This article was downloaded by [University Of Maryland]On 13 December 2011 At 0513Publisher Taylor amp FrancisInforma Ltd Registered in England and Wales Registered Number 1072954 Registered office Mortimer House37-41 Mortimer Street London W1T 3JH UK

Behaviour amp Information TechnologyPublication details including instructions for authors and subscription informationhttpwwwtandfonlinecomloitbit20

The promise of zoomable user interfacesBenjamin B Bederson aa Human-Computer Interaction Lab Computer Science Department University of Maryland3171 AV Williams Building College Park MD 20742 USA

Available online 30 Jun 2011

To cite this article Benjamin B Bederson (2011) The promise of zoomable user interfaces Behaviour amp InformationTechnology 306 853-866

To link to this article httpdxdoiorg1010800144929X2011586724

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use httpwwwtandfonlinecompageterms-and-conditions

This article may be used for research teaching and private study purposes Any substantial or systematicreproduction redistribution reselling loan sub-licensing systematic supply or distribution in any form toanyone is expressly forbidden

The publisher does not give any warranty express or implied or make any representation that the contentswill be complete or accurate or up to date The accuracy of any instructions formulae and drug doses shouldbe independently verified with primary sources The publisher shall not be liable for any loss actions claimsproceedings demand or costs or damages whatsoever or howsoever caused arising directly or indirectly inconnection with or arising out of the use of this material

The promise of zoomable user interfaces

Benjamin B Bederson

Human-Computer Interaction Lab Computer Science Department University of Maryland3171 AV Williams Building College Park MD 20742 USA

(Received 1 September 2010 final version received 21 April 2011)

Zoomable user interfaces (ZUIs) have received a significant amount of attention in the 18 years since they wereintroduced They have enjoyed some success and elements of ZUIs are widely used in computers today althoughthe grand vision of a zoomable desktop has not materialised This paper describes the premise and promise of ZUIsalong with their challenges It describes design guidelines and offers a cautionary tale about research andinnovation

Keywords zoomable user interfaces zooming user interfaces multi-scale information visualisation

1 Introduction

The essential problem that zoomable user interfaces(ZUIs) aim to solve is a fundamental one ndash that there ismore information than fits on the screen The commonsolutions to this problem are roughly scrollinglinking and searching along with denser representa-tions (ie information visualisation) Zooming likefisheye displays (Furnas 1986) is an instance of thelatter ndash a kind of information visualisation that aims totake advantage of human spatial perception andmemory ZUIs place documents in 2D space at anysize enabling (and requiring) animated spatial naviga-tion to move among documents

11 What is a ZUI

Before we go further let us think a little about what itmeans to be a ZUI Many applications include somekind of visual scaling functionality Modern filebrowsers let users control the size of icons Webbrowsers word processors image editors and in factalmost all full-featured document editors and browserslet the user control the magnification of the documentMany let the user zoom far enough out to see smallthumbnails of all the pages of even long documents onmodest size screens However that kind of simplescaling is outside the scope of this paper

This paper defines ZUIs to be those systems thatsupport the multi-scale and spatial organisation of andmagnification-based navigation among multiple

documents or visual objects (examples in Figures 1and 2) Admittedly there is a grey area where it mightnot always be clear whether a particular application isa ZUI according to this definition For example aword processor or document viewer that lets you zoomout and see thumbnails of pages laid out in a gridminimally meets that definition However the pagesare really elements of a whole and not movable inspace individually On the other hand a viewer that letyou zoom out see and move arbitrary numbers ofseparate documents even that were one page eachwould count as a full ZUI according to this definition

According to Cockburn et alrsquos (2008) survey ofapproaches to fitting information on the screen ZUIsdisplay information that is temporally separated Theessence of this approach is that the user moves throughspace and builds up a spatial model of the informationin their head This is distinguished from spatialseparation (found in overview thorn detail interfacessuch as those found in maps) and focus thorn context orlsquofisheyersquo distortion such as that found in Applersquos OS XDock (2010) and with the tabular approaches ofTableLens (Rao and Card 1994) and DateLens(Bederson et al 2004a)

12 Why ZUIs excite people

Based on my own experience and analysis of theliterature I have identified three key characteristicsthat have attracted peoplersquos attention over the years

Email wwwcsumdedubederson bedersoncsumdedu

Behaviour amp Information Technology

Vol 30 No 6 NovemberndashDecember 2011 853ndash866

ISSN 0144-929X printISSN 1362-3001 online

2011 Taylor amp Francis

httpdxdoiorg1010800144929X2011586724

httpwwwtandfonlinecom

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

The promise of ZUIs comes largely from the followinggeneral expectations

121 ZUIs are engaging

The animation is visually attention grabbing It takesadvantage of human visual perception abilities Peoplecan process lsquovisual flowrsquo (Ware 2004) preconsciouslyso it feels easy to build a mental map of theinformation

122 ZUIs are visually rich

There are more degrees of freedom to visually structureobjects and thus they offer the potential for great

creative expression This was identified by Perlin andFox in their original paper on Pad with their exampleof a branching tree story (Perlin and Fox 1993)

123 ZUIs offer the lure of simplicity

The fact that you find information by looking for it ina place implies a promise of simplicity that will solveour organisational and information retrieval problemsThe overview one sees when zoomed out seems like apanacea ndash since one finally knows where everything isThis idea was captured in the conclusion of Perlin andFoxrsquos original paper lsquoAs compared to standardcurrent window models this system makes it easierfor the user to exploit visual memory of places toorganise informationally large workspacesrsquo (Perlin andFox 1993)

Yet even with these qualities the grand vision of azoomable desktop has never been broadly achievedFor example several commercial efforts have disap-peared (eg Cincro Zanvas GeoPhoenix Zoominatorand Innovative iBrowser) Variations such as TaskGallery (Robertson et al 2000) which used linearzooming in a 3D environment among others also havenot been broadly used There has however been aresurgence of commercial interest in ZUIs recently butas we will see they have significantly scaled backexpectations as to where zooming can be useful

2 ZUIsrsquo premise and promise

ZUIs have interested researchers since they wereintroduced by Perlin and Fox (1993) There havebeen a number of widely cited papers focused in severalareas

Figure 2 Recent ZUI Zumobirsquos ZoomCanvas zooms in from the entire canvas on start-up Dragging leftright pans andtapping on a region zooms in for interaction with the detailed content (Zumobi 2010)

Figure 1 Early ZUI Pad shows content at different sizeswith portals that show a remote region of the data surface(Perlin and Fox 1993 Figure 1) 1993 Association forComputing Machinery Inc Reprinted by permission

854 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Systems [eg Pad (Perlin and Fox 1993) Padthornthorn(Bederson and Hollan 1994) Jazz(Bederson et al 2000) and Piccolo (Bedersonet al 2004b)]

Applications [eg PadPrints (Hightower et al1998) PhotoMesa (Bederson 2001) Counter-Point (Good and Bederson 2002) KidPad(Druin et al 1997) AppLensamp LaunchTile(Karlson et al 2005) Microsoft Seadragon(2010) Fly (Lichtschlag et al 2009) MicrosoftpptPlex (2010) Microsoft Canvas for OneNote(2010) Microsoft Pivot (2010) and Prezi (2010)]

Theoretical constructs [eg Space-Scale Diagrams(Furnas and Bederson 1995) and Desert Fog (Juland Furnas 1998)]

Studies (eg Combs and Bederson 1999 Boltmanet al 2002 Hornbaeligk et al 2002 Klein andBederson 2005 and Plumlee and Ware 2006) tosupport and understand them

Many of these have had a fair amount ofaccolades and positive user response yet it is fairto say that none of them have been great commercialsuccesses (defined either monetarily or by largenumbers of users)

In fact I would argue that ZUIs have neverreached the level of broad use envisioned by theiroriginal creators There has been a huge amount ofeffort relating to ZUIs with dozens of publishedpapers and a commensurate amount of technologydeveloped Zooming has been successful in that somekind of zooming is commonly used in a wide rangeof interfaces (eg Google Maps Microsoft WordAdobe Photoshop and Apple iPhone) But a richerkind of zooming that takes over the desktop andbecomes the primary interface to onersquos data nevermaterialised

As one of those early creators and as someone whohad worked on many aspects of ZUIs since nearly theirbeginning I think it is worth reflecting on what thatoriginal excitement was about where ZUIs have beensuccessful where they have not been and why Forexample a number of commercial efforts that pursueda deeper kind of zooming had only modest success (egan effort by Sony to include zooming in early VAIOcomputers (Kunkel 1999 pp 160ndash165) and HillcrestLabsrsquo HoME product that has not had significantdistribution (Hillcrest Labs 2010)

In reflecting on this body of work it may bepossible to increase our understanding of why it issometimes harder to bring innovations to broad usethan initially thought And as researchers why weshould perhaps be a bit more tempered in ourenthusiasm ndash while of course not stifling innovationbefore it has the opportunity to flower

21 The early history

All ZUIs were built with the goal of addressing thefundamental issue that this paper started with ndash thatpeople need to interact with more information than fitsthe screen This is such a fundamental issue incomputing and interface design that it seemed excitingand appropriate to explore a new approach to addressthe problem The challenge was that the problem wasvaguely defined ndash without users or tasks ndash and thesolution was technically very difficult This led to severalyears of building general solutions rather than specificones Let us start by looking at the early history

lsquoPadrsquo was the first system that explored this space(Perlin and Fox 1993) (Figure 1) Pad ran on a SunSPARCstation with black and white graphics dis-played one bit per pixel bitmaps and used non-animated lsquojumprsquo zooming (each mouse click wouldredisplay the view magnified or reduced by a factor oftwo) Pad offered navigation authoring semanticzooming and portals The term lsquosemantic zoomingrsquo(coined by David Fox) refers to how objects can havedifferent visual representations at different sizesPortals were rectangular objects on the surface thatacted like cameras that showed other parts of thesurface Portals were designed to solve the limitationthat spatial layout implied It enabled objects that werephysically far apart to be used near each other

Back in 1992 I was finishing my PhD at New YorkUniversity (NYU) watching Perlinrsquos work closely andwhen I joined Bellcore after my graduation I beganbuilding the first of what turned into a series ofsuccessors to Pad While working with Jim Hollan Istarted with Pa3D a ZUI with richer vector andbitmap colour graphics and smoothly animatedzooming ndash running on much more expensive SGIcomputers Notably Pa3D was in a 3D environmentwhere every polygon could be a zoomable surfaceThere are no papers about Pa3D but a video showingit is available online (Bederson 1993)

Pad and Pa3D assumed that information would beplaced at different scales in this gigantic informationspace and users would access the information bypanning and zooming through the space using portalsto connect otherwise distance objects In other wordswe aimed to build a zoomable desktop We made manydemonstrations of how people might organise theirinformation in this space but notably despite ourconviction that zooming was an incredible solution wehad only limited examples of how people wouldperform day to day tasks

Being confident that zooming was a good idea butjust not sure what for I started building Padthornthorn alsowith Jim Hollan (Bederson and Hollan 1994aBederson et al 1996) Padthornthorn was a successor to

Behaviour amp Information Technology 855

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Pa3D which incorporated a well-defined applicationprogrammerrsquos interface (API) so others could buildzoomable applications and hopefully reveal the powerthat zooming offered

Over the years teams I led (with Jon Meyer JesseGrosjean and Aaron Clamage) extended Padthornthorn andthen built ZUI platforms Jazz (Bederson et al 2000)and Piccolo (Bederson et al 2004b Piccolo2D 2010) Wesupported other platforms and languages and evensupported mobile devices with PocketPiccoloNETThere was also a significant effort to supportsmooth zooming high-quality animated graphicsand overall high performance on what at the timewere fairly meagre computing systems (Bederson andMeyer 1998) Many other researchers and a fewcompanies built extensions and applications on theseplatforms (especially Padthornthorn) as did we In more recentyears commercially available platforms such as JavaNET and Flash made it easier to build zoomableapplications and clientserver solutions have also beenbuilt (Microsoft Seadragon Zoomorama GoogleMaps)

One significant effort was by Sony starting in 1998They started an internal effort to produce ZUIs fortheir VAIO PC computers that were new at the time(Kunkel 1999 pp 160ndash165) Franklin Servan-Schrei-ber who led that effort coined the term lsquoZUIrsquo1 and isthe person who led the recently defunct effort atZoomorama (2010) However Sony never shipped anyZUI products related to that effort

Looking at the core original ZUI ideas (zoomingsemantic zooming and portals) it is interesting to seewhich ones eventually had broad usage Basic zoominghas been used widely in everything from maps todocument viewers Semantic zooming has also hadsome broader use ndash even outside the realm of ZUIs Thelsquoribbonrsquo in some Microsoft Office products (such asWord) uses semantic zooming to display more informa-tion when more space is available It dynamicallychanges as you resize the ribbon to show more buttonsand to show more imagery as the panel gets largerSimilarly several on-screen music players show morecontrols and information when the player is enlarged Itis notable that not a single one of these applications usesportals Many applications (such as Microsoft WordGoogle Maps and Adobe Photoshop) offer multipleviews or a small fixed navigation overview Howevernone of these applications that focus on zooming as acore organisational and navigation technique useportals in a way similar to how they were envisioned

22 ZUI characteristics

There are a few essential characteristics of ZUIs thataffect their usefulness and their usability

Different ZUIs have had various levels of layoutflexibility By this I refer to how much creative controlthe end user has over the layout of information in thespace On one extreme users can put anything at anysize and place This yields complete artistic freedombut can be difficult to author ndash and has the potentialfor creating a visual mess that users can get lost in Onthe other extreme the environment might be con-strained allowing information only in specific placesaccording to a grid or layout algorithm A relatedcharacteristic is whether specific layouts are multi-levelby which I mean that objects in the space appear atsignificantly different sizes requiring a user to navigateto different zoom levels in order to see all the objects

Different ZUIs have also had various navigationmechanisms ndash which are ways for users to movethrough the space Again there is a trade-off betweenflexibility and usability Some interfaces allow users tofly through the space going absolutely anywhere ndashincluding deep into the spaces between objects [result-ing in some researchers labelling this phenomenaDesert Fog (Jul and Furnas 1998)] Very few otherapplications let a user navigate beyond the actualcontent Almost every document browser and editorlimits navigation to the available content (with thenotable exception of Microsoft Excelrsquos scroll bararrows ndash Apple Numbers and Google Spreadsheeton the other hand do limit navigation) On the otherhand some interfaces allow you only to click onobjects to zoom into them and click on a zoom outbutton to zoom out ndash making it impossible to get lostbut also giving less control over exactly where youlook

One approach to managing the complexity ofnavigation is to zoom automatically when neededusing a technique called speed-dependent automaticzooming (SDAZ)

221 Speed-dependent automatic zooming

One of the reasons that navigation is such a big issue isthat there are too many degrees of freedom to easilycontrol In a traditional computer interface onebutton is typically used for selection or action and ascroll bar or specialised key is used for navigation ZUIusers however must pan along two axes instead of justone and they must be able to zoom in and out Formost applications this must be done with the samehardware used by non-zoomable interfaces

Some researchers addressed this by reducing thenumber of navigational degrees of freedom that usersneeded to control in the first place (Igarashi andHinckley 2000 Cockburn et al 2005 Sun andGuimbretiere 2005) One of the reasons for zoomingout is simply to make it faster to pan a large distance

856 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

[as explained with space-scale diagrams (Furnas andBederson 1995) and elaborated on by Van Wijk andNuij (2004)] This observation led to SDAZ where theapplication would automatically zoom out when theuser panned quickly and then zoom back in whenthe user slowed down

Several studies showed the effectiveness of thisapproach (Cockburn et al 2005 Sun and Guimbretiere2005) and yet not a single research or commercial ZUIapplication (outside those that are explicitly exploringSDAZ) uses SDAZ as even an optional navigationmechanism Why is this There is not a clear answerbut it is important to observe that while SDAZsimplifies one thing (degrees of freedom of control)it complicates another (device control and perception)

SDAZ requires the user to engage in a real-timeperceive-think-act loop (Card et al 1986) This real-time interactive manipulation was typical of the lsquoflyingrsquoin many of the earliest ZUI systems ndash and that causedsignificant usability problems People who play videogames like the interaction challenge of real-timeinteraction People doing information processing taskstypically do not They are more likely to want to spendtheir cognitive resources on the underlying task not onanalysing the interface ndash even if it results in them beingslightly slower

222 Mouse wheels

While SDAZ is a somewhat complicated way to controlzooming some clear winners for controlling zoominghave emerged in the marketplace and they are themouse wheel and multi-touch lsquopinchingrsquo The mousewheel has become fairly widespread and is used fornumerous linear navigation actions ndash from scrolling toselection to zooming It offers simple control over theextra degree of freedom and while it is somewhat slow(requiring movement for each further bit of zooming)its simplicity makes it trivial to learn The fact thatzooming navigation often requires only a small scalechange makes for a perfect mapping to themouse wheel

223 Touch screens and multi-touch interaction

The use of multi-touch lsquopinchrsquo gestures on touchscreens is another control mechanism with growingpopularity Ishii and Ullmer originally envisioned thisapproach by indirectly using physical objects to scaleand rotate images on a screen in 1997 (Ishii and Ullmer1997) Then in 2002 Rekimoto developed a directsolution by using multiple fingers on a capacitive touchscreen to scale and rotate images (Rekimoto 2002)Finally this approach was commercialised and is nowbroadly used in Applersquos iPhone and other mobilephones tables as well as Microsoft Surface This kind

of direct multi-touch interaction is easy to learn andoperate More generally the rapidly growing use ofhigh-end touch screen-based smartphones is increasingthe need for zooming (with people frequently readingdocuments designed for larger screens on smallscreens) and pinching seems to be the de factostandard for controlling this

23 Applications

To see something of the range of uses of ZUIs and howtheir characteristics have changed over time Table 1shows a selection of zoomable applications Only lsquotruersquoZUIs are shown (those supporting more than onedocument or object) It captures a range of what peoplehave been using zooming for and makes apparent therange of approaches that people have taken with regardto layout flexibility and navigation It is also clear thatthe essential problem of getting lost in Desert Fog hasnot been consistently avoided Furthermore it is clearthat there is no consistency in the mechanisms that areused to navigate through space

Of broader import there is no obvious focus onusers tasks or problem domains This is characteristicof both the potential and the challenge of ZUIsZooming is a completely general strategy and so oneshould not be surprised to see such a lack of focus anymore than one might see scroll bars and search used inan incredibly diverse set of scenarios That said theredoes seem to be a concentration of ZUIs applied todomains with highly visual documents ndash which ofcourse makes sense given the visual nature of ZUIs

24 Studies

Some of these and other ZUI applications have beenstudied for usability or effectiveness However it is thenature of those application-centric studies that it isdifficult to tease out the relative benefit or cost of thezooming characteristics of those applications

In fact there are relatively few studies that look atthe value of animated zooming more generally But afew are particularly useful so let us look at them

241 Overviews

The first study by Hornbaeligk et al (2002) looked atmulti-scale information structure as well as the use ofsmall overview windows that show a zoomed out viewof the information Surprisingly it found that whileparticipants like the overview window its use resultedin lower performance The study used geographicalmaps as the data set Navigation worked by draggingthe mouse to pan holding down the left button tozoom in and holding down the right button to zoom

Behaviour amp Information Technology 857

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

SomeZUIsystem

sandapplications

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

1993

Pad(PerlinandFox

1993)

First

ZUIplatform

included

several

conceptualisationsof

usagescenarios

Unconstrained

Yes

Leftclickto

zoom

in

Rightclickto

zoom

out

1994

Padthornthorn

(Bederson

andHollan1994)

SecondZUIplatform

included

APIfor

othersto

write

ZUI

apps

Unconstrained

Yes

Leftclickandhold

tozoom

inRightclick

andhold

tozoom

out

1997

KidPad(D

ruin

etal

1997)

Useslsquolocaltoolsrsquo

(Bedersonet

al1996)

toenable

childrento

create

stories

Unconstrained

Drawingsim

ages

textpositioned

manually

Yes

Hyperlinksforpath

Zoom

inoutmodes

Click

toflyinout

1998

PadPrints

(Hightower

etal1998)

Createsavisual

hierarchicalmapof

web

pages

visited

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2001

CounterPoint(G

ood

andBederson2002)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Yes

Click

onslidefornext

Rightclick-and-hold

fliesinout

2001

PhotoMesa(Bederson

2001)

Apersonalphoto

browserdesktopand

mobile

Containsphotosin

adynamicallygenerated

groupofgridsusinga

treemapalgorithm

No

Leftclickzoomsin

Rightclickzoomsout

2002

Seadragon(2010)

Clientserver

high

perfZUIforim

ages

over

anetwork

(web

andmobile)

Containsphotoslaid

out

byasm

allnumber

of

fixed

layout

algorithms

No

Click

tozoom

in

On-screenbutton

zoomsout

2002

Spacetree

(Grosjean

etal2002)

Hierarchyexploration

tool

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2006

Dynapad(Bauer

2006)

Supportsorganisationof

personalcollections

Dynamicveryflexible

No

Multi-touch

2007

Apple

iPhone(2010)

Applicationlauncher

for

Applersquosmobilephone

Containsiconslaid

out

inafixed

grid

No

Tapto

zoom

inPhysical

buttonto

zoom

out

2008

MicrosoftpptPlex(2010)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Nobutwithmanual

override

Click

onslidefornext

Click

tozoom

inEsc

tozoom

outScroll

wheelfliesinout

2008

Zoomorama(2010)

Clientserver

toenable

high-perform

ance

zoomingofim

ages

over

anetwork

Dynamicallygenerated

gridcontaining

images

No

Pop-upon-screenbut-

tonsforzoom

inout

(continued)

858 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

(continued)

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

2008

Prezi

(2010)

Website

forcreatingand

makingfree-form

presentations

Unconstrained

Drawingsim

ages

moviestext

positioned

manually

Yes

TabfornextScroll

wheelon-screen

buttonsandmouse

keyboard

comboto

flyinout

2009

Zumobi(2010)

ZoomCanvas

Launcher

forasm

all

number

ofrelated

contentareason

high-endmobile

devices

Fixed

hand-tuned

layout

turned

tocontent

No

Tapto

zoom

in

on-screenbuttonto

zoom

out

2009

Microsoft

Canvasfor

OneN

ote

(2010)

Plug-inforMicrosoft

OneN

ote

showsvisual

overview

ofalla

userrsquosnotes

Dynamicallygenerated

gridofgrids

containingim

ages

of

notesfrom

Microsoft

OneN

ote

Nobutwithmanual

override

Click

onnote

tozoom

inEscrightclickor

bgndclickto

zoom

out

2009

Schem

atic(2010)

