSIONS





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty














Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty



CONCLUSIONS





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty






Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty



CONCLUSIONS





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty











Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty



CONCLUSIONS





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty



CONCLUSIONS





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty





BIOGRAPHICAL NOTES



ACKNOWLEDGMENTS


REFERENCES





Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty

















































Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Pub

lishe

d by

Man

ey P

ublis

hing

(c)

The

Brit

ish

Car

togr

aphi

c S

ocie

ty









Date post:	22-Mar-2018
Category:	Documents
Upload:	trinhthuan
View:	218 times
Download:	3 times

Novel Method to Measure Inference Affordance in Static ...sara/pubs/fabrikant_etal_caj08.pdf ·...

Documents