Example

ofcustom

website

withfixed

layoutoffixed

content

Fixed

hand-tuned

layout

tuned

tocontent

No

Click

onslideto

zoom

inClick

on

backgroundto

zoom

out

2009

Microsoft

Pivot(2010)

Facetedsearchtoolthat

useszoomingto

navi-

gate

throughsearch

results

Fixed

gridorgridwithin

barchart

layouts

No

Multiple

mechanisms

includingclickingand

on-screenbuttons

NoteNote

thatItheauthoram

acofounder

ofZumobiInc

Behaviour amp Information Technology 859

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

out The study varied whether or not users were shownoverviews and whether the map information wassingle or multi-level That is single level mapsdisplayed all textual information in a single font sizeso when the user zoomed out the text could not beread Multi-level maps displayed textual informationat different sizes Text referring to larger areas on themap was displayed at much larger fonts so that whenusers zoomed out they could read the text coveringlarger geographical areas

Before this study it was believed that overviewswere a good idea and that they improved subjectivesatisfaction (North and Shneiderman 2000) andefficiency (Beard and Walker 1990 Bauer 2006)However this study found a discrepancy betweenpreference and performance Study participants pre-ferred the overview condition but actually performedbetter in the no-overview condition especially whenusing the multi-level map

It turns out that switching between overview anddetail windows was correlated with higher taskcompletion time suggesting that integration of over-view and detail windows requires mental and motoreffort

This result is notable because a fundamentalconcern about ZUIs is that they require time to use(since the animations take time to occur) Worse thereis the potential that ZUIs tax human short-termmemory (STM) because users must integrate in theirheads the spatial layout of the information On theother hand there is the hope that the animatedtransitions are understood preconsciously by the hu-man visual system and thus may not in fact place asignificant load on STM

This study gives some evidence that the cost ofunderstanding an animated zoomable map of informa-tion is in fact less than the use of one with an overviewthat is one of the primary alternative interface designsOf course it does not say anything about the benefit orcost of a spatially organised information spacecompared to a non-spatially organised space It alsodoes not say anything about the benefit or cost ofanimation

242 Animation

A second study looked at simple animated navigationtransitions and found that these animations signifi-cantly increased performance ndash even including the timethe animations took These kinds of animations havelong been expected to provide benefit (Card et al1991) but evidence backing up this understanding hadbeen lacking Klein and Bederson (2005) looked at avery simple but highly controlled use of animationwhen scrolling while reading documents

Understanding animation precisely is important be-cause the potential benefit of animation is underminedby the fact that animation by its nature takes time Soeven if the animation provides some benefit there is acost as well and it is not obvious whether the benefitoutweighs the cost

Participants in this study read documents out loudand when they got to the end of a page they scrolled bypressing the down arrow key The animation speedvaried between 0 (ie unanimated) 100 300 and 500 msThe document type also varied between unformattedtext formatted text and abstract symbols (whereparticipants counted symbols instead of reading) Thereason for varying the document type was to understandhow visual landmarks interact with animation

The essential result was that animation did in factprovide a significant benefit even considering the timespent on the animation Reading errors were reducedby 54 with 500 ms animations Reading task timewas reduced by 3 and counting task time wasreduced by 24 for 300 ms animations The formatteddocuments had a higher overall performance (andlower improvement) implying that visual landmarksappear to be a good idea which is consistent with theHornbaeligk studyrsquos observation about multi-level maps

The lesson from this study is that when designedcorrectly the benefits of simple animated navigationaltransitions can outweigh their costs This supports thedesign decision to make zooming transitions use shortanimations This study is not definitive however since itis possible that the benefit that was seen while scrollingwould not occur while zooming However there is noobvious reason that the same benefits would not occurso this kind of animated transition continues to seemvaluable

243 Presentations

Two other studies looked at how people learned fromzoomable spaces (Boltman et al 2002 Good andBederson 2002) The outcomes here were more mixedwhere the zoomable presentation did not producebetter learning or memory of the presentation ndashalthough participants did remember the structure ofthe presentation better The first study looked at howchildren responded to a story that was presented withor without animated zooming The children with theanimated condition elaborated more during a discus-sion but did not otherwise recall the story better Thesecond looked at how college students responded to aslide presentation ndash again with or without animatedzooming The presentations were structured visually ina 2D hierarchical format with six sections arrangedin a circle where each section contained slides arrangedin a circle These students also did not recall the

860 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

presentation better but the ones who saw the animatedzooming condition did recall the structure of thepresentation better

Together these studies imply that zooming alone isprobably not enough to help people remember contentbetter However users of ZUIs may be more engagedand they may remember the spatial structure of thecontent better ndash even if they do not remember theactual content itself any better However the value ofremembering structure depends on the task If thestructure has value in and of itself then this is apositive value If on the other hand the visualstructure was added as a means to make the presenta-tion more visually interesting and had no intrinsicvalue then such improved retention would provide nobenefit

244 Zooming vs multiple windows

Two related papers (Plumlee and Ware 2002 Plumleeand Ware 2006) look more generally at modelling andthen studying how people perform visual comparisonsof multi-scale data using zoomable and multi-windowinterfaces Informed by an understanding of theextreme limitations of human visual working memorythey modelled how people can search using the twointerfaces The model showed that comparison tasksthat exceeded the capacity of visual working memory(which can hold roughly three visual objects) would bebetter served by adding windows Multiple windowsdecrease visual working memory load because differentviews can be directly compared In contrast ZUIsrequire the user to hold everything outside of thecurrent view in their memory ndash thus making theminappropriate for complex comparison tasks Thestudies that Plumlee and Ware performed demon-strated this essential characteristic and the trade-offbetween ZUIs and multiple windows closely matchedtheir models They also showed that users of ZUIs hadsignificantly higher error rates than users of multiple

windows The practical impact of these studies dependson the frequency and difficulty of performing zoomingcompared to creating and managing windows And italso applies only to visual comparison tasks

So again the lessons are interesting and importantto specific usage cases but there remains no studytelling us lsquowhether zooming is goodrsquo The real answer isthat it depends

3 Discussion

A trend in ZUI applications that appears to be fairlyclear over the years is that over time these applicationshave tended to get simpler and easier to use while atthe same time offering less creative control to endusers This is an important trend because researchersoften have a tendency to create tools that are rich andpowerful After all it is easier to conceptualise aninnovation as offering lsquomorersquo rather than offering lsquolessrsquo

But the marketplace is clear ndash success often requiressimplicity first and power second Of course marryingthe two where the power is under the hood is often thebest solution ndash but also very difficult to design Thesuccess of so many modern applications (eg searchtwitter photo sharing iPhone App Store) are examplesof this in that they can be used with almost no trainingin tiny little bursts

I used to think that only public kiosks had to havethese simplistic characteristics ndash but now it seems thateven applications that people use for hours a week andbecome true experts in follow the same rules If truethis implies that we as researchers have to stronglyconsider not only what new things technologies canhelp people do but also consider simplicity and speedas fundamental goals

31 Why ZUIs are challenging

As summarised in Table 2 the potential benefits ofZUIs are sometimes mirages ZUIs are generally

Table 2 Promise and reality of ZUIs

Promise Reality

Engaging If designed well The animated nature of ZUIs is sometimes referred to as lsquoeye candyrsquo implying that peoplecannot help but pay attention to the movement on the screen However this very movement is a double-edgedsword because if the user does not pay attention and misses the animation then the connection between thedifferent views is lost and the user is likely worse off than without a ZUI

Visually rich Can also make space more complex to navigate and understand ZUIs can have visual objects at any size and atany place ndash thus offering the possibility of placing significant content (scaled down) even in a tiny crackbetween two other objects This kind of structure is very difficult to perceive (since users cannot see the tinycontent in the first place) and conceptualise (since at any given view the vastly different sized objects can notbe seen simultaneously)

Simplicity Does not scale to very large data sets One of the attractions of ZUIs is that they appear conceptually simple(lsquojust zoom in for a closer lookrsquo) But in practice navigating large information spaces in ZUIs is often difficultto execute and requires significant short-term memory to navigate

Behaviour amp Information Technology 861

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

engaging (although they make some people feelphysically sick) and visually rich But the promise ofsimplicity falls short

While human visual perception does make it easy tosee where one is navigating the reality is that it places aheavy load on short-termmemory to remember where inspace you just were and where things are And therequirement of human memory to know how space isorganised means that ZUIs do not scale up very wellZUIs are oftenmotivated by the physical world and howpeople like laying papers out on their desk But no onewants all of their papers on their desk It is much morecommon to have only a relatively small number ofpapers that are actually being worked with

The visual overviews that ZUIs offer for free byzooming out may seem like a solution to the load onhuman memory but in practice it does not becausevisual overviews of any complexity require significantvisual search in order to find anything If there are just asmall number of objects then the visual search task is nothard ndash but of course for a small number of objects youdo not need a ZUI to solve your organisationalproblems

In addition the visual richness of ZUIs is a double-edged sword It requires skill to design a complex spacewith documents of arbitrary size aspect ratio andcolour that people can comprehend and scan Alsopeople are not as good at scanning 2D designs as 1Dlayouts ndash unless the layout is highly structured Buthighly structured 2D layouts do not work well forvisual objects of arbitrary aspect ratios Designers areobligated to leave a lot of unused space scale down thelarge objects so they are unreadable or crop the largeobjects ndash thus losing much of their distinguishability

Finally it is crucial to consider the efficiency ofZUIs and their alternatives It is the nature ofzoomable structures to organise information hierarchi-cally Zooming in to an object shows you moreinformation about that object Of course hierarchicalstructures are also commonly exposed with a simpletree widget as is commonly found in most file systembrowsers The file system browsers exemplified byWindows Explorer are not only very efficient but alsomore easily let the user compare different areas in thehierarchy by simply opening and collapsing the desirednodes So in many cases the alternative to a ZUI is asimple tree widget that is simple compact easy tolearn and fast to use Thus in addition to the broaderchallenges of ZUIs we should always consider thealternatives that are also powerful

32 Design guidelines

So how should ZUIs be designed to best overcomethese obstacles Let us start with the big issues

321 Support the right tasks

ZUIs like most visualisations have strong potentialfor supporting users in understanding the big pictureidentifying trends patterns and outliers But theytypically are not good at helping users simply get thebest answer So be sure that some sort of overview-based design is called for in the first place

322 Be cautious about using zooming as the primaryinterface

Zooming navigation can be difficult and is uncom-mon Unless the data set is fundamentally spatial (egmaps) users probably do not want zooming as theirprimary interaction with the data Instead think abouthow people search and create data sets ndash which maythen be accessed partially by zooming ie createhybrid systems that combine mechanisms such asfaceted search and zooming [eg PhotoMesa (Beder-son 2001) the Hard Rock Memorabilia (2010) orMicrosoft Pivot (2010)]

323 Only use ZUIs when a small visual representa-tion of the data is available

Zooming works best when the objects can be recognisedwhen they are zoomed out So certain domains arebetter suited to ZUIs Photos are good purely textualdocuments are bad and audio recordings are terrible(unless there are associated images) Recent researchshows how a small visual snippet may be automaticallycreated from web pages (Teevan et al 2009)

324 Donrsquot limit yourself to fixed data sets

People rarely want to interact long with a fixed set ofdata But it is often technically difficult to dynamicallycreate and modify large zoomable spaces Be sure thatthe user interface does not get blocked or slowed downwhen new data is added to a zoomable space

And then there are several more specificsuggestions

325 Maintain aspect ratio

It is tempting to use semantic zooming to signifi-cantly change the visual representation of objects atdifferent sizes For example a visual chart may becondensed to a single value or a text document maybe replaced by its title when small However thesmall object must have the same aspect ratio ndash or atleast fit in the same box as the full object Otherwisenearby objects can end up overlapping when the userzooms out

862 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

326 Use meaningful layout

The problem of finding a specific object among a largenumberof themcanbe alleviated if theobjects are laidoutin a semantically meaningful way If the user can predictwhere an object will be based on attributes that they arelikely to be familiar with then they can trim the searchspace Traditional paper phones books with theiralphabetic ordering are a good example of this approach

327 Use consistent layout

Ideally once an object is placed in space it should notmove That way the user can build up an internal mapover time of where objects are Unfortunately thisideal is often in direct conflict with the previous one ofmeaningful layout If objects are to be laid outmeaningfully then the layout must change over timeas the data changes and as the organisationalmechanism changes PhotoMesa (Bederson 2001) isan example of a ZUI that went for a meaningful layoutbut had no consistency at all

328 Use scannable layout

Objects must be positioned in such a manner as to enablethe user to visually scan the space to look for an objectwithout missing or repeating objects Simple 1D lists ofobjects are easy to scanbecause the person just runs downthe list and the only state that must be kept is the currentitem and the direction of scanning In 2D however forany layout except for a grid there is the potential for amuch heavier load on short-term memory

329 Use breadth over depth

It is tempting to place objects in space at manydifferent sizes so more information can be discoveredby zooming in further ndash and in fact many of the earlierZUI applications did this However when objects areplaced at many different levels the user would notknow that the small objects exist (unless the space iscarefully designed to indicate them) and they can getlost Further even if the user knows they are there it istypically burdensome to zoom in and out repeatedlySpaceTree (Grosjean et al 2002) is an example of anapplication that was designed explicitly with breadth inmind Even with hundreds of thousands of objects allof them are at the same size

3210 Use small data sets

The problems that come with a large number of objectsin a zoomable space can be mitigated by not buildingapplications around large data sets ZUIs appear to

work better for users when they have no more thanabout 100 screens worth of information HoweverZUIs with 1000 or more screens worth of informationbecome problematic because of the issues raised in theprevious section (why ZUIs are challenging) This is afundamental problem since ZUIs are often designed towork for user generated data ndash which may not easily belimited So application designers may choose topartition the space into manageable subsets

3211 Use simple and consistent navigationinteraction

People must be able to discover how to navigate thespace with no training and it must be difficult orimpossible to get lost in Desert Fog Unfortunatelythere is significant inconsistency among applicationsThey vary between single and double clicks to zoom inhow much the screen is zoomed per action whetherzooms scale by a fixed percentage or whether the targetobject fills the screen And mechanisms are all over themap to zoom out ndash from right click left click on thebackground fixed on-screen buttons and pop-upbuttons And as with zoom in the actual amountthat is zoomed out is inconsistent across applicationsas well

4 Conclusion

Understanding ZUIs is difficult They have clearlybeen a success by traditional research measures(significant visibility publications and citations) Andelements of ZUIs have been broadly deployed incommonly used applications But the original broadvision of zooming as a basic information organisa-tional principle [such as Raskinrsquos vision of hospitalmanagement (Raskin 2000)] has not happened Ofcourse lack of commercial success does not mean anidea is not a good one (maybe the business reasons didnot line up or maybe the right tools got built but withthe wrong marketing strategy the timing may havebeen wrong or any number of other reasons mighthave stalled a successful business) But still the lack ofbroad success coupled with the fundamental issuesraised in this paper appears significant

I definitely would not rule out further success ofZUIs and I think that the successes that haveoccurred such as Google Maps Microsoft Word andApple iPhone are substantial However their applic-ability is much narrower than original envisioned Andthe design and technical challenges of building effectiveZUIs make it that much less likely that they will bebroadly deployed At the same time there are clearcases where zooming is valuable ndash such as withdocuments that are larger than screens (ie photos

Behaviour amp Information Technology 863

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

The promise of ZUIs comes largely from the followinggeneral expectations

121 ZUIs are engaging

The animation is visually attention grabbing It takesadvantage of human visual perception abilities Peoplecan process lsquovisual flowrsquo (Ware 2004) preconsciouslyso it feels easy to build a mental map of theinformation

122 ZUIs are visually rich

There are more degrees of freedom to visually structureobjects and thus they offer the potential for great

creative expression This was identified by Perlin andFox in their original paper on Pad with their exampleof a branching tree story (Perlin and Fox 1993)

123 ZUIs offer the lure of simplicity

The fact that you find information by looking for it ina place implies a promise of simplicity that will solveour organisational and information retrieval problemsThe overview one sees when zoomed out seems like apanacea ndash since one finally knows where everything isThis idea was captured in the conclusion of Perlin andFoxrsquos original paper lsquoAs compared to standardcurrent window models this system makes it easierfor the user to exploit visual memory of places toorganise informationally large workspacesrsquo (Perlin andFox 1993)

Yet even with these qualities the grand vision of azoomable desktop has never been broadly achievedFor example several commercial efforts have disap-peared (eg Cincro Zanvas GeoPhoenix Zoominatorand Innovative iBrowser) Variations such as TaskGallery (Robertson et al 2000) which used linearzooming in a 3D environment among others also havenot been broadly used There has however been aresurgence of commercial interest in ZUIs recently butas we will see they have significantly scaled backexpectations as to where zooming can be useful

2 ZUIsrsquo premise and promise

ZUIs have interested researchers since they wereintroduced by Perlin and Fox (1993) There havebeen a number of widely cited papers focused in severalareas

Figure 2 Recent ZUI Zumobirsquos ZoomCanvas zooms in from the entire canvas on start-up Dragging leftright pans andtapping on a region zooms in for interaction with the detailed content (Zumobi 2010)

Figure 1 Early ZUI Pad shows content at different sizeswith portals that show a remote region of the data surface(Perlin and Fox 1993 Figure 1) 1993 Association forComputing Machinery Inc Reprinted by permission

854 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Systems [eg Pad (Perlin and Fox 1993) Padthornthorn(Bederson and Hollan 1994) Jazz(Bederson et al 2000) and Piccolo (Bedersonet al 2004b)]

Applications [eg PadPrints (Hightower et al1998) PhotoMesa (Bederson 2001) Counter-Point (Good and Bederson 2002) KidPad(Druin et al 1997) AppLensamp LaunchTile(Karlson et al 2005) Microsoft Seadragon(2010) Fly (Lichtschlag et al 2009) MicrosoftpptPlex (2010) Microsoft Canvas for OneNote(2010) Microsoft Pivot (2010) and Prezi (2010)]

Theoretical constructs [eg Space-Scale Diagrams(Furnas and Bederson 1995) and Desert Fog (Juland Furnas 1998)]

Studies (eg Combs and Bederson 1999 Boltmanet al 2002 Hornbaeligk et al 2002 Klein andBederson 2005 and Plumlee and Ware 2006) tosupport and understand them

Many of these have had a fair amount ofaccolades and positive user response yet it is fairto say that none of them have been great commercialsuccesses (defined either monetarily or by largenumbers of users)

In fact I would argue that ZUIs have neverreached the level of broad use envisioned by theiroriginal creators There has been a huge amount ofeffort relating to ZUIs with dozens of publishedpapers and a commensurate amount of technologydeveloped Zooming has been successful in that somekind of zooming is commonly used in a wide rangeof interfaces (eg Google Maps Microsoft WordAdobe Photoshop and Apple iPhone) But a richerkind of zooming that takes over the desktop andbecomes the primary interface to onersquos data nevermaterialised

As one of those early creators and as someone whohad worked on many aspects of ZUIs since nearly theirbeginning I think it is worth reflecting on what thatoriginal excitement was about where ZUIs have beensuccessful where they have not been and why Forexample a number of commercial efforts that pursueda deeper kind of zooming had only modest success (egan effort by Sony to include zooming in early VAIOcomputers (Kunkel 1999 pp 160ndash165) and HillcrestLabsrsquo HoME product that has not had significantdistribution (Hillcrest Labs 2010)

In reflecting on this body of work it may bepossible to increase our understanding of why it issometimes harder to bring innovations to broad usethan initially thought And as researchers why weshould perhaps be a bit more tempered in ourenthusiasm ndash while of course not stifling innovationbefore it has the opportunity to flower

21 The early history

All ZUIs were built with the goal of addressing thefundamental issue that this paper started with ndash thatpeople need to interact with more information than fitsthe screen This is such a fundamental issue incomputing and interface design that it seemed excitingand appropriate to explore a new approach to addressthe problem The challenge was that the problem wasvaguely defined ndash without users or tasks ndash and thesolution was technically very difficult This led to severalyears of building general solutions rather than specificones Let us start by looking at the early history

lsquoPadrsquo was the first system that explored this space(Perlin and Fox 1993) (Figure 1) Pad ran on a SunSPARCstation with black and white graphics dis-played one bit per pixel bitmaps and used non-animated lsquojumprsquo zooming (each mouse click wouldredisplay the view magnified or reduced by a factor oftwo) Pad offered navigation authoring semanticzooming and portals The term lsquosemantic zoomingrsquo(coined by David Fox) refers to how objects can havedifferent visual representations at different sizesPortals were rectangular objects on the surface thatacted like cameras that showed other parts of thesurface Portals were designed to solve the limitationthat spatial layout implied It enabled objects that werephysically far apart to be used near each other

Back in 1992 I was finishing my PhD at New YorkUniversity (NYU) watching Perlinrsquos work closely andwhen I joined Bellcore after my graduation I beganbuilding the first of what turned into a series ofsuccessors to Pad While working with Jim Hollan Istarted with Pa3D a ZUI with richer vector andbitmap colour graphics and smoothly animatedzooming ndash running on much more expensive SGIcomputers Notably Pa3D was in a 3D environmentwhere every polygon could be a zoomable surfaceThere are no papers about Pa3D but a video showingit is available online (Bederson 1993)

Pad and Pa3D assumed that information would beplaced at different scales in this gigantic informationspace and users would access the information bypanning and zooming through the space using portalsto connect otherwise distance objects In other wordswe aimed to build a zoomable desktop We made manydemonstrations of how people might organise theirinformation in this space but notably despite ourconviction that zooming was an incredible solution wehad only limited examples of how people wouldperform day to day tasks

Being confident that zooming was a good idea butjust not sure what for I started building Padthornthorn alsowith Jim Hollan (Bederson and Hollan 1994aBederson et al 1996) Padthornthorn was a successor to

Behaviour amp Information Technology 855

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Pa3D which incorporated a well-defined applicationprogrammerrsquos interface (API) so others could buildzoomable applications and hopefully reveal the powerthat zooming offered

Over the years teams I led (with Jon Meyer JesseGrosjean and Aaron Clamage) extended Padthornthorn andthen built ZUI platforms Jazz (Bederson et al 2000)and Piccolo (Bederson et al 2004b Piccolo2D 2010) Wesupported other platforms and languages and evensupported mobile devices with PocketPiccoloNETThere was also a significant effort to supportsmooth zooming high-quality animated graphicsand overall high performance on what at the timewere fairly meagre computing systems (Bederson andMeyer 1998) Many other researchers and a fewcompanies built extensions and applications on theseplatforms (especially Padthornthorn) as did we In more recentyears commercially available platforms such as JavaNET and Flash made it easier to build zoomableapplications and clientserver solutions have also beenbuilt (Microsoft Seadragon Zoomorama GoogleMaps)

One significant effort was by Sony starting in 1998They started an internal effort to produce ZUIs fortheir VAIO PC computers that were new at the time(Kunkel 1999 pp 160ndash165) Franklin Servan-Schrei-ber who led that effort coined the term lsquoZUIrsquo1 and isthe person who led the recently defunct effort atZoomorama (2010) However Sony never shipped anyZUI products related to that effort

Looking at the core original ZUI ideas (zoomingsemantic zooming and portals) it is interesting to seewhich ones eventually had broad usage Basic zoominghas been used widely in everything from maps todocument viewers Semantic zooming has also hadsome broader use ndash even outside the realm of ZUIs Thelsquoribbonrsquo in some Microsoft Office products (such asWord) uses semantic zooming to display more informa-tion when more space is available It dynamicallychanges as you resize the ribbon to show more buttonsand to show more imagery as the panel gets largerSimilarly several on-screen music players show morecontrols and information when the player is enlarged Itis notable that not a single one of these applications usesportals Many applications (such as Microsoft WordGoogle Maps and Adobe Photoshop) offer multipleviews or a small fixed navigation overview Howevernone of these applications that focus on zooming as acore organisational and navigation technique useportals in a way similar to how they were envisioned

22 ZUI characteristics

There are a few essential characteristics of ZUIs thataffect their usefulness and their usability

Different ZUIs have had various levels of layoutflexibility By this I refer to how much creative controlthe end user has over the layout of information in thespace On one extreme users can put anything at anysize and place This yields complete artistic freedombut can be difficult to author ndash and has the potentialfor creating a visual mess that users can get lost in Onthe other extreme the environment might be con-strained allowing information only in specific placesaccording to a grid or layout algorithm A relatedcharacteristic is whether specific layouts are multi-levelby which I mean that objects in the space appear atsignificantly different sizes requiring a user to navigateto different zoom levels in order to see all the objects

Different ZUIs have also had various navigationmechanisms ndash which are ways for users to movethrough the space Again there is a trade-off betweenflexibility and usability Some interfaces allow users tofly through the space going absolutely anywhere ndashincluding deep into the spaces between objects [result-ing in some researchers labelling this phenomenaDesert Fog (Jul and Furnas 1998)] Very few otherapplications let a user navigate beyond the actualcontent Almost every document browser and editorlimits navigation to the available content (with thenotable exception of Microsoft Excelrsquos scroll bararrows ndash Apple Numbers and Google Spreadsheeton the other hand do limit navigation) On the otherhand some interfaces allow you only to click onobjects to zoom into them and click on a zoom outbutton to zoom out ndash making it impossible to get lostbut also giving less control over exactly where youlook

One approach to managing the complexity ofnavigation is to zoom automatically when neededusing a technique called speed-dependent automaticzooming (SDAZ)

221 Speed-dependent automatic zooming

One of the reasons that navigation is such a big issue isthat there are too many degrees of freedom to easilycontrol In a traditional computer interface onebutton is typically used for selection or action and ascroll bar or specialised key is used for navigation ZUIusers however must pan along two axes instead of justone and they must be able to zoom in and out Formost applications this must be done with the samehardware used by non-zoomable interfaces

Some researchers addressed this by reducing thenumber of navigational degrees of freedom that usersneeded to control in the first place (Igarashi andHinckley 2000 Cockburn et al 2005 Sun andGuimbretiere 2005) One of the reasons for zoomingout is simply to make it faster to pan a large distance

856 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

[as explained with space-scale diagrams (Furnas andBederson 1995) and elaborated on by Van Wijk andNuij (2004)] This observation led to SDAZ where theapplication would automatically zoom out when theuser panned quickly and then zoom back in whenthe user slowed down

Several studies showed the effectiveness of thisapproach (Cockburn et al 2005 Sun and Guimbretiere2005) and yet not a single research or commercial ZUIapplication (outside those that are explicitly exploringSDAZ) uses SDAZ as even an optional navigationmechanism Why is this There is not a clear answerbut it is important to observe that while SDAZsimplifies one thing (degrees of freedom of control)it complicates another (device control and perception)

SDAZ requires the user to engage in a real-timeperceive-think-act loop (Card et al 1986) This real-time interactive manipulation was typical of the lsquoflyingrsquoin many of the earliest ZUI systems ndash and that causedsignificant usability problems People who play videogames like the interaction challenge of real-timeinteraction People doing information processing taskstypically do not They are more likely to want to spendtheir cognitive resources on the underlying task not onanalysing the interface ndash even if it results in them beingslightly slower

222 Mouse wheels

While SDAZ is a somewhat complicated way to controlzooming some clear winners for controlling zoominghave emerged in the marketplace and they are themouse wheel and multi-touch lsquopinchingrsquo The mousewheel has become fairly widespread and is used fornumerous linear navigation actions ndash from scrolling toselection to zooming It offers simple control over theextra degree of freedom and while it is somewhat slow(requiring movement for each further bit of zooming)its simplicity makes it trivial to learn The fact thatzooming navigation often requires only a small scalechange makes for a perfect mapping to themouse wheel

223 Touch screens and multi-touch interaction

The use of multi-touch lsquopinchrsquo gestures on touchscreens is another control mechanism with growingpopularity Ishii and Ullmer originally envisioned thisapproach by indirectly using physical objects to scaleand rotate images on a screen in 1997 (Ishii and Ullmer1997) Then in 2002 Rekimoto developed a directsolution by using multiple fingers on a capacitive touchscreen to scale and rotate images (Rekimoto 2002)Finally this approach was commercialised and is nowbroadly used in Applersquos iPhone and other mobilephones tables as well as Microsoft Surface This kind

of direct multi-touch interaction is easy to learn andoperate More generally the rapidly growing use ofhigh-end touch screen-based smartphones is increasingthe need for zooming (with people frequently readingdocuments designed for larger screens on smallscreens) and pinching seems to be the de factostandard for controlling this

23 Applications

To see something of the range of uses of ZUIs and howtheir characteristics have changed over time Table 1shows a selection of zoomable applications Only lsquotruersquoZUIs are shown (those supporting more than onedocument or object) It captures a range of what peoplehave been using zooming for and makes apparent therange of approaches that people have taken with regardto layout flexibility and navigation It is also clear thatthe essential problem of getting lost in Desert Fog hasnot been consistently avoided Furthermore it is clearthat there is no consistency in the mechanisms that areused to navigate through space

Of broader import there is no obvious focus onusers tasks or problem domains This is characteristicof both the potential and the challenge of ZUIsZooming is a completely general strategy and so oneshould not be surprised to see such a lack of focus anymore than one might see scroll bars and search used inan incredibly diverse set of scenarios That said theredoes seem to be a concentration of ZUIs applied todomains with highly visual documents ndash which ofcourse makes sense given the visual nature of ZUIs

24 Studies

Some of these and other ZUI applications have beenstudied for usability or effectiveness However it is thenature of those application-centric studies that it isdifficult to tease out the relative benefit or cost of thezooming characteristics of those applications

In fact there are relatively few studies that look atthe value of animated zooming more generally But afew are particularly useful so let us look at them

241 Overviews

The first study by Hornbaeligk et al (2002) looked atmulti-scale information structure as well as the use ofsmall overview windows that show a zoomed out viewof the information Surprisingly it found that whileparticipants like the overview window its use resultedin lower performance The study used geographicalmaps as the data set Navigation worked by draggingthe mouse to pan holding down the left button tozoom in and holding down the right button to zoom

Behaviour amp Information Technology 857

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

SomeZUIsystem

sandapplications

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

1993

Pad(PerlinandFox

1993)

First

ZUIplatform

included

several

conceptualisationsof

usagescenarios

Unconstrained

Yes

Leftclickto

zoom

in

Rightclickto

zoom

out

1994

Padthornthorn

(Bederson

andHollan1994)

SecondZUIplatform

included

APIfor

othersto

write

ZUI

apps

Unconstrained

Yes

Leftclickandhold

tozoom

inRightclick

andhold

tozoom

out

1997

KidPad(D

ruin

etal

1997)

Useslsquolocaltoolsrsquo

(Bedersonet

al1996)

toenable

childrento

create

stories

Unconstrained

Drawingsim

ages

textpositioned

manually

Yes

Hyperlinksforpath

Zoom

inoutmodes

Click

toflyinout

1998

PadPrints

(Hightower

etal1998)

Createsavisual

hierarchicalmapof

web

pages

visited

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2001

CounterPoint(G

ood

andBederson2002)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Yes

Click

onslidefornext

Rightclick-and-hold

fliesinout

2001

PhotoMesa(Bederson

2001)

Apersonalphoto

browserdesktopand

mobile

Containsphotosin

adynamicallygenerated

groupofgridsusinga

treemapalgorithm

No

Leftclickzoomsin

Rightclickzoomsout

2002

Seadragon(2010)

Clientserver

high

perfZUIforim

ages

over

anetwork

(web

andmobile)

Containsphotoslaid

out

byasm

allnumber

of

fixed

layout

algorithms

No

Click

tozoom

in

On-screenbutton

zoomsout

2002

Spacetree

(Grosjean

etal2002)

Hierarchyexploration

tool

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2006

Dynapad(Bauer

2006)

Supportsorganisationof

personalcollections

Dynamicveryflexible

No

Multi-touch

2007

Apple

iPhone(2010)

Applicationlauncher

for

Applersquosmobilephone

Containsiconslaid

out

inafixed

grid

No

Tapto

zoom

inPhysical

buttonto

zoom

out

2008

MicrosoftpptPlex(2010)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Nobutwithmanual

override

Click

onslidefornext

Click

tozoom

inEsc

tozoom

outScroll

wheelfliesinout

2008

Zoomorama(2010)

Clientserver

toenable

high-perform

ance

zoomingofim

ages

over

anetwork

Dynamicallygenerated

gridcontaining

images

No

Pop-upon-screenbut-

tonsforzoom

inout

(continued)

858 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

(continued)

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

2008

Prezi

(2010)

Website

forcreatingand

makingfree-form

presentations

Unconstrained

Drawingsim

ages

moviestext

positioned

manually

Yes

TabfornextScroll

wheelon-screen

buttonsandmouse

keyboard

comboto

flyinout

2009

Zumobi(2010)

ZoomCanvas

Launcher

forasm

all

number

ofrelated

contentareason

high-endmobile

devices

Fixed

hand-tuned

layout

turned

tocontent

No

Tapto

zoom

in

on-screenbuttonto

zoom

out

2009

Microsoft

Canvasfor

OneN

ote

(2010)

Plug-inforMicrosoft

OneN

ote

showsvisual

overview

ofalla

userrsquosnotes

Dynamicallygenerated

gridofgrids

containingim

ages

of

notesfrom

Microsoft

OneN

ote

Nobutwithmanual

override

Click

onnote

tozoom

inEscrightclickor

bgndclickto

zoom

out

2009

Schem

atic(2010)

Example

ofcustom

website

withfixed

layoutoffixed

content

Fixed

hand-tuned

layout

tuned

tocontent

No

Click

onslideto

zoom

inClick

on

backgroundto

zoom

out

2009

Microsoft

Pivot(2010)

Facetedsearchtoolthat

useszoomingto

navi-

gate

throughsearch

results

Fixed

gridorgridwithin

barchart

layouts

No

Multiple

mechanisms

includingclickingand

on-screenbuttons

NoteNote

thatItheauthoram

acofounder

ofZumobiInc

Behaviour amp Information Technology 859

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

out The study varied whether or not users were shownoverviews and whether the map information wassingle or multi-level That is single level mapsdisplayed all textual information in a single font sizeso when the user zoomed out the text could not beread Multi-level maps displayed textual informationat different sizes Text referring to larger areas on themap was displayed at much larger fonts so that whenusers zoomed out they could read the text coveringlarger geographical areas

Before this study it was believed that overviewswere a good idea and that they improved subjectivesatisfaction (North and Shneiderman 2000) andefficiency (Beard and Walker 1990 Bauer 2006)However this study found a discrepancy betweenpreference and performance Study participants pre-ferred the overview condition but actually performedbetter in the no-overview condition especially whenusing the multi-level map

It turns out that switching between overview anddetail windows was correlated with higher taskcompletion time suggesting that integration of over-view and detail windows requires mental and motoreffort

This result is notable because a fundamentalconcern about ZUIs is that they require time to use(since the animations take time to occur) Worse thereis the potential that ZUIs tax human short-termmemory (STM) because users must integrate in theirheads the spatial layout of the information On theother hand there is the hope that the animatedtransitions are understood preconsciously by the hu-man visual system and thus may not in fact place asignificant load on STM

This study gives some evidence that the cost ofunderstanding an animated zoomable map of informa-tion is in fact less than the use of one with an overviewthat is one of the primary alternative interface designsOf course it does not say anything about the benefit orcost of a spatially organised information spacecompared to a non-spatially organised space It alsodoes not say anything about the benefit or cost ofanimation

242 Animation

A second study looked at simple animated navigationtransitions and found that these animations signifi-cantly increased performance ndash even including the timethe animations took These kinds of animations havelong been expected to provide benefit (Card et al1991) but evidence backing up this understanding hadbeen lacking Klein and Bederson (2005) looked at avery simple but highly controlled use of animationwhen scrolling while reading documents

Understanding animation precisely is important be-cause the potential benefit of animation is underminedby the fact that animation by its nature takes time Soeven if the animation provides some benefit there is acost as well and it is not obvious whether the benefitoutweighs the cost

Participants in this study read documents out loudand when they got to the end of a page they scrolled bypressing the down arrow key The animation speedvaried between 0 (ie unanimated) 100 300 and 500 msThe document type also varied between unformattedtext formatted text and abstract symbols (whereparticipants counted symbols instead of reading) Thereason for varying the document type was to understandhow visual landmarks interact with animation

The essential result was that animation did in factprovide a significant benefit even considering the timespent on the animation Reading errors were reducedby 54 with 500 ms animations Reading task timewas reduced by 3 and counting task time wasreduced by 24 for 300 ms animations The formatteddocuments had a higher overall performance (andlower improvement) implying that visual landmarksappear to be a good idea which is consistent with theHornbaeligk studyrsquos observation about multi-level maps

The lesson from this study is that when designedcorrectly the benefits of simple animated navigationaltransitions can outweigh their costs This supports thedesign decision to make zooming transitions use shortanimations This study is not definitive however since itis possible that the benefit that was seen while scrollingwould not occur while zooming However there is noobvious reason that the same benefits would not occurso this kind of animated transition continues to seemvaluable

243 Presentations

Two other studies looked at how people learned fromzoomable spaces (Boltman et al 2002 Good andBederson 2002) The outcomes here were more mixedwhere the zoomable presentation did not producebetter learning or memory of the presentation ndashalthough participants did remember the structure ofthe presentation better The first study looked at howchildren responded to a story that was presented withor without animated zooming The children with theanimated condition elaborated more during a discus-sion but did not otherwise recall the story better Thesecond looked at how college students responded to aslide presentation ndash again with or without animatedzooming The presentations were structured visually ina 2D hierarchical format with six sections arrangedin a circle where each section contained slides arrangedin a circle These students also did not recall the

860 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

presentation better but the ones who saw the animatedzooming condition did recall the structure of thepresentation better

Together these studies imply that zooming alone isprobably not enough to help people remember contentbetter However users of ZUIs may be more engagedand they may remember the spatial structure of thecontent better ndash even if they do not remember theactual content itself any better However the value ofremembering structure depends on the task If thestructure has value in and of itself then this is apositive value If on the other hand the visualstructure was added as a means to make the presenta-tion more visually interesting and had no intrinsicvalue then such improved retention would provide nobenefit

244 Zooming vs multiple windows

Two related papers (Plumlee and Ware 2002 Plumleeand Ware 2006) look more generally at modelling andthen studying how people perform visual comparisonsof multi-scale data using zoomable and multi-windowinterfaces Informed by an understanding of theextreme limitations of human visual working memorythey modelled how people can search using the twointerfaces The model showed that comparison tasksthat exceeded the capacity of visual working memory(which can hold roughly three visual objects) would bebetter served by adding windows Multiple windowsdecrease visual working memory load because differentviews can be directly compared In contrast ZUIsrequire the user to hold everything outside of thecurrent view in their memory ndash thus making theminappropriate for complex comparison tasks Thestudies that Plumlee and Ware performed demon-strated this essential characteristic and the trade-offbetween ZUIs and multiple windows closely matchedtheir models They also showed that users of ZUIs hadsignificantly higher error rates than users of multiple

windows The practical impact of these studies dependson the frequency and difficulty of performing zoomingcompared to creating and managing windows And italso applies only to visual comparison tasks

So again the lessons are interesting and importantto specific usage cases but there remains no studytelling us lsquowhether zooming is goodrsquo The real answer isthat it depends

3 Discussion

A trend in ZUI applications that appears to be fairlyclear over the years is that over time these applicationshave tended to get simpler and easier to use while atthe same time offering less creative control to endusers This is an important trend because researchersoften have a tendency to create tools that are rich andpowerful After all it is easier to conceptualise aninnovation as offering lsquomorersquo rather than offering lsquolessrsquo

But the marketplace is clear ndash success often requiressimplicity first and power second Of course marryingthe two where the power is under the hood is often thebest solution ndash but also very difficult to design Thesuccess of so many modern applications (eg searchtwitter photo sharing iPhone App Store) are examplesof this in that they can be used with almost no trainingin tiny little bursts

I used to think that only public kiosks had to havethese simplistic characteristics ndash but now it seems thateven applications that people use for hours a week andbecome true experts in follow the same rules If truethis implies that we as researchers have to stronglyconsider not only what new things technologies canhelp people do but also consider simplicity and speedas fundamental goals

31 Why ZUIs are challenging

As summarised in Table 2 the potential benefits ofZUIs are sometimes mirages ZUIs are generally

Table 2 Promise and reality of ZUIs

Promise Reality

Engaging If designed well The animated nature of ZUIs is sometimes referred to as lsquoeye candyrsquo implying that peoplecannot help but pay attention to the movement on the screen However this very movement is a double-edgedsword because if the user does not pay attention and misses the animation then the connection between thedifferent views is lost and the user is likely worse off than without a ZUI

Visually rich Can also make space more complex to navigate and understand ZUIs can have visual objects at any size and atany place ndash thus offering the possibility of placing significant content (scaled down) even in a tiny crackbetween two other objects This kind of structure is very difficult to perceive (since users cannot see the tinycontent in the first place) and conceptualise (since at any given view the vastly different sized objects can notbe seen simultaneously)

Simplicity Does not scale to very large data sets One of the attractions of ZUIs is that they appear conceptually simple(lsquojust zoom in for a closer lookrsquo) But in practice navigating large information spaces in ZUIs is often difficultto execute and requires significant short-term memory to navigate

Behaviour amp Information Technology 861

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

engaging (although they make some people feelphysically sick) and visually rich But the promise ofsimplicity falls short

While human visual perception does make it easy tosee where one is navigating the reality is that it places aheavy load on short-termmemory to remember where inspace you just were and where things are And therequirement of human memory to know how space isorganised means that ZUIs do not scale up very wellZUIs are oftenmotivated by the physical world and howpeople like laying papers out on their desk But no onewants all of their papers on their desk It is much morecommon to have only a relatively small number ofpapers that are actually being worked with

The visual overviews that ZUIs offer for free byzooming out may seem like a solution to the load onhuman memory but in practice it does not becausevisual overviews of any complexity require significantvisual search in order to find anything If there are just asmall number of objects then the visual search task is nothard ndash but of course for a small number of objects youdo not need a ZUI to solve your organisationalproblems

In addition the visual richness of ZUIs is a double-edged sword It requires skill to design a complex spacewith documents of arbitrary size aspect ratio andcolour that people can comprehend and scan Alsopeople are not as good at scanning 2D designs as 1Dlayouts ndash unless the layout is highly structured Buthighly structured 2D layouts do not work well forvisual objects of arbitrary aspect ratios Designers areobligated to leave a lot of unused space scale down thelarge objects so they are unreadable or crop the largeobjects ndash thus losing much of their distinguishability

Finally it is crucial to consider the efficiency ofZUIs and their alternatives It is the nature ofzoomable structures to organise information hierarchi-cally Zooming in to an object shows you moreinformation about that object Of course hierarchicalstructures are also commonly exposed with a simpletree widget as is commonly found in most file systembrowsers The file system browsers exemplified byWindows Explorer are not only very efficient but alsomore easily let the user compare different areas in thehierarchy by simply opening and collapsing the desirednodes So in many cases the alternative to a ZUI is asimple tree widget that is simple compact easy tolearn and fast to use Thus in addition to the broaderchallenges of ZUIs we should always consider thealternatives that are also powerful

32 Design guidelines

So how should ZUIs be designed to best overcomethese obstacles Let us start with the big issues

321 Support the right tasks

ZUIs like most visualisations have strong potentialfor supporting users in understanding the big pictureidentifying trends patterns and outliers But theytypically are not good at helping users simply get thebest answer So be sure that some sort of overview-based design is called for in the first place

322 Be cautious about using zooming as the primaryinterface

Zooming navigation can be difficult and is uncom-mon Unless the data set is fundamentally spatial (egmaps) users probably do not want zooming as theirprimary interaction with the data Instead think abouthow people search and create data sets ndash which maythen be accessed partially by zooming ie createhybrid systems that combine mechanisms such asfaceted search and zooming [eg PhotoMesa (Beder-son 2001) the Hard Rock Memorabilia (2010) orMicrosoft Pivot (2010)]

323 Only use ZUIs when a small visual representa-tion of the data is available

Zooming works best when the objects can be recognisedwhen they are zoomed out So certain domains arebetter suited to ZUIs Photos are good purely textualdocuments are bad and audio recordings are terrible(unless there are associated images) Recent researchshows how a small visual snippet may be automaticallycreated from web pages (Teevan et al 2009)

324 Donrsquot limit yourself to fixed data sets

People rarely want to interact long with a fixed set ofdata But it is often technically difficult to dynamicallycreate and modify large zoomable spaces Be sure thatthe user interface does not get blocked or slowed downwhen new data is added to a zoomable space

And then there are several more specificsuggestions

325 Maintain aspect ratio

It is tempting to use semantic zooming to signifi-cantly change the visual representation of objects atdifferent sizes For example a visual chart may becondensed to a single value or a text document maybe replaced by its title when small However thesmall object must have the same aspect ratio ndash or atleast fit in the same box as the full object Otherwisenearby objects can end up overlapping when the userzooms out

862 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

326 Use meaningful layout

The problem of finding a specific object among a largenumberof themcanbe alleviated if theobjects are laidoutin a semantically meaningful way If the user can predictwhere an object will be based on attributes that they arelikely to be familiar with then they can trim the searchspace Traditional paper phones books with theiralphabetic ordering are a good example of this approach

327 Use consistent layout

Ideally once an object is placed in space it should notmove That way the user can build up an internal mapover time of where objects are Unfortunately thisideal is often in direct conflict with the previous one ofmeaningful layout If objects are to be laid outmeaningfully then the layout must change over timeas the data changes and as the organisationalmechanism changes PhotoMesa (Bederson 2001) isan example of a ZUI that went for a meaningful layoutbut had no consistency at all

328 Use scannable layout

Objects must be positioned in such a manner as to enablethe user to visually scan the space to look for an objectwithout missing or repeating objects Simple 1D lists ofobjects are easy to scanbecause the person just runs downthe list and the only state that must be kept is the currentitem and the direction of scanning In 2D however forany layout except for a grid there is the potential for amuch heavier load on short-term memory

329 Use breadth over depth

It is tempting to place objects in space at manydifferent sizes so more information can be discoveredby zooming in further ndash and in fact many of the earlierZUI applications did this However when objects areplaced at many different levels the user would notknow that the small objects exist (unless the space iscarefully designed to indicate them) and they can getlost Further even if the user knows they are there it istypically burdensome to zoom in and out repeatedlySpaceTree (Grosjean et al 2002) is an example of anapplication that was designed explicitly with breadth inmind Even with hundreds of thousands of objects allof them are at the same size

3210 Use small data sets

The problems that come with a large number of objectsin a zoomable space can be mitigated by not buildingapplications around large data sets ZUIs appear to

work better for users when they have no more thanabout 100 screens worth of information HoweverZUIs with 1000 or more screens worth of informationbecome problematic because of the issues raised in theprevious section (why ZUIs are challenging) This is afundamental problem since ZUIs are often designed towork for user generated data ndash which may not easily belimited So application designers may choose topartition the space into manageable subsets

3211 Use simple and consistent navigationinteraction

People must be able to discover how to navigate thespace with no training and it must be difficult orimpossible to get lost in Desert Fog Unfortunatelythere is significant inconsistency among applicationsThey vary between single and double clicks to zoom inhow much the screen is zoomed per action whetherzooms scale by a fixed percentage or whether the targetobject fills the screen And mechanisms are all over themap to zoom out ndash from right click left click on thebackground fixed on-screen buttons and pop-upbuttons And as with zoom in the actual amountthat is zoomed out is inconsistent across applicationsas well

4 Conclusion

Understanding ZUIs is difficult They have clearlybeen a success by traditional research measures(significant visibility publications and citations) Andelements of ZUIs have been broadly deployed incommonly used applications But the original broadvision of zooming as a basic information organisa-tional principle [such as Raskinrsquos vision of hospitalmanagement (Raskin 2000)] has not happened Ofcourse lack of commercial success does not mean anidea is not a good one (maybe the business reasons didnot line up or maybe the right tools got built but withthe wrong marketing strategy the timing may havebeen wrong or any number of other reasons mighthave stalled a successful business) But still the lack ofbroad success coupled with the fundamental issuesraised in this paper appears significant

I definitely would not rule out further success ofZUIs and I think that the successes that haveoccurred such as Google Maps Microsoft Word andApple iPhone are substantial However their applic-ability is much narrower than original envisioned Andthe design and technical challenges of building effectiveZUIs make it that much less likely that they will bebroadly deployed At the same time there are clearcases where zooming is valuable ndash such as withdocuments that are larger than screens (ie photos

Behaviour amp Information Technology 863

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Systems [eg Pad (Perlin and Fox 1993) Padthornthorn(Bederson and Hollan 1994) Jazz(Bederson et al 2000) and Piccolo (Bedersonet al 2004b)]

Applications [eg PadPrints (Hightower et al1998) PhotoMesa (Bederson 2001) Counter-Point (Good and Bederson 2002) KidPad(Druin et al 1997) AppLensamp LaunchTile(Karlson et al 2005) Microsoft Seadragon(2010) Fly (Lichtschlag et al 2009) MicrosoftpptPlex (2010) Microsoft Canvas for OneNote(2010) Microsoft Pivot (2010) and Prezi (2010)]

Theoretical constructs [eg Space-Scale Diagrams(Furnas and Bederson 1995) and Desert Fog (Juland Furnas 1998)]

Studies (eg Combs and Bederson 1999 Boltmanet al 2002 Hornbaeligk et al 2002 Klein andBederson 2005 and Plumlee and Ware 2006) tosupport and understand them

Many of these have had a fair amount ofaccolades and positive user response yet it is fairto say that none of them have been great commercialsuccesses (defined either monetarily or by largenumbers of users)

In fact I would argue that ZUIs have neverreached the level of broad use envisioned by theiroriginal creators There has been a huge amount ofeffort relating to ZUIs with dozens of publishedpapers and a commensurate amount of technologydeveloped Zooming has been successful in that somekind of zooming is commonly used in a wide rangeof interfaces (eg Google Maps Microsoft WordAdobe Photoshop and Apple iPhone) But a richerkind of zooming that takes over the desktop andbecomes the primary interface to onersquos data nevermaterialised

As one of those early creators and as someone whohad worked on many aspects of ZUIs since nearly theirbeginning I think it is worth reflecting on what thatoriginal excitement was about where ZUIs have beensuccessful where they have not been and why Forexample a number of commercial efforts that pursueda deeper kind of zooming had only modest success (egan effort by Sony to include zooming in early VAIOcomputers (Kunkel 1999 pp 160ndash165) and HillcrestLabsrsquo HoME product that has not had significantdistribution (Hillcrest Labs 2010)

In reflecting on this body of work it may bepossible to increase our understanding of why it issometimes harder to bring innovations to broad usethan initially thought And as researchers why weshould perhaps be a bit more tempered in ourenthusiasm ndash while of course not stifling innovationbefore it has the opportunity to flower

21 The early history

All ZUIs were built with the goal of addressing thefundamental issue that this paper started with ndash thatpeople need to interact with more information than fitsthe screen This is such a fundamental issue incomputing and interface design that it seemed excitingand appropriate to explore a new approach to addressthe problem The challenge was that the problem wasvaguely defined ndash without users or tasks ndash and thesolution was technically very difficult This led to severalyears of building general solutions rather than specificones Let us start by looking at the early history

lsquoPadrsquo was the first system that explored this space(Perlin and Fox 1993) (Figure 1) Pad ran on a SunSPARCstation with black and white graphics dis-played one bit per pixel bitmaps and used non-animated lsquojumprsquo zooming (each mouse click wouldredisplay the view magnified or reduced by a factor oftwo) Pad offered navigation authoring semanticzooming and portals The term lsquosemantic zoomingrsquo(coined by David Fox) refers to how objects can havedifferent visual representations at different sizesPortals were rectangular objects on the surface thatacted like cameras that showed other parts of thesurface Portals were designed to solve the limitationthat spatial layout implied It enabled objects that werephysically far apart to be used near each other

Back in 1992 I was finishing my PhD at New YorkUniversity (NYU) watching Perlinrsquos work closely andwhen I joined Bellcore after my graduation I beganbuilding the first of what turned into a series ofsuccessors to Pad While working with Jim Hollan Istarted with Pa3D a ZUI with richer vector andbitmap colour graphics and smoothly animatedzooming ndash running on much more expensive SGIcomputers Notably Pa3D was in a 3D environmentwhere every polygon could be a zoomable surfaceThere are no papers about Pa3D but a video showingit is available online (Bederson 1993)

Pad and Pa3D assumed that information would beplaced at different scales in this gigantic informationspace and users would access the information bypanning and zooming through the space using portalsto connect otherwise distance objects In other wordswe aimed to build a zoomable desktop We made manydemonstrations of how people might organise theirinformation in this space but notably despite ourconviction that zooming was an incredible solution wehad only limited examples of how people wouldperform day to day tasks

Being confident that zooming was a good idea butjust not sure what for I started building Padthornthorn alsowith Jim Hollan (Bederson and Hollan 1994aBederson et al 1996) Padthornthorn was a successor to

Behaviour amp Information Technology 855

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Pa3D which incorporated a well-defined applicationprogrammerrsquos interface (API) so others could buildzoomable applications and hopefully reveal the powerthat zooming offered

Over the years teams I led (with Jon Meyer JesseGrosjean and Aaron Clamage) extended Padthornthorn andthen built ZUI platforms Jazz (Bederson et al 2000)and Piccolo (Bederson et al 2004b Piccolo2D 2010) Wesupported other platforms and languages and evensupported mobile devices with PocketPiccoloNETThere was also a significant effort to supportsmooth zooming high-quality animated graphicsand overall high performance on what at the timewere fairly meagre computing systems (Bederson andMeyer 1998) Many other researchers and a fewcompanies built extensions and applications on theseplatforms (especially Padthornthorn) as did we In more recentyears commercially available platforms such as JavaNET and Flash made it easier to build zoomableapplications and clientserver solutions have also beenbuilt (Microsoft Seadragon Zoomorama GoogleMaps)

One significant effort was by Sony starting in 1998They started an internal effort to produce ZUIs fortheir VAIO PC computers that were new at the time(Kunkel 1999 pp 160ndash165) Franklin Servan-Schrei-ber who led that effort coined the term lsquoZUIrsquo1 and isthe person who led the recently defunct effort atZoomorama (2010) However Sony never shipped anyZUI products related to that effort

Looking at the core original ZUI ideas (zoomingsemantic zooming and portals) it is interesting to seewhich ones eventually had broad usage Basic zoominghas been used widely in everything from maps todocument viewers Semantic zooming has also hadsome broader use ndash even outside the realm of ZUIs Thelsquoribbonrsquo in some Microsoft Office products (such asWord) uses semantic zooming to display more informa-tion when more space is available It dynamicallychanges as you resize the ribbon to show more buttonsand to show more imagery as the panel gets largerSimilarly several on-screen music players show morecontrols and information when the player is enlarged Itis notable that not a single one of these applications usesportals Many applications (such as Microsoft WordGoogle Maps and Adobe Photoshop) offer multipleviews or a small fixed navigation overview Howevernone of these applications that focus on zooming as acore organisational and navigation technique useportals in a way similar to how they were envisioned

22 ZUI characteristics

There are a few essential characteristics of ZUIs thataffect their usefulness and their usability

Different ZUIs have had various levels of layoutflexibility By this I refer to how much creative controlthe end user has over the layout of information in thespace On one extreme users can put anything at anysize and place This yields complete artistic freedombut can be difficult to author ndash and has the potentialfor creating a visual mess that users can get lost in Onthe other extreme the environment might be con-strained allowing information only in specific placesaccording to a grid or layout algorithm A relatedcharacteristic is whether specific layouts are multi-levelby which I mean that objects in the space appear atsignificantly different sizes requiring a user to navigateto different zoom levels in order to see all the objects

Different ZUIs have also had various navigationmechanisms ndash which are ways for users to movethrough the space Again there is a trade-off betweenflexibility and usability Some interfaces allow users tofly through the space going absolutely anywhere ndashincluding deep into the spaces between objects [result-ing in some researchers labelling this phenomenaDesert Fog (Jul and Furnas 1998)] Very few otherapplications let a user navigate beyond the actualcontent Almost every document browser and editorlimits navigation to the available content (with thenotable exception of Microsoft Excelrsquos scroll bararrows ndash Apple Numbers and Google Spreadsheeton the other hand do limit navigation) On the otherhand some interfaces allow you only to click onobjects to zoom into them and click on a zoom outbutton to zoom out ndash making it impossible to get lostbut also giving less control over exactly where youlook

One approach to managing the complexity ofnavigation is to zoom automatically when neededusing a technique called speed-dependent automaticzooming (SDAZ)

221 Speed-dependent automatic zooming

One of the reasons that navigation is such a big issue isthat there are too many degrees of freedom to easilycontrol In a traditional computer interface onebutton is typically used for selection or action and ascroll bar or specialised key is used for navigation ZUIusers however must pan along two axes instead of justone and they must be able to zoom in and out Formost applications this must be done with the samehardware used by non-zoomable interfaces

Some researchers addressed this by reducing thenumber of navigational degrees of freedom that usersneeded to control in the first place (Igarashi andHinckley 2000 Cockburn et al 2005 Sun andGuimbretiere 2005) One of the reasons for zoomingout is simply to make it faster to pan a large distance

856 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

[as explained with space-scale diagrams (Furnas andBederson 1995) and elaborated on by Van Wijk andNuij (2004)] This observation led to SDAZ where theapplication would automatically zoom out when theuser panned quickly and then zoom back in whenthe user slowed down

Several studies showed the effectiveness of thisapproach (Cockburn et al 2005 Sun and Guimbretiere2005) and yet not a single research or commercial ZUIapplication (outside those that are explicitly exploringSDAZ) uses SDAZ as even an optional navigationmechanism Why is this There is not a clear answerbut it is important to observe that while SDAZsimplifies one thing (degrees of freedom of control)it complicates another (device control and perception)

SDAZ requires the user to engage in a real-timeperceive-think-act loop (Card et al 1986) This real-time interactive manipulation was typical of the lsquoflyingrsquoin many of the earliest ZUI systems ndash and that causedsignificant usability problems People who play videogames like the interaction challenge of real-timeinteraction People doing information processing taskstypically do not They are more likely to want to spendtheir cognitive resources on the underlying task not onanalysing the interface ndash even if it results in them beingslightly slower

222 Mouse wheels

While SDAZ is a somewhat complicated way to controlzooming some clear winners for controlling zoominghave emerged in the marketplace and they are themouse wheel and multi-touch lsquopinchingrsquo The mousewheel has become fairly widespread and is used fornumerous linear navigation actions ndash from scrolling toselection to zooming It offers simple control over theextra degree of freedom and while it is somewhat slow(requiring movement for each further bit of zooming)its simplicity makes it trivial to learn The fact thatzooming navigation often requires only a small scalechange makes for a perfect mapping to themouse wheel

223 Touch screens and multi-touch interaction

The use of multi-touch lsquopinchrsquo gestures on touchscreens is another control mechanism with growingpopularity Ishii and Ullmer originally envisioned thisapproach by indirectly using physical objects to scaleand rotate images on a screen in 1997 (Ishii and Ullmer1997) Then in 2002 Rekimoto developed a directsolution by using multiple fingers on a capacitive touchscreen to scale and rotate images (Rekimoto 2002)Finally this approach was commercialised and is nowbroadly used in Applersquos iPhone and other mobilephones tables as well as Microsoft Surface This kind

of direct multi-touch interaction is easy to learn andoperate More generally the rapidly growing use ofhigh-end touch screen-based smartphones is increasingthe need for zooming (with people frequently readingdocuments designed for larger screens on smallscreens) and pinching seems to be the de factostandard for controlling this

23 Applications

To see something of the range of uses of ZUIs and howtheir characteristics have changed over time Table 1shows a selection of zoomable applications Only lsquotruersquoZUIs are shown (those supporting more than onedocument or object) It captures a range of what peoplehave been using zooming for and makes apparent therange of approaches that people have taken with regardto layout flexibility and navigation It is also clear thatthe essential problem of getting lost in Desert Fog hasnot been consistently avoided Furthermore it is clearthat there is no consistency in the mechanisms that areused to navigate through space

Of broader import there is no obvious focus onusers tasks or problem domains This is characteristicof both the potential and the challenge of ZUIsZooming is a completely general strategy and so oneshould not be surprised to see such a lack of focus anymore than one might see scroll bars and search used inan incredibly diverse set of scenarios That said theredoes seem to be a concentration of ZUIs applied todomains with highly visual documents ndash which ofcourse makes sense given the visual nature of ZUIs

24 Studies

Some of these and other ZUI applications have beenstudied for usability or effectiveness However it is thenature of those application-centric studies that it isdifficult to tease out the relative benefit or cost of thezooming characteristics of those applications

In fact there are relatively few studies that look atthe value of animated zooming more generally But afew are particularly useful so let us look at them

241 Overviews

The first study by Hornbaeligk et al (2002) looked atmulti-scale information structure as well as the use ofsmall overview windows that show a zoomed out viewof the information Surprisingly it found that whileparticipants like the overview window its use resultedin lower performance The study used geographicalmaps as the data set Navigation worked by draggingthe mouse to pan holding down the left button tozoom in and holding down the right button to zoom

Behaviour amp Information Technology 857

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

SomeZUIsystem

sandapplications

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

1993

Pad(PerlinandFox

1993)

First

ZUIplatform

included

several

conceptualisationsof

usagescenarios

Unconstrained

Yes

Leftclickto

zoom

in

Rightclickto

zoom

out

1994

Padthornthorn

(Bederson

andHollan1994)

SecondZUIplatform

included

APIfor

othersto

write

ZUI

apps

Unconstrained

Yes

Leftclickandhold

tozoom

inRightclick

andhold

tozoom

out

1997

KidPad(D

ruin

etal

1997)

Useslsquolocaltoolsrsquo

(Bedersonet

al1996)

toenable

childrento

create

stories

Unconstrained

Drawingsim

ages

textpositioned

manually

Yes

Hyperlinksforpath

Zoom

inoutmodes

Click

toflyinout

1998

PadPrints

(Hightower

etal1998)

Createsavisual

hierarchicalmapof

web

pages

visited

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2001

CounterPoint(G

ood

andBederson2002)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Yes

Click

onslidefornext

Rightclick-and-hold

fliesinout

2001

PhotoMesa(Bederson

2001)

Apersonalphoto

browserdesktopand

mobile

Containsphotosin

adynamicallygenerated

groupofgridsusinga

treemapalgorithm

No

Leftclickzoomsin

Rightclickzoomsout

2002

Seadragon(2010)

Clientserver

high

perfZUIforim

ages

over

anetwork

(web

andmobile)

Containsphotoslaid

out

byasm

allnumber

of

fixed

layout

algorithms

No

Click

tozoom

in

On-screenbutton

zoomsout

2002

Spacetree

(Grosjean

etal2002)

Hierarchyexploration

tool

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2006

Dynapad(Bauer

2006)

Supportsorganisationof

personalcollections

Dynamicveryflexible

No

Multi-touch

2007

Apple

iPhone(2010)

Applicationlauncher

for

Applersquosmobilephone

Containsiconslaid

out

inafixed

grid

No

Tapto

zoom

inPhysical

buttonto

zoom

out

2008

MicrosoftpptPlex(2010)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Nobutwithmanual

override

Click

onslidefornext

Click

tozoom

inEsc

tozoom

outScroll

wheelfliesinout

2008

Zoomorama(2010)

Clientserver

toenable

high-perform

ance

zoomingofim

ages

over

anetwork

Dynamicallygenerated

gridcontaining

images

No

Pop-upon-screenbut-

tonsforzoom

inout

(continued)

858 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

(continued)

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

2008

Prezi

(2010)

Website

forcreatingand

makingfree-form

presentations

Unconstrained

Drawingsim

ages

moviestext

positioned

manually

Yes

TabfornextScroll

wheelon-screen

buttonsandmouse

keyboard

comboto

flyinout

2009

Zumobi(2010)

ZoomCanvas

Launcher

forasm

all

number

ofrelated

contentareason

high-endmobile

devices

Fixed

hand-tuned

layout

turned

tocontent

No

Tapto

zoom

in

on-screenbuttonto

zoom

out

2009

Microsoft

Canvasfor

OneN

ote

(2010)

Plug-inforMicrosoft

OneN

ote

showsvisual

overview

ofalla

userrsquosnotes

Dynamicallygenerated

gridofgrids

containingim

ages

of

notesfrom

Microsoft

OneN

ote

Nobutwithmanual

override

Click

onnote

tozoom

inEscrightclickor

bgndclickto

zoom

out

2009

Schem

atic(2010)

Example

ofcustom

website

withfixed

layoutoffixed

content

Fixed

hand-tuned

layout

tuned

tocontent

No

Click

onslideto

zoom

inClick

on

backgroundto

zoom

out

2009

Microsoft

Pivot(2010)

Facetedsearchtoolthat

useszoomingto

navi-

gate

throughsearch

results

Fixed

gridorgridwithin

barchart

layouts

No

Multiple

mechanisms

includingclickingand

on-screenbuttons

NoteNote

thatItheauthoram

acofounder

ofZumobiInc

Behaviour amp Information Technology 859

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

out The study varied whether or not users were shownoverviews and whether the map information wassingle or multi-level That is single level mapsdisplayed all textual information in a single font sizeso when the user zoomed out the text could not beread Multi-level maps displayed textual informationat different sizes Text referring to larger areas on themap was displayed at much larger fonts so that whenusers zoomed out they could read the text coveringlarger geographical areas

Before this study it was believed that overviewswere a good idea and that they improved subjectivesatisfaction (North and Shneiderman 2000) andefficiency (Beard and Walker 1990 Bauer 2006)However this study found a discrepancy betweenpreference and performance Study participants pre-ferred the overview condition but actually performedbetter in the no-overview condition especially whenusing the multi-level map

It turns out that switching between overview anddetail windows was correlated with higher taskcompletion time suggesting that integration of over-view and detail windows requires mental and motoreffort

This result is notable because a fundamentalconcern about ZUIs is that they require time to use(since the animations take time to occur) Worse thereis the potential that ZUIs tax human short-termmemory (STM) because users must integrate in theirheads the spatial layout of the information On theother hand there is the hope that the animatedtransitions are understood preconsciously by the hu-man visual system and thus may not in fact place asignificant load on STM

This study gives some evidence that the cost ofunderstanding an animated zoomable map of informa-tion is in fact less than the use of one with an overviewthat is one of the primary alternative interface designsOf course it does not say anything about the benefit orcost of a spatially organised information spacecompared to a non-spatially organised space It alsodoes not say anything about the benefit or cost ofanimation

242 Animation

A second study looked at simple animated navigationtransitions and found that these animations signifi-cantly increased performance ndash even including the timethe animations took These kinds of animations havelong been expected to provide benefit (Card et al1991) but evidence backing up this understanding hadbeen lacking Klein and Bederson (2005) looked at avery simple but highly controlled use of animationwhen scrolling while reading documents

Understanding animation precisely is important be-cause the potential benefit of animation is underminedby the fact that animation by its nature takes time Soeven if the animation provides some benefit there is acost as well and it is not obvious whether the benefitoutweighs the cost

Participants in this study read documents out loudand when they got to the end of a page they scrolled bypressing the down arrow key The animation speedvaried between 0 (ie unanimated) 100 300 and 500 msThe document type also varied between unformattedtext formatted text and abstract symbols (whereparticipants counted symbols instead of reading) Thereason for varying the document type was to understandhow visual landmarks interact with animation

The essential result was that animation did in factprovide a significant benefit even considering the timespent on the animation Reading errors were reducedby 54 with 500 ms animations Reading task timewas reduced by 3 and counting task time wasreduced by 24 for 300 ms animations The formatteddocuments had a higher overall performance (andlower improvement) implying that visual landmarksappear to be a good idea which is consistent with theHornbaeligk studyrsquos observation about multi-level maps

The lesson from this study is that when designedcorrectly the benefits of simple animated navigationaltransitions can outweigh their costs This supports thedesign decision to make zooming transitions use shortanimations This study is not definitive however since itis possible that the benefit that was seen while scrollingwould not occur while zooming However there is noobvious reason that the same benefits would not occurso this kind of animated transition continues to seemvaluable

243 Presentations

Two other studies looked at how people learned fromzoomable spaces (Boltman et al 2002 Good andBederson 2002) The outcomes here were more mixedwhere the zoomable presentation did not producebetter learning or memory of the presentation ndashalthough participants did remember the structure ofthe presentation better The first study looked at howchildren responded to a story that was presented withor without animated zooming The children with theanimated condition elaborated more during a discus-sion but did not otherwise recall the story better Thesecond looked at how college students responded to aslide presentation ndash again with or without animatedzooming The presentations were structured visually ina 2D hierarchical format with six sections arrangedin a circle where each section contained slides arrangedin a circle These students also did not recall the

860 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

presentation better but the ones who saw the animatedzooming condition did recall the structure of thepresentation better

Together these studies imply that zooming alone isprobably not enough to help people remember contentbetter However users of ZUIs may be more engagedand they may remember the spatial structure of thecontent better ndash even if they do not remember theactual content itself any better However the value ofremembering structure depends on the task If thestructure has value in and of itself then this is apositive value If on the other hand the visualstructure was added as a means to make the presenta-tion more visually interesting and had no intrinsicvalue then such improved retention would provide nobenefit

244 Zooming vs multiple windows

Two related papers (Plumlee and Ware 2002 Plumleeand Ware 2006) look more generally at modelling andthen studying how people perform visual comparisonsof multi-scale data using zoomable and multi-windowinterfaces Informed by an understanding of theextreme limitations of human visual working memorythey modelled how people can search using the twointerfaces The model showed that comparison tasksthat exceeded the capacity of visual working memory(which can hold roughly three visual objects) would bebetter served by adding windows Multiple windowsdecrease visual working memory load because differentviews can be directly compared In contrast ZUIsrequire the user to hold everything outside of thecurrent view in their memory ndash thus making theminappropriate for complex comparison tasks Thestudies that Plumlee and Ware performed demon-strated this essential characteristic and the trade-offbetween ZUIs and multiple windows closely matchedtheir models They also showed that users of ZUIs hadsignificantly higher error rates than users of multiple

windows The practical impact of these studies dependson the frequency and difficulty of performing zoomingcompared to creating and managing windows And italso applies only to visual comparison tasks

So again the lessons are interesting and importantto specific usage cases but there remains no studytelling us lsquowhether zooming is goodrsquo The real answer isthat it depends

3 Discussion

A trend in ZUI applications that appears to be fairlyclear over the years is that over time these applicationshave tended to get simpler and easier to use while atthe same time offering less creative control to endusers This is an important trend because researchersoften have a tendency to create tools that are rich andpowerful After all it is easier to conceptualise aninnovation as offering lsquomorersquo rather than offering lsquolessrsquo

But the marketplace is clear ndash success often requiressimplicity first and power second Of course marryingthe two where the power is under the hood is often thebest solution ndash but also very difficult to design Thesuccess of so many modern applications (eg searchtwitter photo sharing iPhone App Store) are examplesof this in that they can be used with almost no trainingin tiny little bursts

I used to think that only public kiosks had to havethese simplistic characteristics ndash but now it seems thateven applications that people use for hours a week andbecome true experts in follow the same rules If truethis implies that we as researchers have to stronglyconsider not only what new things technologies canhelp people do but also consider simplicity and speedas fundamental goals

31 Why ZUIs are challenging

As summarised in Table 2 the potential benefits ofZUIs are sometimes mirages ZUIs are generally

Table 2 Promise and reality of ZUIs

Promise Reality

Engaging If designed well The animated nature of ZUIs is sometimes referred to as lsquoeye candyrsquo implying that peoplecannot help but pay attention to the movement on the screen However this very movement is a double-edgedsword because if the user does not pay attention and misses the animation then the connection between thedifferent views is lost and the user is likely worse off than without a ZUI

Visually rich Can also make space more complex to navigate and understand ZUIs can have visual objects at any size and atany place ndash thus offering the possibility of placing significant content (scaled down) even in a tiny crackbetween two other objects This kind of structure is very difficult to perceive (since users cannot see the tinycontent in the first place) and conceptualise (since at any given view the vastly different sized objects can notbe seen simultaneously)

Simplicity Does not scale to very large data sets One of the attractions of ZUIs is that they appear conceptually simple(lsquojust zoom in for a closer lookrsquo) But in practice navigating large information spaces in ZUIs is often difficultto execute and requires significant short-term memory to navigate

Behaviour amp Information Technology 861

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

engaging (although they make some people feelphysically sick) and visually rich But the promise ofsimplicity falls short

While human visual perception does make it easy tosee where one is navigating the reality is that it places aheavy load on short-termmemory to remember where inspace you just were and where things are And therequirement of human memory to know how space isorganised means that ZUIs do not scale up very wellZUIs are oftenmotivated by the physical world and howpeople like laying papers out on their desk But no onewants all of their papers on their desk It is much morecommon to have only a relatively small number ofpapers that are actually being worked with

The visual overviews that ZUIs offer for free byzooming out may seem like a solution to the load onhuman memory but in practice it does not becausevisual overviews of any complexity require significantvisual search in order to find anything If there are just asmall number of objects then the visual search task is nothard ndash but of course for a small number of objects youdo not need a ZUI to solve your organisationalproblems

In addition the visual richness of ZUIs is a double-edged sword It requires skill to design a complex spacewith documents of arbitrary size aspect ratio andcolour that people can comprehend and scan Alsopeople are not as good at scanning 2D designs as 1Dlayouts ndash unless the layout is highly structured Buthighly structured 2D layouts do not work well forvisual objects of arbitrary aspect ratios Designers areobligated to leave a lot of unused space scale down thelarge objects so they are unreadable or crop the largeobjects ndash thus losing much of their distinguishability

Finally it is crucial to consider the efficiency ofZUIs and their alternatives It is the nature ofzoomable structures to organise information hierarchi-cally Zooming in to an object shows you moreinformation about that object Of course hierarchicalstructures are also commonly exposed with a simpletree widget as is commonly found in most file systembrowsers The file system browsers exemplified byWindows Explorer are not only very efficient but alsomore easily let the user compare different areas in thehierarchy by simply opening and collapsing the desirednodes So in many cases the alternative to a ZUI is asimple tree widget that is simple compact easy tolearn and fast to use Thus in addition to the broaderchallenges of ZUIs we should always consider thealternatives that are also powerful

32 Design guidelines

So how should ZUIs be designed to best overcomethese obstacles Let us start with the big issues

321 Support the right tasks

ZUIs like most visualisations have strong potentialfor supporting users in understanding the big pictureidentifying trends patterns and outliers But theytypically are not good at helping users simply get thebest answer So be sure that some sort of overview-based design is called for in the first place

322 Be cautious about using zooming as the primaryinterface

Zooming navigation can be difficult and is uncom-mon Unless the data set is fundamentally spatial (egmaps) users probably do not want zooming as theirprimary interaction with the data Instead think abouthow people search and create data sets ndash which maythen be accessed partially by zooming ie createhybrid systems that combine mechanisms such asfaceted search and zooming [eg PhotoMesa (Beder-son 2001) the Hard Rock Memorabilia (2010) orMicrosoft Pivot (2010)]

323 Only use ZUIs when a small visual representa-tion of the data is available

Zooming works best when the objects can be recognisedwhen they are zoomed out So certain domains arebetter suited to ZUIs Photos are good purely textualdocuments are bad and audio recordings are terrible(unless there are associated images) Recent researchshows how a small visual snippet may be automaticallycreated from web pages (Teevan et al 2009)

324 Donrsquot limit yourself to fixed data sets

People rarely want to interact long with a fixed set ofdata But it is often technically difficult to dynamicallycreate and modify large zoomable spaces Be sure thatthe user interface does not get blocked or slowed downwhen new data is added to a zoomable space

And then there are several more specificsuggestions

325 Maintain aspect ratio

It is tempting to use semantic zooming to signifi-cantly change the visual representation of objects atdifferent sizes For example a visual chart may becondensed to a single value or a text document maybe replaced by its title when small However thesmall object must have the same aspect ratio ndash or atleast fit in the same box as the full object Otherwisenearby objects can end up overlapping when the userzooms out

862 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

326 Use meaningful layout

The problem of finding a specific object among a largenumberof themcanbe alleviated if theobjects are laidoutin a semantically meaningful way If the user can predictwhere an object will be based on attributes that they arelikely to be familiar with then they can trim the searchspace Traditional paper phones books with theiralphabetic ordering are a good example of this approach

327 Use consistent layout

Ideally once an object is placed in space it should notmove That way the user can build up an internal mapover time of where objects are Unfortunately thisideal is often in direct conflict with the previous one ofmeaningful layout If objects are to be laid outmeaningfully then the layout must change over timeas the data changes and as the organisationalmechanism changes PhotoMesa (Bederson 2001) isan example of a ZUI that went for a meaningful layoutbut had no consistency at all

328 Use scannable layout

Objects must be positioned in such a manner as to enablethe user to visually scan the space to look for an objectwithout missing or repeating objects Simple 1D lists ofobjects are easy to scanbecause the person just runs downthe list and the only state that must be kept is the currentitem and the direction of scanning In 2D however forany layout except for a grid there is the potential for amuch heavier load on short-term memory

329 Use breadth over depth

It is tempting to place objects in space at manydifferent sizes so more information can be discoveredby zooming in further ndash and in fact many of the earlierZUI applications did this However when objects areplaced at many different levels the user would notknow that the small objects exist (unless the space iscarefully designed to indicate them) and they can getlost Further even if the user knows they are there it istypically burdensome to zoom in and out repeatedlySpaceTree (Grosjean et al 2002) is an example of anapplication that was designed explicitly with breadth inmind Even with hundreds of thousands of objects allof them are at the same size

3210 Use small data sets

The problems that come with a large number of objectsin a zoomable space can be mitigated by not buildingapplications around large data sets ZUIs appear to

work better for users when they have no more thanabout 100 screens worth of information HoweverZUIs with 1000 or more screens worth of informationbecome problematic because of the issues raised in theprevious section (why ZUIs are challenging) This is afundamental problem since ZUIs are often designed towork for user generated data ndash which may not easily belimited So application designers may choose topartition the space into manageable subsets

3211 Use simple and consistent navigationinteraction

People must be able to discover how to navigate thespace with no training and it must be difficult orimpossible to get lost in Desert Fog Unfortunatelythere is significant inconsistency among applicationsThey vary between single and double clicks to zoom inhow much the screen is zoomed per action whetherzooms scale by a fixed percentage or whether the targetobject fills the screen And mechanisms are all over themap to zoom out ndash from right click left click on thebackground fixed on-screen buttons and pop-upbuttons And as with zoom in the actual amountthat is zoomed out is inconsistent across applicationsas well

4 Conclusion

Understanding ZUIs is difficult They have clearlybeen a success by traditional research measures(significant visibility publications and citations) Andelements of ZUIs have been broadly deployed incommonly used applications But the original broadvision of zooming as a basic information organisa-tional principle [such as Raskinrsquos vision of hospitalmanagement (Raskin 2000)] has not happened Ofcourse lack of commercial success does not mean anidea is not a good one (maybe the business reasons didnot line up or maybe the right tools got built but withthe wrong marketing strategy the timing may havebeen wrong or any number of other reasons mighthave stalled a successful business) But still the lack ofbroad success coupled with the fundamental issuesraised in this paper appears significant

I definitely would not rule out further success ofZUIs and I think that the successes that haveoccurred such as Google Maps Microsoft Word andApple iPhone are substantial However their applic-ability is much narrower than original envisioned Andthe design and technical challenges of building effectiveZUIs make it that much less likely that they will bebroadly deployed At the same time there are clearcases where zooming is valuable ndash such as withdocuments that are larger than screens (ie photos

Behaviour amp Information Technology 863

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Pa3D which incorporated a well-defined applicationprogrammerrsquos interface (API) so others could buildzoomable applications and hopefully reveal the powerthat zooming offered

Over the years teams I led (with Jon Meyer JesseGrosjean and Aaron Clamage) extended Padthornthorn andthen built ZUI platforms Jazz (Bederson et al 2000)and Piccolo (Bederson et al 2004b Piccolo2D 2010) Wesupported other platforms and languages and evensupported mobile devices with PocketPiccoloNETThere was also a significant effort to supportsmooth zooming high-quality animated graphicsand overall high performance on what at the timewere fairly meagre computing systems (Bederson andMeyer 1998) Many other researchers and a fewcompanies built extensions and applications on theseplatforms (especially Padthornthorn) as did we In more recentyears commercially available platforms such as JavaNET and Flash made it easier to build zoomableapplications and clientserver solutions have also beenbuilt (Microsoft Seadragon Zoomorama GoogleMaps)

One significant effort was by Sony starting in 1998They started an internal effort to produce ZUIs fortheir VAIO PC computers that were new at the time(Kunkel 1999 pp 160ndash165) Franklin Servan-Schrei-ber who led that effort coined the term lsquoZUIrsquo1 and isthe person who led the recently defunct effort atZoomorama (2010) However Sony never shipped anyZUI products related to that effort

Looking at the core original ZUI ideas (zoomingsemantic zooming and portals) it is interesting to seewhich ones eventually had broad usage Basic zoominghas been used widely in everything from maps todocument viewers Semantic zooming has also hadsome broader use ndash even outside the realm of ZUIs Thelsquoribbonrsquo in some Microsoft Office products (such asWord) uses semantic zooming to display more informa-tion when more space is available It dynamicallychanges as you resize the ribbon to show more buttonsand to show more imagery as the panel gets largerSimilarly several on-screen music players show morecontrols and information when the player is enlarged Itis notable that not a single one of these applications usesportals Many applications (such as Microsoft WordGoogle Maps and Adobe Photoshop) offer multipleviews or a small fixed navigation overview Howevernone of these applications that focus on zooming as acore organisational and navigation technique useportals in a way similar to how they were envisioned

22 ZUI characteristics

There are a few essential characteristics of ZUIs thataffect their usefulness and their usability

Different ZUIs have had various levels of layoutflexibility By this I refer to how much creative controlthe end user has over the layout of information in thespace On one extreme users can put anything at anysize and place This yields complete artistic freedombut can be difficult to author ndash and has the potentialfor creating a visual mess that users can get lost in Onthe other extreme the environment might be con-strained allowing information only in specific placesaccording to a grid or layout algorithm A relatedcharacteristic is whether specific layouts are multi-levelby which I mean that objects in the space appear atsignificantly different sizes requiring a user to navigateto different zoom levels in order to see all the objects

Different ZUIs have also had various navigationmechanisms ndash which are ways for users to movethrough the space Again there is a trade-off betweenflexibility and usability Some interfaces allow users tofly through the space going absolutely anywhere ndashincluding deep into the spaces between objects [result-ing in some researchers labelling this phenomenaDesert Fog (Jul and Furnas 1998)] Very few otherapplications let a user navigate beyond the actualcontent Almost every document browser and editorlimits navigation to the available content (with thenotable exception of Microsoft Excelrsquos scroll bararrows ndash Apple Numbers and Google Spreadsheeton the other hand do limit navigation) On the otherhand some interfaces allow you only to click onobjects to zoom into them and click on a zoom outbutton to zoom out ndash making it impossible to get lostbut also giving less control over exactly where youlook

One approach to managing the complexity ofnavigation is to zoom automatically when neededusing a technique called speed-dependent automaticzooming (SDAZ)

221 Speed-dependent automatic zooming

One of the reasons that navigation is such a big issue isthat there are too many degrees of freedom to easilycontrol In a traditional computer interface onebutton is typically used for selection or action and ascroll bar or specialised key is used for navigation ZUIusers however must pan along two axes instead of justone and they must be able to zoom in and out Formost applications this must be done with the samehardware used by non-zoomable interfaces

Some researchers addressed this by reducing thenumber of navigational degrees of freedom that usersneeded to control in the first place (Igarashi andHinckley 2000 Cockburn et al 2005 Sun andGuimbretiere 2005) One of the reasons for zoomingout is simply to make it faster to pan a large distance

856 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

[as explained with space-scale diagrams (Furnas andBederson 1995) and elaborated on by Van Wijk andNuij (2004)] This observation led to SDAZ where theapplication would automatically zoom out when theuser panned quickly and then zoom back in whenthe user slowed down

Several studies showed the effectiveness of thisapproach (Cockburn et al 2005 Sun and Guimbretiere2005) and yet not a single research or commercial ZUIapplication (outside those that are explicitly exploringSDAZ) uses SDAZ as even an optional navigationmechanism Why is this There is not a clear answerbut it is important to observe that while SDAZsimplifies one thing (degrees of freedom of control)it complicates another (device control and perception)

SDAZ requires the user to engage in a real-timeperceive-think-act loop (Card et al 1986) This real-time interactive manipulation was typical of the lsquoflyingrsquoin many of the earliest ZUI systems ndash and that causedsignificant usability problems People who play videogames like the interaction challenge of real-timeinteraction People doing information processing taskstypically do not They are more likely to want to spendtheir cognitive resources on the underlying task not onanalysing the interface ndash even if it results in them beingslightly slower

222 Mouse wheels

While SDAZ is a somewhat complicated way to controlzooming some clear winners for controlling zoominghave emerged in the marketplace and they are themouse wheel and multi-touch lsquopinchingrsquo The mousewheel has become fairly widespread and is used fornumerous linear navigation actions ndash from scrolling toselection to zooming It offers simple control over theextra degree of freedom and while it is somewhat slow(requiring movement for each further bit of zooming)its simplicity makes it trivial to learn The fact thatzooming navigation often requires only a small scalechange makes for a perfect mapping to themouse wheel

223 Touch screens and multi-touch interaction

The use of multi-touch lsquopinchrsquo gestures on touchscreens is another control mechanism with growingpopularity Ishii and Ullmer originally envisioned thisapproach by indirectly using physical objects to scaleand rotate images on a screen in 1997 (Ishii and Ullmer1997) Then in 2002 Rekimoto developed a directsolution by using multiple fingers on a capacitive touchscreen to scale and rotate images (Rekimoto 2002)Finally this approach was commercialised and is nowbroadly used in Applersquos iPhone and other mobilephones tables as well as Microsoft Surface This kind

of direct multi-touch interaction is easy to learn andoperate More generally the rapidly growing use ofhigh-end touch screen-based smartphones is increasingthe need for zooming (with people frequently readingdocuments designed for larger screens on smallscreens) and pinching seems to be the de factostandard for controlling this

23 Applications

To see something of the range of uses of ZUIs and howtheir characteristics have changed over time Table 1shows a selection of zoomable applications Only lsquotruersquoZUIs are shown (those supporting more than onedocument or object) It captures a range of what peoplehave been using zooming for and makes apparent therange of approaches that people have taken with regardto layout flexibility and navigation It is also clear thatthe essential problem of getting lost in Desert Fog hasnot been consistently avoided Furthermore it is clearthat there is no consistency in the mechanisms that areused to navigate through space

Of broader import there is no obvious focus onusers tasks or problem domains This is characteristicof both the potential and the challenge of ZUIsZooming is a completely general strategy and so oneshould not be surprised to see such a lack of focus anymore than one might see scroll bars and search used inan incredibly diverse set of scenarios That said theredoes seem to be a concentration of ZUIs applied todomains with highly visual documents ndash which ofcourse makes sense given the visual nature of ZUIs

24 Studies

Some of these and other ZUI applications have beenstudied for usability or effectiveness However it is thenature of those application-centric studies that it isdifficult to tease out the relative benefit or cost of thezooming characteristics of those applications

In fact there are relatively few studies that look atthe value of animated zooming more generally But afew are particularly useful so let us look at them

241 Overviews

The first study by Hornbaeligk et al (2002) looked atmulti-scale information structure as well as the use ofsmall overview windows that show a zoomed out viewof the information Surprisingly it found that whileparticipants like the overview window its use resultedin lower performance The study used geographicalmaps as the data set Navigation worked by draggingthe mouse to pan holding down the left button tozoom in and holding down the right button to zoom

Behaviour amp Information Technology 857

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

SomeZUIsystem

sandapplications

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

1993

Pad(PerlinandFox

1993)

First

ZUIplatform

included

several

conceptualisationsof

usagescenarios

Unconstrained

Yes

Leftclickto

zoom

in

Rightclickto

zoom

out

1994

Padthornthorn

(Bederson

andHollan1994)

SecondZUIplatform

included

APIfor

othersto

write

ZUI

apps

Unconstrained

Yes

Leftclickandhold

tozoom

inRightclick

andhold

tozoom

out

1997

KidPad(D

ruin

etal

1997)

Useslsquolocaltoolsrsquo

(Bedersonet

al1996)

toenable

childrento

create

stories

Unconstrained

Drawingsim

ages

textpositioned

manually

Yes

Hyperlinksforpath

Zoom

inoutmodes

Click

toflyinout

1998

PadPrints

(Hightower

etal1998)

Createsavisual

hierarchicalmapof

web

pages

visited

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2001

CounterPoint(G

ood

andBederson2002)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Yes

Click

onslidefornext

Rightclick-and-hold

fliesinout

2001

PhotoMesa(Bederson

2001)

Apersonalphoto

browserdesktopand

mobile

Containsphotosin

adynamicallygenerated

groupofgridsusinga

treemapalgorithm

No

Leftclickzoomsin

Rightclickzoomsout

2002

Seadragon(2010)

Clientserver

high

perfZUIforim

ages

over

anetwork

(web

andmobile)

Containsphotoslaid

out

byasm

allnumber

of

fixed

layout

algorithms

No

Click

tozoom

in

On-screenbutton

zoomsout

2002

Spacetree

(Grosjean

etal2002)

Hierarchyexploration

tool

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2006

Dynapad(Bauer

2006)

Supportsorganisationof

personalcollections

Dynamicveryflexible

No

Multi-touch

2007

Apple

iPhone(2010)

Applicationlauncher

for

Applersquosmobilephone

Containsiconslaid

out

inafixed

grid

No

Tapto

zoom

inPhysical

buttonto

zoom

out

2008

MicrosoftpptPlex(2010)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Nobutwithmanual

override

Click

onslidefornext

Click

tozoom

inEsc

tozoom

outScroll

wheelfliesinout

2008

Zoomorama(2010)

Clientserver

toenable

high-perform

ance

zoomingofim

ages

over

anetwork

Dynamicallygenerated

gridcontaining

images

No

Pop-upon-screenbut-

tonsforzoom

inout

(continued)

858 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

(continued)

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

2008

Prezi

(2010)

Website

forcreatingand

makingfree-form

presentations

Unconstrained

Drawingsim

ages

moviestext

positioned

manually

Yes

TabfornextScroll

wheelon-screen

buttonsandmouse

keyboard

comboto

flyinout

2009

Zumobi(2010)

ZoomCanvas

Launcher

forasm

all

number

ofrelated

contentareason

high-endmobile

devices

Fixed

hand-tuned

layout

turned

tocontent

No

Tapto

zoom

in

on-screenbuttonto

zoom

out

2009

Microsoft

Canvasfor

OneN

ote

(2010)

Plug-inforMicrosoft

OneN

ote

showsvisual

overview

ofalla

userrsquosnotes

Dynamicallygenerated

gridofgrids

containingim

ages

of

notesfrom

Microsoft

OneN

ote

Nobutwithmanual

override

Click

onnote

tozoom

inEscrightclickor

bgndclickto

zoom

out

2009

Schem

atic(2010)

Example

ofcustom

website

withfixed

layoutoffixed

content

Fixed

hand-tuned

layout

tuned

tocontent

No

Click

onslideto

zoom

inClick

on

backgroundto

zoom

out

2009

Microsoft

Pivot(2010)

Facetedsearchtoolthat

useszoomingto

navi-

gate

throughsearch

results

Fixed

gridorgridwithin

barchart

layouts

No

Multiple

mechanisms

includingclickingand

on-screenbuttons

NoteNote

thatItheauthoram

acofounder

ofZumobiInc

Behaviour amp Information Technology 859

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

out The study varied whether or not users were shownoverviews and whether the map information wassingle or multi-level That is single level mapsdisplayed all textual information in a single font sizeso when the user zoomed out the text could not beread Multi-level maps displayed textual informationat different sizes Text referring to larger areas on themap was displayed at much larger fonts so that whenusers zoomed out they could read the text coveringlarger geographical areas

Before this study it was believed that overviewswere a good idea and that they improved subjectivesatisfaction (North and Shneiderman 2000) andefficiency (Beard and Walker 1990 Bauer 2006)However this study found a discrepancy betweenpreference and performance Study participants pre-ferred the overview condition but actually performedbetter in the no-overview condition especially whenusing the multi-level map

It turns out that switching between overview anddetail windows was correlated with higher taskcompletion time suggesting that integration of over-view and detail windows requires mental and motoreffort

This result is notable because a fundamentalconcern about ZUIs is that they require time to use(since the animations take time to occur) Worse thereis the potential that ZUIs tax human short-termmemory (STM) because users must integrate in theirheads the spatial layout of the information On theother hand there is the hope that the animatedtransitions are understood preconsciously by the hu-man visual system and thus may not in fact place asignificant load on STM

This study gives some evidence that the cost ofunderstanding an animated zoomable map of informa-tion is in fact less than the use of one with an overviewthat is one of the primary alternative interface designsOf course it does not say anything about the benefit orcost of a spatially organised information spacecompared to a non-spatially organised space It alsodoes not say anything about the benefit or cost ofanimation

242 Animation

A second study looked at simple animated navigationtransitions and found that these animations signifi-cantly increased performance ndash even including the timethe animations took These kinds of animations havelong been expected to provide benefit (Card et al1991) but evidence backing up this understanding hadbeen lacking Klein and Bederson (2005) looked at avery simple but highly controlled use of animationwhen scrolling while reading documents

Understanding animation precisely is important be-cause the potential benefit of animation is underminedby the fact that animation by its nature takes time Soeven if the animation provides some benefit there is acost as well and it is not obvious whether the benefitoutweighs the cost

Participants in this study read documents out loudand when they got to the end of a page they scrolled bypressing the down arrow key The animation speedvaried between 0 (ie unanimated) 100 300 and 500 msThe document type also varied between unformattedtext formatted text and abstract symbols (whereparticipants counted symbols instead of reading) Thereason for varying the document type was to understandhow visual landmarks interact with animation

The essential result was that animation did in factprovide a significant benefit even considering the timespent on the animation Reading errors were reducedby 54 with 500 ms animations Reading task timewas reduced by 3 and counting task time wasreduced by 24 for 300 ms animations The formatteddocuments had a higher overall performance (andlower improvement) implying that visual landmarksappear to be a good idea which is consistent with theHornbaeligk studyrsquos observation about multi-level maps

The lesson from this study is that when designedcorrectly the benefits of simple animated navigationaltransitions can outweigh their costs This supports thedesign decision to make zooming transitions use shortanimations This study is not definitive however since itis possible that the benefit that was seen while scrollingwould not occur while zooming However there is noobvious reason that the same benefits would not occurso this kind of animated transition continues to seemvaluable

243 Presentations

Two other studies looked at how people learned fromzoomable spaces (Boltman et al 2002 Good andBederson 2002) The outcomes here were more mixedwhere the zoomable presentation did not producebetter learning or memory of the presentation ndashalthough participants did remember the structure ofthe presentation better The first study looked at howchildren responded to a story that was presented withor without animated zooming The children with theanimated condition elaborated more during a discus-sion but did not otherwise recall the story better Thesecond looked at how college students responded to aslide presentation ndash again with or without animatedzooming The presentations were structured visually ina 2D hierarchical format with six sections arrangedin a circle where each section contained slides arrangedin a circle These students also did not recall the

860 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

presentation better but the ones who saw the animatedzooming condition did recall the structure of thepresentation better

Together these studies imply that zooming alone isprobably not enough to help people remember contentbetter However users of ZUIs may be more engagedand they may remember the spatial structure of thecontent better ndash even if they do not remember theactual content itself any better However the value ofremembering structure depends on the task If thestructure has value in and of itself then this is apositive value If on the other hand the visualstructure was added as a means to make the presenta-tion more visually interesting and had no intrinsicvalue then such improved retention would provide nobenefit

244 Zooming vs multiple windows

Two related papers (Plumlee and Ware 2002 Plumleeand Ware 2006) look more generally at modelling andthen studying how people perform visual comparisonsof multi-scale data using zoomable and multi-windowinterfaces Informed by an understanding of theextreme limitations of human visual working memorythey modelled how people can search using the twointerfaces The model showed that comparison tasksthat exceeded the capacity of visual working memory(which can hold roughly three visual objects) would bebetter served by adding windows Multiple windowsdecrease visual working memory load because differentviews can be directly compared In contrast ZUIsrequire the user to hold everything outside of thecurrent view in their memory ndash thus making theminappropriate for complex comparison tasks Thestudies that Plumlee and Ware performed demon-strated this essential characteristic and the trade-offbetween ZUIs and multiple windows closely matchedtheir models They also showed that users of ZUIs hadsignificantly higher error rates than users of multiple

windows The practical impact of these studies dependson the frequency and difficulty of performing zoomingcompared to creating and managing windows And italso applies only to visual comparison tasks

So again the lessons are interesting and importantto specific usage cases but there remains no studytelling us lsquowhether zooming is goodrsquo The real answer isthat it depends

3 Discussion

A trend in ZUI applications that appears to be fairlyclear over the years is that over time these applicationshave tended to get simpler and easier to use while atthe same time offering less creative control to endusers This is an important trend because researchersoften have a tendency to create tools that are rich andpowerful After all it is easier to conceptualise aninnovation as offering lsquomorersquo rather than offering lsquolessrsquo

But the marketplace is clear ndash success often requiressimplicity first and power second Of course marryingthe two where the power is under the hood is often thebest solution ndash but also very difficult to design Thesuccess of so many modern applications (eg searchtwitter photo sharing iPhone App Store) are examplesof this in that they can be used with almost no trainingin tiny little bursts

I used to think that only public kiosks had to havethese simplistic characteristics ndash but now it seems thateven applications that people use for hours a week andbecome true experts in follow the same rules If truethis implies that we as researchers have to stronglyconsider not only what new things technologies canhelp people do but also consider simplicity and speedas fundamental goals

31 Why ZUIs are challenging

As summarised in Table 2 the potential benefits ofZUIs are sometimes mirages ZUIs are generally

Table 2 Promise and reality of ZUIs

Promise Reality

Engaging If designed well The animated nature of ZUIs is sometimes referred to as lsquoeye candyrsquo implying that peoplecannot help but pay attention to the movement on the screen However this very movement is a double-edgedsword because if the user does not pay attention and misses the animation then the connection between thedifferent views is lost and the user is likely worse off than without a ZUI

Visually rich Can also make space more complex to navigate and understand ZUIs can have visual objects at any size and atany place ndash thus offering the possibility of placing significant content (scaled down) even in a tiny crackbetween two other objects This kind of structure is very difficult to perceive (since users cannot see the tinycontent in the first place) and conceptualise (since at any given view the vastly different sized objects can notbe seen simultaneously)

Simplicity Does not scale to very large data sets One of the attractions of ZUIs is that they appear conceptually simple(lsquojust zoom in for a closer lookrsquo) But in practice navigating large information spaces in ZUIs is often difficultto execute and requires significant short-term memory to navigate

Behaviour amp Information Technology 861

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

engaging (although they make some people feelphysically sick) and visually rich But the promise ofsimplicity falls short

While human visual perception does make it easy tosee where one is navigating the reality is that it places aheavy load on short-termmemory to remember where inspace you just were and where things are And therequirement of human memory to know how space isorganised means that ZUIs do not scale up very wellZUIs are oftenmotivated by the physical world and howpeople like laying papers out on their desk But no onewants all of their papers on their desk It is much morecommon to have only a relatively small number ofpapers that are actually being worked with

The visual overviews that ZUIs offer for free byzooming out may seem like a solution to the load onhuman memory but in practice it does not becausevisual overviews of any complexity require significantvisual search in order to find anything If there are just asmall number of objects then the visual search task is nothard ndash but of course for a small number of objects youdo not need a ZUI to solve your organisationalproblems

In addition the visual richness of ZUIs is a double-edged sword It requires skill to design a complex spacewith documents of arbitrary size aspect ratio andcolour that people can comprehend and scan Alsopeople are not as good at scanning 2D designs as 1Dlayouts ndash unless the layout is highly structured Buthighly structured 2D layouts do not work well forvisual objects of arbitrary aspect ratios Designers areobligated to leave a lot of unused space scale down thelarge objects so they are unreadable or crop the largeobjects ndash thus losing much of their distinguishability

Finally it is crucial to consider the efficiency ofZUIs and their alternatives It is the nature ofzoomable structures to organise information hierarchi-cally Zooming in to an object shows you moreinformation about that object Of course hierarchicalstructures are also commonly exposed with a simpletree widget as is commonly found in most file systembrowsers The file system browsers exemplified byWindows Explorer are not only very efficient but alsomore easily let the user compare different areas in thehierarchy by simply opening and collapsing the desirednodes So in many cases the alternative to a ZUI is asimple tree widget that is simple compact easy tolearn and fast to use Thus in addition to the broaderchallenges of ZUIs we should always consider thealternatives that are also powerful

32 Design guidelines

So how should ZUIs be designed to best overcomethese obstacles Let us start with the big issues

321 Support the right tasks

ZUIs like most visualisations have strong potentialfor supporting users in understanding the big pictureidentifying trends patterns and outliers But theytypically are not good at helping users simply get thebest answer So be sure that some sort of overview-based design is called for in the first place

322 Be cautious about using zooming as the primaryinterface

Zooming navigation can be difficult and is uncom-mon Unless the data set is fundamentally spatial (egmaps) users probably do not want zooming as theirprimary interaction with the data Instead think abouthow people search and create data sets ndash which maythen be accessed partially by zooming ie createhybrid systems that combine mechanisms such asfaceted search and zooming [eg PhotoMesa (Beder-son 2001) the Hard Rock Memorabilia (2010) orMicrosoft Pivot (2010)]

323 Only use ZUIs when a small visual representa-tion of the data is available

Zooming works best when the objects can be recognisedwhen they are zoomed out So certain domains arebetter suited to ZUIs Photos are good purely textualdocuments are bad and audio recordings are terrible(unless there are associated images) Recent researchshows how a small visual snippet may be automaticallycreated from web pages (Teevan et al 2009)

324 Donrsquot limit yourself to fixed data sets

People rarely want to interact long with a fixed set ofdata But it is often technically difficult to dynamicallycreate and modify large zoomable spaces Be sure thatthe user interface does not get blocked or slowed downwhen new data is added to a zoomable space

And then there are several more specificsuggestions

325 Maintain aspect ratio

It is tempting to use semantic zooming to signifi-cantly change the visual representation of objects atdifferent sizes For example a visual chart may becondensed to a single value or a text document maybe replaced by its title when small However thesmall object must have the same aspect ratio ndash or atleast fit in the same box as the full object Otherwisenearby objects can end up overlapping when the userzooms out

862 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

326 Use meaningful layout

The problem of finding a specific object among a largenumberof themcanbe alleviated if theobjects are laidoutin a semantically meaningful way If the user can predictwhere an object will be based on attributes that they arelikely to be familiar with then they can trim the searchspace Traditional paper phones books with theiralphabetic ordering are a good example of this approach

327 Use consistent layout

Ideally once an object is placed in space it should notmove That way the user can build up an internal mapover time of where objects are Unfortunately thisideal is often in direct conflict with the previous one ofmeaningful layout If objects are to be laid outmeaningfully then the layout must change over timeas the data changes and as the organisationalmechanism changes PhotoMesa (Bederson 2001) isan example of a ZUI that went for a meaningful layoutbut had no consistency at all

328 Use scannable layout

Objects must be positioned in such a manner as to enablethe user to visually scan the space to look for an objectwithout missing or repeating objects Simple 1D lists ofobjects are easy to scanbecause the person just runs downthe list and the only state that must be kept is the currentitem and the direction of scanning In 2D however forany layout except for a grid there is the potential for amuch heavier load on short-term memory

329 Use breadth over depth

It is tempting to place objects in space at manydifferent sizes so more information can be discoveredby zooming in further ndash and in fact many of the earlierZUI applications did this However when objects areplaced at many different levels the user would notknow that the small objects exist (unless the space iscarefully designed to indicate them) and they can getlost Further even if the user knows they are there it istypically burdensome to zoom in and out repeatedlySpaceTree (Grosjean et al 2002) is an example of anapplication that was designed explicitly with breadth inmind Even with hundreds of thousands of objects allof them are at the same size

3210 Use small data sets

The problems that come with a large number of objectsin a zoomable space can be mitigated by not buildingapplications around large data sets ZUIs appear to

work better for users when they have no more thanabout 100 screens worth of information HoweverZUIs with 1000 or more screens worth of informationbecome problematic because of the issues raised in theprevious section (why ZUIs are challenging) This is afundamental problem since ZUIs are often designed towork for user generated data ndash which may not easily belimited So application designers may choose topartition the space into manageable subsets

3211 Use simple and consistent navigationinteraction

People must be able to discover how to navigate thespace with no training and it must be difficult orimpossible to get lost in Desert Fog Unfortunatelythere is significant inconsistency among applicationsThey vary between single and double clicks to zoom inhow much the screen is zoomed per action whetherzooms scale by a fixed percentage or whether the targetobject fills the screen And mechanisms are all over themap to zoom out ndash from right click left click on thebackground fixed on-screen buttons and pop-upbuttons And as with zoom in the actual amountthat is zoomed out is inconsistent across applicationsas well

4 Conclusion

Understanding ZUIs is difficult They have clearlybeen a success by traditional research measures(significant visibility publications and citations) Andelements of ZUIs have been broadly deployed incommonly used applications But the original broadvision of zooming as a basic information organisa-tional principle [such as Raskinrsquos vision of hospitalmanagement (Raskin 2000)] has not happened Ofcourse lack of commercial success does not mean anidea is not a good one (maybe the business reasons didnot line up or maybe the right tools got built but withthe wrong marketing strategy the timing may havebeen wrong or any number of other reasons mighthave stalled a successful business) But still the lack ofbroad success coupled with the fundamental issuesraised in this paper appears significant

I definitely would not rule out further success ofZUIs and I think that the successes that haveoccurred such as Google Maps Microsoft Word andApple iPhone are substantial However their applic-ability is much narrower than original envisioned Andthe design and technical challenges of building effectiveZUIs make it that much less likely that they will bebroadly deployed At the same time there are clearcases where zooming is valuable ndash such as withdocuments that are larger than screens (ie photos

Behaviour amp Information Technology 863

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

[as explained with space-scale diagrams (Furnas andBederson 1995) and elaborated on by Van Wijk andNuij (2004)] This observation led to SDAZ where theapplication would automatically zoom out when theuser panned quickly and then zoom back in whenthe user slowed down

Several studies showed the effectiveness of thisapproach (Cockburn et al 2005 Sun and Guimbretiere2005) and yet not a single research or commercial ZUIapplication (outside those that are explicitly exploringSDAZ) uses SDAZ as even an optional navigationmechanism Why is this There is not a clear answerbut it is important to observe that while SDAZsimplifies one thing (degrees of freedom of control)it complicates another (device control and perception)

SDAZ requires the user to engage in a real-timeperceive-think-act loop (Card et al 1986) This real-time interactive manipulation was typical of the lsquoflyingrsquoin many of the earliest ZUI systems ndash and that causedsignificant usability problems People who play videogames like the interaction challenge of real-timeinteraction People doing information processing taskstypically do not They are more likely to want to spendtheir cognitive resources on the underlying task not onanalysing the interface ndash even if it results in them beingslightly slower

222 Mouse wheels

While SDAZ is a somewhat complicated way to controlzooming some clear winners for controlling zoominghave emerged in the marketplace and they are themouse wheel and multi-touch lsquopinchingrsquo The mousewheel has become fairly widespread and is used fornumerous linear navigation actions ndash from scrolling toselection to zooming It offers simple control over theextra degree of freedom and while it is somewhat slow(requiring movement for each further bit of zooming)its simplicity makes it trivial to learn The fact thatzooming navigation often requires only a small scalechange makes for a perfect mapping to themouse wheel

223 Touch screens and multi-touch interaction

The use of multi-touch lsquopinchrsquo gestures on touchscreens is another control mechanism with growingpopularity Ishii and Ullmer originally envisioned thisapproach by indirectly using physical objects to scaleand rotate images on a screen in 1997 (Ishii and Ullmer1997) Then in 2002 Rekimoto developed a directsolution by using multiple fingers on a capacitive touchscreen to scale and rotate images (Rekimoto 2002)Finally this approach was commercialised and is nowbroadly used in Applersquos iPhone and other mobilephones tables as well as Microsoft Surface This kind

of direct multi-touch interaction is easy to learn andoperate More generally the rapidly growing use ofhigh-end touch screen-based smartphones is increasingthe need for zooming (with people frequently readingdocuments designed for larger screens on smallscreens) and pinching seems to be the de factostandard for controlling this

23 Applications

To see something of the range of uses of ZUIs and howtheir characteristics have changed over time Table 1shows a selection of zoomable applications Only lsquotruersquoZUIs are shown (those supporting more than onedocument or object) It captures a range of what peoplehave been using zooming for and makes apparent therange of approaches that people have taken with regardto layout flexibility and navigation It is also clear thatthe essential problem of getting lost in Desert Fog hasnot been consistently avoided Furthermore it is clearthat there is no consistency in the mechanisms that areused to navigate through space

Of broader import there is no obvious focus onusers tasks or problem domains This is characteristicof both the potential and the challenge of ZUIsZooming is a completely general strategy and so oneshould not be surprised to see such a lack of focus anymore than one might see scroll bars and search used inan incredibly diverse set of scenarios That said theredoes seem to be a concentration of ZUIs applied todomains with highly visual documents ndash which ofcourse makes sense given the visual nature of ZUIs

24 Studies

Some of these and other ZUI applications have beenstudied for usability or effectiveness However it is thenature of those application-centric studies that it isdifficult to tease out the relative benefit or cost of thezooming characteristics of those applications

In fact there are relatively few studies that look atthe value of animated zooming more generally But afew are particularly useful so let us look at them

241 Overviews

The first study by Hornbaeligk et al (2002) looked atmulti-scale information structure as well as the use ofsmall overview windows that show a zoomed out viewof the information Surprisingly it found that whileparticipants like the overview window its use resultedin lower performance The study used geographicalmaps as the data set Navigation worked by draggingthe mouse to pan holding down the left button tozoom in and holding down the right button to zoom

Behaviour amp Information Technology 857

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

SomeZUIsystem

sandapplications

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

1993

Pad(PerlinandFox

1993)

First

ZUIplatform

included

several

conceptualisationsof

usagescenarios

Unconstrained

Yes

Leftclickto

zoom

in

Rightclickto

zoom

out

1994

Padthornthorn

(Bederson

andHollan1994)

SecondZUIplatform

included

APIfor

othersto

write

ZUI

apps

Unconstrained

Yes

Leftclickandhold

tozoom

inRightclick

andhold

tozoom

out

1997

KidPad(D

ruin

etal

1997)

Useslsquolocaltoolsrsquo

(Bedersonet

al1996)

toenable

childrento

create

stories

Unconstrained

Drawingsim

ages

textpositioned

manually

Yes

Hyperlinksforpath

Zoom

inoutmodes

Click

toflyinout

1998

PadPrints

(Hightower

etal1998)

Createsavisual

hierarchicalmapof

web

pages

visited

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2001

CounterPoint(G

ood

andBederson2002)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Yes

Click

onslidefornext

Rightclick-and-hold

fliesinout

2001

PhotoMesa(Bederson

2001)

Apersonalphoto

browserdesktopand

mobile

Containsphotosin

adynamicallygenerated

groupofgridsusinga

treemapalgorithm

No

Leftclickzoomsin

Rightclickzoomsout

2002

Seadragon(2010)

Clientserver

high

perfZUIforim

ages

over

anetwork

(web

andmobile)

Containsphotoslaid

out

byasm

allnumber

of

fixed

layout

algorithms

No

Click

tozoom

in

On-screenbutton

zoomsout

2002

Spacetree

(Grosjean

etal2002)

Hierarchyexploration

tool

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2006

Dynapad(Bauer

2006)

Supportsorganisationof

personalcollections

Dynamicveryflexible

No

Multi-touch

2007

Apple

iPhone(2010)

Applicationlauncher

for

Applersquosmobilephone

Containsiconslaid

out

inafixed

grid

No

Tapto

zoom

inPhysical

buttonto

zoom

out

2008

MicrosoftpptPlex(2010)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Nobutwithmanual

override

Click

onslidefornext

Click

tozoom

inEsc

tozoom

outScroll

wheelfliesinout

2008

Zoomorama(2010)

Clientserver

toenable

high-perform

ance

zoomingofim

ages

over

anetwork

Dynamicallygenerated

gridcontaining

images

No

Pop-upon-screenbut-

tonsforzoom

inout

(continued)

858 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

(continued)

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

2008

Prezi

(2010)

Website

forcreatingand

makingfree-form

presentations

Unconstrained

Drawingsim

ages

moviestext

positioned

manually

Yes

TabfornextScroll

wheelon-screen

buttonsandmouse

keyboard

comboto

flyinout

2009

Zumobi(2010)

ZoomCanvas

Launcher

forasm

all

number

ofrelated

contentareason

high-endmobile

devices

Fixed

hand-tuned

layout

turned

tocontent

No

Tapto

zoom

in

on-screenbuttonto

zoom

out

2009

Microsoft

Canvasfor

OneN

ote

(2010)

Plug-inforMicrosoft

OneN

ote

showsvisual

overview

ofalla

userrsquosnotes

Dynamicallygenerated

gridofgrids

containingim

ages

of

notesfrom

Microsoft

OneN

ote

Nobutwithmanual

override

Click

onnote

tozoom

inEscrightclickor

bgndclickto

zoom

out

2009

Schem

atic(2010)

Example

ofcustom

website

withfixed

layoutoffixed

content

Fixed

hand-tuned

layout

tuned

tocontent

No

Click

onslideto

zoom

inClick

on

backgroundto

zoom

out

2009

Microsoft

Pivot(2010)

Facetedsearchtoolthat

useszoomingto

navi-

gate

throughsearch

results

Fixed

gridorgridwithin

barchart

layouts

No

Multiple

mechanisms

includingclickingand

on-screenbuttons

NoteNote

thatItheauthoram

acofounder

ofZumobiInc

Behaviour amp Information Technology 859

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

out The study varied whether or not users were shownoverviews and whether the map information wassingle or multi-level That is single level mapsdisplayed all textual information in a single font sizeso when the user zoomed out the text could not beread Multi-level maps displayed textual informationat different sizes Text referring to larger areas on themap was displayed at much larger fonts so that whenusers zoomed out they could read the text coveringlarger geographical areas

Before this study it was believed that overviewswere a good idea and that they improved subjectivesatisfaction (North and Shneiderman 2000) andefficiency (Beard and Walker 1990 Bauer 2006)However this study found a discrepancy betweenpreference and performance Study participants pre-ferred the overview condition but actually performedbetter in the no-overview condition especially whenusing the multi-level map

It turns out that switching between overview anddetail windows was correlated with higher taskcompletion time suggesting that integration of over-view and detail windows requires mental and motoreffort

This result is notable because a fundamentalconcern about ZUIs is that they require time to use(since the animations take time to occur) Worse thereis the potential that ZUIs tax human short-termmemory (STM) because users must integrate in theirheads the spatial layout of the information On theother hand there is the hope that the animatedtransitions are understood preconsciously by the hu-man visual system and thus may not in fact place asignificant load on STM

This study gives some evidence that the cost ofunderstanding an animated zoomable map of informa-tion is in fact less than the use of one with an overviewthat is one of the primary alternative interface designsOf course it does not say anything about the benefit orcost of a spatially organised information spacecompared to a non-spatially organised space It alsodoes not say anything about the benefit or cost ofanimation

242 Animation

A second study looked at simple animated navigationtransitions and found that these animations signifi-cantly increased performance ndash even including the timethe animations took These kinds of animations havelong been expected to provide benefit (Card et al1991) but evidence backing up this understanding hadbeen lacking Klein and Bederson (2005) looked at avery simple but highly controlled use of animationwhen scrolling while reading documents

Understanding animation precisely is important be-cause the potential benefit of animation is underminedby the fact that animation by its nature takes time Soeven if the animation provides some benefit there is acost as well and it is not obvious whether the benefitoutweighs the cost

Participants in this study read documents out loudand when they got to the end of a page they scrolled bypressing the down arrow key The animation speedvaried between 0 (ie unanimated) 100 300 and 500 msThe document type also varied between unformattedtext formatted text and abstract symbols (whereparticipants counted symbols instead of reading) Thereason for varying the document type was to understandhow visual landmarks interact with animation

The essential result was that animation did in factprovide a significant benefit even considering the timespent on the animation Reading errors were reducedby 54 with 500 ms animations Reading task timewas reduced by 3 and counting task time wasreduced by 24 for 300 ms animations The formatteddocuments had a higher overall performance (andlower improvement) implying that visual landmarksappear to be a good idea which is consistent with theHornbaeligk studyrsquos observation about multi-level maps

The lesson from this study is that when designedcorrectly the benefits of simple animated navigationaltransitions can outweigh their costs This supports thedesign decision to make zooming transitions use shortanimations This study is not definitive however since itis possible that the benefit that was seen while scrollingwould not occur while zooming However there is noobvious reason that the same benefits would not occurso this kind of animated transition continues to seemvaluable

243 Presentations

Two other studies looked at how people learned fromzoomable spaces (Boltman et al 2002 Good andBederson 2002) The outcomes here were more mixedwhere the zoomable presentation did not producebetter learning or memory of the presentation ndashalthough participants did remember the structure ofthe presentation better The first study looked at howchildren responded to a story that was presented withor without animated zooming The children with theanimated condition elaborated more during a discus-sion but did not otherwise recall the story better Thesecond looked at how college students responded to aslide presentation ndash again with or without animatedzooming The presentations were structured visually ina 2D hierarchical format with six sections arrangedin a circle where each section contained slides arrangedin a circle These students also did not recall the

860 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

presentation better but the ones who saw the animatedzooming condition did recall the structure of thepresentation better

Together these studies imply that zooming alone isprobably not enough to help people remember contentbetter However users of ZUIs may be more engagedand they may remember the spatial structure of thecontent better ndash even if they do not remember theactual content itself any better However the value ofremembering structure depends on the task If thestructure has value in and of itself then this is apositive value If on the other hand the visualstructure was added as a means to make the presenta-tion more visually interesting and had no intrinsicvalue then such improved retention would provide nobenefit

244 Zooming vs multiple windows

Two related papers (Plumlee and Ware 2002 Plumleeand Ware 2006) look more generally at modelling andthen studying how people perform visual comparisonsof multi-scale data using zoomable and multi-windowinterfaces Informed by an understanding of theextreme limitations of human visual working memorythey modelled how people can search using the twointerfaces The model showed that comparison tasksthat exceeded the capacity of visual working memory(which can hold roughly three visual objects) would bebetter served by adding windows Multiple windowsdecrease visual working memory load because differentviews can be directly compared In contrast ZUIsrequire the user to hold everything outside of thecurrent view in their memory ndash thus making theminappropriate for complex comparison tasks Thestudies that Plumlee and Ware performed demon-strated this essential characteristic and the trade-offbetween ZUIs and multiple windows closely matchedtheir models They also showed that users of ZUIs hadsignificantly higher error rates than users of multiple

windows The practical impact of these studies dependson the frequency and difficulty of performing zoomingcompared to creating and managing windows And italso applies only to visual comparison tasks

So again the lessons are interesting and importantto specific usage cases but there remains no studytelling us lsquowhether zooming is goodrsquo The real answer isthat it depends

3 Discussion

A trend in ZUI applications that appears to be fairlyclear over the years is that over time these applicationshave tended to get simpler and easier to use while atthe same time offering less creative control to endusers This is an important trend because researchersoften have a tendency to create tools that are rich andpowerful After all it is easier to conceptualise aninnovation as offering lsquomorersquo rather than offering lsquolessrsquo

But the marketplace is clear ndash success often requiressimplicity first and power second Of course marryingthe two where the power is under the hood is often thebest solution ndash but also very difficult to design Thesuccess of so many modern applications (eg searchtwitter photo sharing iPhone App Store) are examplesof this in that they can be used with almost no trainingin tiny little bursts

I used to think that only public kiosks had to havethese simplistic characteristics ndash but now it seems thateven applications that people use for hours a week andbecome true experts in follow the same rules If truethis implies that we as researchers have to stronglyconsider not only what new things technologies canhelp people do but also consider simplicity and speedas fundamental goals

31 Why ZUIs are challenging

As summarised in Table 2 the potential benefits ofZUIs are sometimes mirages ZUIs are generally

Table 2 Promise and reality of ZUIs

Promise Reality

Engaging If designed well The animated nature of ZUIs is sometimes referred to as lsquoeye candyrsquo implying that peoplecannot help but pay attention to the movement on the screen However this very movement is a double-edgedsword because if the user does not pay attention and misses the animation then the connection between thedifferent views is lost and the user is likely worse off than without a ZUI

Visually rich Can also make space more complex to navigate and understand ZUIs can have visual objects at any size and atany place ndash thus offering the possibility of placing significant content (scaled down) even in a tiny crackbetween two other objects This kind of structure is very difficult to perceive (since users cannot see the tinycontent in the first place) and conceptualise (since at any given view the vastly different sized objects can notbe seen simultaneously)

Simplicity Does not scale to very large data sets One of the attractions of ZUIs is that they appear conceptually simple(lsquojust zoom in for a closer lookrsquo) But in practice navigating large information spaces in ZUIs is often difficultto execute and requires significant short-term memory to navigate

Behaviour amp Information Technology 861

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

engaging (although they make some people feelphysically sick) and visually rich But the promise ofsimplicity falls short

While human visual perception does make it easy tosee where one is navigating the reality is that it places aheavy load on short-termmemory to remember where inspace you just were and where things are And therequirement of human memory to know how space isorganised means that ZUIs do not scale up very wellZUIs are oftenmotivated by the physical world and howpeople like laying papers out on their desk But no onewants all of their papers on their desk It is much morecommon to have only a relatively small number ofpapers that are actually being worked with

The visual overviews that ZUIs offer for free byzooming out may seem like a solution to the load onhuman memory but in practice it does not becausevisual overviews of any complexity require significantvisual search in order to find anything If there are just asmall number of objects then the visual search task is nothard ndash but of course for a small number of objects youdo not need a ZUI to solve your organisationalproblems

In addition the visual richness of ZUIs is a double-edged sword It requires skill to design a complex spacewith documents of arbitrary size aspect ratio andcolour that people can comprehend and scan Alsopeople are not as good at scanning 2D designs as 1Dlayouts ndash unless the layout is highly structured Buthighly structured 2D layouts do not work well forvisual objects of arbitrary aspect ratios Designers areobligated to leave a lot of unused space scale down thelarge objects so they are unreadable or crop the largeobjects ndash thus losing much of their distinguishability

Finally it is crucial to consider the efficiency ofZUIs and their alternatives It is the nature ofzoomable structures to organise information hierarchi-cally Zooming in to an object shows you moreinformation about that object Of course hierarchicalstructures are also commonly exposed with a simpletree widget as is commonly found in most file systembrowsers The file system browsers exemplified byWindows Explorer are not only very efficient but alsomore easily let the user compare different areas in thehierarchy by simply opening and collapsing the desirednodes So in many cases the alternative to a ZUI is asimple tree widget that is simple compact easy tolearn and fast to use Thus in addition to the broaderchallenges of ZUIs we should always consider thealternatives that are also powerful

32 Design guidelines

So how should ZUIs be designed to best overcomethese obstacles Let us start with the big issues

321 Support the right tasks

ZUIs like most visualisations have strong potentialfor supporting users in understanding the big pictureidentifying trends patterns and outliers But theytypically are not good at helping users simply get thebest answer So be sure that some sort of overview-based design is called for in the first place

322 Be cautious about using zooming as the primaryinterface

Zooming navigation can be difficult and is uncom-mon Unless the data set is fundamentally spatial (egmaps) users probably do not want zooming as theirprimary interaction with the data Instead think abouthow people search and create data sets ndash which maythen be accessed partially by zooming ie createhybrid systems that combine mechanisms such asfaceted search and zooming [eg PhotoMesa (Beder-son 2001) the Hard Rock Memorabilia (2010) orMicrosoft Pivot (2010)]

323 Only use ZUIs when a small visual representa-tion of the data is available

Zooming works best when the objects can be recognisedwhen they are zoomed out So certain domains arebetter suited to ZUIs Photos are good purely textualdocuments are bad and audio recordings are terrible(unless there are associated images) Recent researchshows how a small visual snippet may be automaticallycreated from web pages (Teevan et al 2009)

324 Donrsquot limit yourself to fixed data sets

People rarely want to interact long with a fixed set ofdata But it is often technically difficult to dynamicallycreate and modify large zoomable spaces Be sure thatthe user interface does not get blocked or slowed downwhen new data is added to a zoomable space

And then there are several more specificsuggestions

325 Maintain aspect ratio

It is tempting to use semantic zooming to signifi-cantly change the visual representation of objects atdifferent sizes For example a visual chart may becondensed to a single value or a text document maybe replaced by its title when small However thesmall object must have the same aspect ratio ndash or atleast fit in the same box as the full object Otherwisenearby objects can end up overlapping when the userzooms out

862 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

326 Use meaningful layout

The problem of finding a specific object among a largenumberof themcanbe alleviated if theobjects are laidoutin a semantically meaningful way If the user can predictwhere an object will be based on attributes that they arelikely to be familiar with then they can trim the searchspace Traditional paper phones books with theiralphabetic ordering are a good example of this approach

327 Use consistent layout

Ideally once an object is placed in space it should notmove That way the user can build up an internal mapover time of where objects are Unfortunately thisideal is often in direct conflict with the previous one ofmeaningful layout If objects are to be laid outmeaningfully then the layout must change over timeas the data changes and as the organisationalmechanism changes PhotoMesa (Bederson 2001) isan example of a ZUI that went for a meaningful layoutbut had no consistency at all

328 Use scannable layout

Objects must be positioned in such a manner as to enablethe user to visually scan the space to look for an objectwithout missing or repeating objects Simple 1D lists ofobjects are easy to scanbecause the person just runs downthe list and the only state that must be kept is the currentitem and the direction of scanning In 2D however forany layout except for a grid there is the potential for amuch heavier load on short-term memory

329 Use breadth over depth

It is tempting to place objects in space at manydifferent sizes so more information can be discoveredby zooming in further ndash and in fact many of the earlierZUI applications did this However when objects areplaced at many different levels the user would notknow that the small objects exist (unless the space iscarefully designed to indicate them) and they can getlost Further even if the user knows they are there it istypically burdensome to zoom in and out repeatedlySpaceTree (Grosjean et al 2002) is an example of anapplication that was designed explicitly with breadth inmind Even with hundreds of thousands of objects allof them are at the same size

3210 Use small data sets

The problems that come with a large number of objectsin a zoomable space can be mitigated by not buildingapplications around large data sets ZUIs appear to

work better for users when they have no more thanabout 100 screens worth of information HoweverZUIs with 1000 or more screens worth of informationbecome problematic because of the issues raised in theprevious section (why ZUIs are challenging) This is afundamental problem since ZUIs are often designed towork for user generated data ndash which may not easily belimited So application designers may choose topartition the space into manageable subsets

3211 Use simple and consistent navigationinteraction

People must be able to discover how to navigate thespace with no training and it must be difficult orimpossible to get lost in Desert Fog Unfortunatelythere is significant inconsistency among applicationsThey vary between single and double clicks to zoom inhow much the screen is zoomed per action whetherzooms scale by a fixed percentage or whether the targetobject fills the screen And mechanisms are all over themap to zoom out ndash from right click left click on thebackground fixed on-screen buttons and pop-upbuttons And as with zoom in the actual amountthat is zoomed out is inconsistent across applicationsas well

4 Conclusion

Understanding ZUIs is difficult They have clearlybeen a success by traditional research measures(significant visibility publications and citations) Andelements of ZUIs have been broadly deployed incommonly used applications But the original broadvision of zooming as a basic information organisa-tional principle [such as Raskinrsquos vision of hospitalmanagement (Raskin 2000)] has not happened Ofcourse lack of commercial success does not mean anidea is not a good one (maybe the business reasons didnot line up or maybe the right tools got built but withthe wrong marketing strategy the timing may havebeen wrong or any number of other reasons mighthave stalled a successful business) But still the lack ofbroad success coupled with the fundamental issuesraised in this paper appears significant

I definitely would not rule out further success ofZUIs and I think that the successes that haveoccurred such as Google Maps Microsoft Word andApple iPhone are substantial However their applic-ability is much narrower than original envisioned Andthe design and technical challenges of building effectiveZUIs make it that much less likely that they will bebroadly deployed At the same time there are clearcases where zooming is valuable ndash such as withdocuments that are larger than screens (ie photos

Behaviour amp Information Technology 863

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

SomeZUIsystem

sandapplications

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

1993

Pad(PerlinandFox

1993)

First

ZUIplatform

included

several

conceptualisationsof

usagescenarios

Unconstrained

Yes

Leftclickto

zoom

in

Rightclickto

zoom

out

1994

Padthornthorn

(Bederson

andHollan1994)

SecondZUIplatform

included

APIfor

othersto

write

ZUI

apps

Unconstrained

Yes

Leftclickandhold

tozoom

inRightclick

andhold

tozoom

out

1997

KidPad(D

ruin

etal

1997)

Useslsquolocaltoolsrsquo

(Bedersonet

al1996)

toenable

childrento

create

stories

Unconstrained

Drawingsim

ages

textpositioned

manually

Yes

Hyperlinksforpath

Zoom

inoutmodes

Click

toflyinout

1998

PadPrints

(Hightower

etal1998)

Createsavisual

hierarchicalmapof

web

pages

visited

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2001

CounterPoint(G

ood

andBederson2002)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Yes

Click

onslidefornext

Rightclick-and-hold

fliesinout

2001

PhotoMesa(Bederson

2001)

Apersonalphoto

browserdesktopand

mobile

Containsphotosin

adynamicallygenerated

groupofgridsusinga

treemapalgorithm

No

Leftclickzoomsin

Rightclickzoomsout

2002

Seadragon(2010)

Clientserver

high

perfZUIforim

ages

over

anetwork

(web

andmobile)

Containsphotoslaid

out

byasm

allnumber

of

fixed

layout

algorithms

No

Click

tozoom

in

On-screenbutton

zoomsout

2002

Spacetree

(Grosjean

etal2002)

Hierarchyexploration

tool

Dynamicallygenerated

tree-basednode-link

diagram

No

Rightclick-and-hold

fliesinout

2006

Dynapad(Bauer

2006)

Supportsorganisationof

personalcollections

Dynamicveryflexible

No

Multi-touch

2007

Apple

iPhone(2010)

Applicationlauncher

for

Applersquosmobilephone

Containsiconslaid

out

inafixed

grid

No

Tapto

zoom

inPhysical

buttonto

zoom

out

2008

MicrosoftpptPlex(2010)

Aplug-into

PowerPoint

thatenables

presentationauthors

tolayoutslides

inzoomable

space

ContainsPowerPoint

slidesSmallnumber

oflayoutalgorithms

withmanualoverride

Nobutwithmanual

override

Click

onslidefornext

Click

tozoom

inEsc

tozoom

outScroll

wheelfliesinout

2008

Zoomorama(2010)

Clientserver

toenable

high-perform

ance

zoomingofim

ages

over

anetwork

Dynamicallygenerated

gridcontaining

images

No

Pop-upon-screenbut-

tonsforzoom

inout

(continued)

858 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

(continued)

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

2008

Prezi

(2010)

Website

forcreatingand

makingfree-form

presentations

Unconstrained

Drawingsim

ages

moviestext

positioned

manually

Yes

TabfornextScroll

wheelon-screen

buttonsandmouse

keyboard

comboto

flyinout

2009

Zumobi(2010)

ZoomCanvas

Launcher

forasm

all

number

ofrelated

contentareason

high-endmobile

devices

Fixed

hand-tuned

layout

turned

tocontent

No

Tapto

zoom

in

on-screenbuttonto

zoom

out

2009

Microsoft

Canvasfor

OneN

ote

(2010)

Plug-inforMicrosoft

OneN

ote

showsvisual

overview

ofalla

userrsquosnotes

Dynamicallygenerated

gridofgrids

containingim

ages

of

notesfrom

Microsoft

OneN

ote

Nobutwithmanual

override

Click

onnote

tozoom

inEscrightclickor

bgndclickto

zoom

out

2009

Schem

atic(2010)

Example

ofcustom

website

withfixed

layoutoffixed

content

Fixed

hand-tuned

layout

tuned

tocontent

No

Click

onslideto

zoom

inClick

on

backgroundto

zoom

out

2009

Microsoft

Pivot(2010)

Facetedsearchtoolthat

useszoomingto

navi-

gate

throughsearch

results

Fixed

gridorgridwithin

barchart

layouts

No

Multiple

mechanisms

includingclickingand

on-screenbuttons

NoteNote

thatItheauthoram

acofounder

ofZumobiInc

Behaviour amp Information Technology 859

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

out The study varied whether or not users were shownoverviews and whether the map information wassingle or multi-level That is single level mapsdisplayed all textual information in a single font sizeso when the user zoomed out the text could not beread Multi-level maps displayed textual informationat different sizes Text referring to larger areas on themap was displayed at much larger fonts so that whenusers zoomed out they could read the text coveringlarger geographical areas

Before this study it was believed that overviewswere a good idea and that they improved subjectivesatisfaction (North and Shneiderman 2000) andefficiency (Beard and Walker 1990 Bauer 2006)However this study found a discrepancy betweenpreference and performance Study participants pre-ferred the overview condition but actually performedbetter in the no-overview condition especially whenusing the multi-level map

It turns out that switching between overview anddetail windows was correlated with higher taskcompletion time suggesting that integration of over-view and detail windows requires mental and motoreffort

This result is notable because a fundamentalconcern about ZUIs is that they require time to use(since the animations take time to occur) Worse thereis the potential that ZUIs tax human short-termmemory (STM) because users must integrate in theirheads the spatial layout of the information On theother hand there is the hope that the animatedtransitions are understood preconsciously by the hu-man visual system and thus may not in fact place asignificant load on STM

This study gives some evidence that the cost ofunderstanding an animated zoomable map of informa-tion is in fact less than the use of one with an overviewthat is one of the primary alternative interface designsOf course it does not say anything about the benefit orcost of a spatially organised information spacecompared to a non-spatially organised space It alsodoes not say anything about the benefit or cost ofanimation

242 Animation

A second study looked at simple animated navigationtransitions and found that these animations signifi-cantly increased performance ndash even including the timethe animations took These kinds of animations havelong been expected to provide benefit (Card et al1991) but evidence backing up this understanding hadbeen lacking Klein and Bederson (2005) looked at avery simple but highly controlled use of animationwhen scrolling while reading documents

Understanding animation precisely is important be-cause the potential benefit of animation is underminedby the fact that animation by its nature takes time Soeven if the animation provides some benefit there is acost as well and it is not obvious whether the benefitoutweighs the cost

Participants in this study read documents out loudand when they got to the end of a page they scrolled bypressing the down arrow key The animation speedvaried between 0 (ie unanimated) 100 300 and 500 msThe document type also varied between unformattedtext formatted text and abstract symbols (whereparticipants counted symbols instead of reading) Thereason for varying the document type was to understandhow visual landmarks interact with animation

The essential result was that animation did in factprovide a significant benefit even considering the timespent on the animation Reading errors were reducedby 54 with 500 ms animations Reading task timewas reduced by 3 and counting task time wasreduced by 24 for 300 ms animations The formatteddocuments had a higher overall performance (andlower improvement) implying that visual landmarksappear to be a good idea which is consistent with theHornbaeligk studyrsquos observation about multi-level maps

The lesson from this study is that when designedcorrectly the benefits of simple animated navigationaltransitions can outweigh their costs This supports thedesign decision to make zooming transitions use shortanimations This study is not definitive however since itis possible that the benefit that was seen while scrollingwould not occur while zooming However there is noobvious reason that the same benefits would not occurso this kind of animated transition continues to seemvaluable

243 Presentations

Two other studies looked at how people learned fromzoomable spaces (Boltman et al 2002 Good andBederson 2002) The outcomes here were more mixedwhere the zoomable presentation did not producebetter learning or memory of the presentation ndashalthough participants did remember the structure ofthe presentation better The first study looked at howchildren responded to a story that was presented withor without animated zooming The children with theanimated condition elaborated more during a discus-sion but did not otherwise recall the story better Thesecond looked at how college students responded to aslide presentation ndash again with or without animatedzooming The presentations were structured visually ina 2D hierarchical format with six sections arrangedin a circle where each section contained slides arrangedin a circle These students also did not recall the

860 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

presentation better but the ones who saw the animatedzooming condition did recall the structure of thepresentation better

Together these studies imply that zooming alone isprobably not enough to help people remember contentbetter However users of ZUIs may be more engagedand they may remember the spatial structure of thecontent better ndash even if they do not remember theactual content itself any better However the value ofremembering structure depends on the task If thestructure has value in and of itself then this is apositive value If on the other hand the visualstructure was added as a means to make the presenta-tion more visually interesting and had no intrinsicvalue then such improved retention would provide nobenefit

244 Zooming vs multiple windows

Two related papers (Plumlee and Ware 2002 Plumleeand Ware 2006) look more generally at modelling andthen studying how people perform visual comparisonsof multi-scale data using zoomable and multi-windowinterfaces Informed by an understanding of theextreme limitations of human visual working memorythey modelled how people can search using the twointerfaces The model showed that comparison tasksthat exceeded the capacity of visual working memory(which can hold roughly three visual objects) would bebetter served by adding windows Multiple windowsdecrease visual working memory load because differentviews can be directly compared In contrast ZUIsrequire the user to hold everything outside of thecurrent view in their memory ndash thus making theminappropriate for complex comparison tasks Thestudies that Plumlee and Ware performed demon-strated this essential characteristic and the trade-offbetween ZUIs and multiple windows closely matchedtheir models They also showed that users of ZUIs hadsignificantly higher error rates than users of multiple

windows The practical impact of these studies dependson the frequency and difficulty of performing zoomingcompared to creating and managing windows And italso applies only to visual comparison tasks

So again the lessons are interesting and importantto specific usage cases but there remains no studytelling us lsquowhether zooming is goodrsquo The real answer isthat it depends

3 Discussion

A trend in ZUI applications that appears to be fairlyclear over the years is that over time these applicationshave tended to get simpler and easier to use while atthe same time offering less creative control to endusers This is an important trend because researchersoften have a tendency to create tools that are rich andpowerful After all it is easier to conceptualise aninnovation as offering lsquomorersquo rather than offering lsquolessrsquo

But the marketplace is clear ndash success often requiressimplicity first and power second Of course marryingthe two where the power is under the hood is often thebest solution ndash but also very difficult to design Thesuccess of so many modern applications (eg searchtwitter photo sharing iPhone App Store) are examplesof this in that they can be used with almost no trainingin tiny little bursts

I used to think that only public kiosks had to havethese simplistic characteristics ndash but now it seems thateven applications that people use for hours a week andbecome true experts in follow the same rules If truethis implies that we as researchers have to stronglyconsider not only what new things technologies canhelp people do but also consider simplicity and speedas fundamental goals

31 Why ZUIs are challenging

As summarised in Table 2 the potential benefits ofZUIs are sometimes mirages ZUIs are generally

Table 2 Promise and reality of ZUIs

Promise Reality

Engaging If designed well The animated nature of ZUIs is sometimes referred to as lsquoeye candyrsquo implying that peoplecannot help but pay attention to the movement on the screen However this very movement is a double-edgedsword because if the user does not pay attention and misses the animation then the connection between thedifferent views is lost and the user is likely worse off than without a ZUI

Visually rich Can also make space more complex to navigate and understand ZUIs can have visual objects at any size and atany place ndash thus offering the possibility of placing significant content (scaled down) even in a tiny crackbetween two other objects This kind of structure is very difficult to perceive (since users cannot see the tinycontent in the first place) and conceptualise (since at any given view the vastly different sized objects can notbe seen simultaneously)

Simplicity Does not scale to very large data sets One of the attractions of ZUIs is that they appear conceptually simple(lsquojust zoom in for a closer lookrsquo) But in practice navigating large information spaces in ZUIs is often difficultto execute and requires significant short-term memory to navigate

Behaviour amp Information Technology 861

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

engaging (although they make some people feelphysically sick) and visually rich But the promise ofsimplicity falls short

While human visual perception does make it easy tosee where one is navigating the reality is that it places aheavy load on short-termmemory to remember where inspace you just were and where things are And therequirement of human memory to know how space isorganised means that ZUIs do not scale up very wellZUIs are oftenmotivated by the physical world and howpeople like laying papers out on their desk But no onewants all of their papers on their desk It is much morecommon to have only a relatively small number ofpapers that are actually being worked with

The visual overviews that ZUIs offer for free byzooming out may seem like a solution to the load onhuman memory but in practice it does not becausevisual overviews of any complexity require significantvisual search in order to find anything If there are just asmall number of objects then the visual search task is nothard ndash but of course for a small number of objects youdo not need a ZUI to solve your organisationalproblems

In addition the visual richness of ZUIs is a double-edged sword It requires skill to design a complex spacewith documents of arbitrary size aspect ratio andcolour that people can comprehend and scan Alsopeople are not as good at scanning 2D designs as 1Dlayouts ndash unless the layout is highly structured Buthighly structured 2D layouts do not work well forvisual objects of arbitrary aspect ratios Designers areobligated to leave a lot of unused space scale down thelarge objects so they are unreadable or crop the largeobjects ndash thus losing much of their distinguishability

Finally it is crucial to consider the efficiency ofZUIs and their alternatives It is the nature ofzoomable structures to organise information hierarchi-cally Zooming in to an object shows you moreinformation about that object Of course hierarchicalstructures are also commonly exposed with a simpletree widget as is commonly found in most file systembrowsers The file system browsers exemplified byWindows Explorer are not only very efficient but alsomore easily let the user compare different areas in thehierarchy by simply opening and collapsing the desirednodes So in many cases the alternative to a ZUI is asimple tree widget that is simple compact easy tolearn and fast to use Thus in addition to the broaderchallenges of ZUIs we should always consider thealternatives that are also powerful

32 Design guidelines

So how should ZUIs be designed to best overcomethese obstacles Let us start with the big issues

321 Support the right tasks

ZUIs like most visualisations have strong potentialfor supporting users in understanding the big pictureidentifying trends patterns and outliers But theytypically are not good at helping users simply get thebest answer So be sure that some sort of overview-based design is called for in the first place

322 Be cautious about using zooming as the primaryinterface

Zooming navigation can be difficult and is uncom-mon Unless the data set is fundamentally spatial (egmaps) users probably do not want zooming as theirprimary interaction with the data Instead think abouthow people search and create data sets ndash which maythen be accessed partially by zooming ie createhybrid systems that combine mechanisms such asfaceted search and zooming [eg PhotoMesa (Beder-son 2001) the Hard Rock Memorabilia (2010) orMicrosoft Pivot (2010)]

323 Only use ZUIs when a small visual representa-tion of the data is available

Zooming works best when the objects can be recognisedwhen they are zoomed out So certain domains arebetter suited to ZUIs Photos are good purely textualdocuments are bad and audio recordings are terrible(unless there are associated images) Recent researchshows how a small visual snippet may be automaticallycreated from web pages (Teevan et al 2009)

324 Donrsquot limit yourself to fixed data sets

People rarely want to interact long with a fixed set ofdata But it is often technically difficult to dynamicallycreate and modify large zoomable spaces Be sure thatthe user interface does not get blocked or slowed downwhen new data is added to a zoomable space

And then there are several more specificsuggestions

325 Maintain aspect ratio

It is tempting to use semantic zooming to signifi-cantly change the visual representation of objects atdifferent sizes For example a visual chart may becondensed to a single value or a text document maybe replaced by its title when small However thesmall object must have the same aspect ratio ndash or atleast fit in the same box as the full object Otherwisenearby objects can end up overlapping when the userzooms out

862 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

326 Use meaningful layout

The problem of finding a specific object among a largenumberof themcanbe alleviated if theobjects are laidoutin a semantically meaningful way If the user can predictwhere an object will be based on attributes that they arelikely to be familiar with then they can trim the searchspace Traditional paper phones books with theiralphabetic ordering are a good example of this approach

327 Use consistent layout

Ideally once an object is placed in space it should notmove That way the user can build up an internal mapover time of where objects are Unfortunately thisideal is often in direct conflict with the previous one ofmeaningful layout If objects are to be laid outmeaningfully then the layout must change over timeas the data changes and as the organisationalmechanism changes PhotoMesa (Bederson 2001) isan example of a ZUI that went for a meaningful layoutbut had no consistency at all

328 Use scannable layout

Objects must be positioned in such a manner as to enablethe user to visually scan the space to look for an objectwithout missing or repeating objects Simple 1D lists ofobjects are easy to scanbecause the person just runs downthe list and the only state that must be kept is the currentitem and the direction of scanning In 2D however forany layout except for a grid there is the potential for amuch heavier load on short-term memory

329 Use breadth over depth

It is tempting to place objects in space at manydifferent sizes so more information can be discoveredby zooming in further ndash and in fact many of the earlierZUI applications did this However when objects areplaced at many different levels the user would notknow that the small objects exist (unless the space iscarefully designed to indicate them) and they can getlost Further even if the user knows they are there it istypically burdensome to zoom in and out repeatedlySpaceTree (Grosjean et al 2002) is an example of anapplication that was designed explicitly with breadth inmind Even with hundreds of thousands of objects allof them are at the same size

3210 Use small data sets

The problems that come with a large number of objectsin a zoomable space can be mitigated by not buildingapplications around large data sets ZUIs appear to

work better for users when they have no more thanabout 100 screens worth of information HoweverZUIs with 1000 or more screens worth of informationbecome problematic because of the issues raised in theprevious section (why ZUIs are challenging) This is afundamental problem since ZUIs are often designed towork for user generated data ndash which may not easily belimited So application designers may choose topartition the space into manageable subsets

3211 Use simple and consistent navigationinteraction

People must be able to discover how to navigate thespace with no training and it must be difficult orimpossible to get lost in Desert Fog Unfortunatelythere is significant inconsistency among applicationsThey vary between single and double clicks to zoom inhow much the screen is zoomed per action whetherzooms scale by a fixed percentage or whether the targetobject fills the screen And mechanisms are all over themap to zoom out ndash from right click left click on thebackground fixed on-screen buttons and pop-upbuttons And as with zoom in the actual amountthat is zoomed out is inconsistent across applicationsas well

4 Conclusion

Understanding ZUIs is difficult They have clearlybeen a success by traditional research measures(significant visibility publications and citations) Andelements of ZUIs have been broadly deployed incommonly used applications But the original broadvision of zooming as a basic information organisa-tional principle [such as Raskinrsquos vision of hospitalmanagement (Raskin 2000)] has not happened Ofcourse lack of commercial success does not mean anidea is not a good one (maybe the business reasons didnot line up or maybe the right tools got built but withthe wrong marketing strategy the timing may havebeen wrong or any number of other reasons mighthave stalled a successful business) But still the lack ofbroad success coupled with the fundamental issuesraised in this paper appears significant

I definitely would not rule out further success ofZUIs and I think that the successes that haveoccurred such as Google Maps Microsoft Word andApple iPhone are substantial However their applic-ability is much narrower than original envisioned Andthe design and technical challenges of building effectiveZUIs make it that much less likely that they will bebroadly deployed At the same time there are clearcases where zooming is valuable ndash such as withdocuments that are larger than screens (ie photos

Behaviour amp Information Technology 863

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Table

1

(continued)

Year

Name

Description

Layoutflexibility

Multi-level

Navigationmechanism

forzooming

2008

Prezi

(2010)

Website

forcreatingand

makingfree-form

presentations

Unconstrained

Drawingsim

ages

moviestext

positioned

manually

Yes

TabfornextScroll

wheelon-screen

buttonsandmouse

keyboard

comboto

flyinout

2009

Zumobi(2010)

ZoomCanvas

Launcher

forasm

all

number

ofrelated

contentareason

high-endmobile

devices

Fixed

hand-tuned

layout

turned

tocontent

No

Tapto

zoom

in

on-screenbuttonto

zoom

out

2009

Microsoft

Canvasfor

OneN

ote

(2010)

Plug-inforMicrosoft

OneN

ote

showsvisual

overview

ofalla

userrsquosnotes

Dynamicallygenerated

gridofgrids

containingim

ages

of

notesfrom

Microsoft

OneN

ote

Nobutwithmanual

override

Click

onnote

tozoom

inEscrightclickor

bgndclickto

zoom

out

2009

Schem

atic(2010)

Example

ofcustom

website

withfixed

layoutoffixed

content

Fixed

hand-tuned

layout

tuned

tocontent

No

Click

onslideto

zoom

inClick

on

backgroundto

zoom

out

2009

Microsoft

Pivot(2010)

Facetedsearchtoolthat

useszoomingto

navi-

gate

throughsearch

results

Fixed

gridorgridwithin

barchart

layouts

No

Multiple

mechanisms

includingclickingand

on-screenbuttons

NoteNote

thatItheauthoram

acofounder

ofZumobiInc

Behaviour amp Information Technology 859

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

out The study varied whether or not users were shownoverviews and whether the map information wassingle or multi-level That is single level mapsdisplayed all textual information in a single font sizeso when the user zoomed out the text could not beread Multi-level maps displayed textual informationat different sizes Text referring to larger areas on themap was displayed at much larger fonts so that whenusers zoomed out they could read the text coveringlarger geographical areas

Before this study it was believed that overviewswere a good idea and that they improved subjectivesatisfaction (North and Shneiderman 2000) andefficiency (Beard and Walker 1990 Bauer 2006)However this study found a discrepancy betweenpreference and performance Study participants pre-ferred the overview condition but actually performedbetter in the no-overview condition especially whenusing the multi-level map

It turns out that switching between overview anddetail windows was correlated with higher taskcompletion time suggesting that integration of over-view and detail windows requires mental and motoreffort

This result is notable because a fundamentalconcern about ZUIs is that they require time to use(since the animations take time to occur) Worse thereis the potential that ZUIs tax human short-termmemory (STM) because users must integrate in theirheads the spatial layout of the information On theother hand there is the hope that the animatedtransitions are understood preconsciously by the hu-man visual system and thus may not in fact place asignificant load on STM

This study gives some evidence that the cost ofunderstanding an animated zoomable map of informa-tion is in fact less than the use of one with an overviewthat is one of the primary alternative interface designsOf course it does not say anything about the benefit orcost of a spatially organised information spacecompared to a non-spatially organised space It alsodoes not say anything about the benefit or cost ofanimation

242 Animation

A second study looked at simple animated navigationtransitions and found that these animations signifi-cantly increased performance ndash even including the timethe animations took These kinds of animations havelong been expected to provide benefit (Card et al1991) but evidence backing up this understanding hadbeen lacking Klein and Bederson (2005) looked at avery simple but highly controlled use of animationwhen scrolling while reading documents

Understanding animation precisely is important be-cause the potential benefit of animation is underminedby the fact that animation by its nature takes time Soeven if the animation provides some benefit there is acost as well and it is not obvious whether the benefitoutweighs the cost

Participants in this study read documents out loudand when they got to the end of a page they scrolled bypressing the down arrow key The animation speedvaried between 0 (ie unanimated) 100 300 and 500 msThe document type also varied between unformattedtext formatted text and abstract symbols (whereparticipants counted symbols instead of reading) Thereason for varying the document type was to understandhow visual landmarks interact with animation

The essential result was that animation did in factprovide a significant benefit even considering the timespent on the animation Reading errors were reducedby 54 with 500 ms animations Reading task timewas reduced by 3 and counting task time wasreduced by 24 for 300 ms animations The formatteddocuments had a higher overall performance (andlower improvement) implying that visual landmarksappear to be a good idea which is consistent with theHornbaeligk studyrsquos observation about multi-level maps

The lesson from this study is that when designedcorrectly the benefits of simple animated navigationaltransitions can outweigh their costs This supports thedesign decision to make zooming transitions use shortanimations This study is not definitive however since itis possible that the benefit that was seen while scrollingwould not occur while zooming However there is noobvious reason that the same benefits would not occurso this kind of animated transition continues to seemvaluable

243 Presentations

Two other studies looked at how people learned fromzoomable spaces (Boltman et al 2002 Good andBederson 2002) The outcomes here were more mixedwhere the zoomable presentation did not producebetter learning or memory of the presentation ndashalthough participants did remember the structure ofthe presentation better The first study looked at howchildren responded to a story that was presented withor without animated zooming The children with theanimated condition elaborated more during a discus-sion but did not otherwise recall the story better Thesecond looked at how college students responded to aslide presentation ndash again with or without animatedzooming The presentations were structured visually ina 2D hierarchical format with six sections arrangedin a circle where each section contained slides arrangedin a circle These students also did not recall the

860 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

presentation better but the ones who saw the animatedzooming condition did recall the structure of thepresentation better

Together these studies imply that zooming alone isprobably not enough to help people remember contentbetter However users of ZUIs may be more engagedand they may remember the spatial structure of thecontent better ndash even if they do not remember theactual content itself any better However the value ofremembering structure depends on the task If thestructure has value in and of itself then this is apositive value If on the other hand the visualstructure was added as a means to make the presenta-tion more visually interesting and had no intrinsicvalue then such improved retention would provide nobenefit

244 Zooming vs multiple windows

Two related papers (Plumlee and Ware 2002 Plumleeand Ware 2006) look more generally at modelling andthen studying how people perform visual comparisonsof multi-scale data using zoomable and multi-windowinterfaces Informed by an understanding of theextreme limitations of human visual working memorythey modelled how people can search using the twointerfaces The model showed that comparison tasksthat exceeded the capacity of visual working memory(which can hold roughly three visual objects) would bebetter served by adding windows Multiple windowsdecrease visual working memory load because differentviews can be directly compared In contrast ZUIsrequire the user to hold everything outside of thecurrent view in their memory ndash thus making theminappropriate for complex comparison tasks Thestudies that Plumlee and Ware performed demon-strated this essential characteristic and the trade-offbetween ZUIs and multiple windows closely matchedtheir models They also showed that users of ZUIs hadsignificantly higher error rates than users of multiple

windows The practical impact of these studies dependson the frequency and difficulty of performing zoomingcompared to creating and managing windows And italso applies only to visual comparison tasks

So again the lessons are interesting and importantto specific usage cases but there remains no studytelling us lsquowhether zooming is goodrsquo The real answer isthat it depends

3 Discussion

A trend in ZUI applications that appears to be fairlyclear over the years is that over time these applicationshave tended to get simpler and easier to use while atthe same time offering less creative control to endusers This is an important trend because researchersoften have a tendency to create tools that are rich andpowerful After all it is easier to conceptualise aninnovation as offering lsquomorersquo rather than offering lsquolessrsquo

But the marketplace is clear ndash success often requiressimplicity first and power second Of course marryingthe two where the power is under the hood is often thebest solution ndash but also very difficult to design Thesuccess of so many modern applications (eg searchtwitter photo sharing iPhone App Store) are examplesof this in that they can be used with almost no trainingin tiny little bursts

I used to think that only public kiosks had to havethese simplistic characteristics ndash but now it seems thateven applications that people use for hours a week andbecome true experts in follow the same rules If truethis implies that we as researchers have to stronglyconsider not only what new things technologies canhelp people do but also consider simplicity and speedas fundamental goals

31 Why ZUIs are challenging

As summarised in Table 2 the potential benefits ofZUIs are sometimes mirages ZUIs are generally

Table 2 Promise and reality of ZUIs

Promise Reality

Engaging If designed well The animated nature of ZUIs is sometimes referred to as lsquoeye candyrsquo implying that peoplecannot help but pay attention to the movement on the screen However this very movement is a double-edgedsword because if the user does not pay attention and misses the animation then the connection between thedifferent views is lost and the user is likely worse off than without a ZUI

Visually rich Can also make space more complex to navigate and understand ZUIs can have visual objects at any size and atany place ndash thus offering the possibility of placing significant content (scaled down) even in a tiny crackbetween two other objects This kind of structure is very difficult to perceive (since users cannot see the tinycontent in the first place) and conceptualise (since at any given view the vastly different sized objects can notbe seen simultaneously)

Simplicity Does not scale to very large data sets One of the attractions of ZUIs is that they appear conceptually simple(lsquojust zoom in for a closer lookrsquo) But in practice navigating large information spaces in ZUIs is often difficultto execute and requires significant short-term memory to navigate

Behaviour amp Information Technology 861

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

engaging (although they make some people feelphysically sick) and visually rich But the promise ofsimplicity falls short

While human visual perception does make it easy tosee where one is navigating the reality is that it places aheavy load on short-termmemory to remember where inspace you just were and where things are And therequirement of human memory to know how space isorganised means that ZUIs do not scale up very wellZUIs are oftenmotivated by the physical world and howpeople like laying papers out on their desk But no onewants all of their papers on their desk It is much morecommon to have only a relatively small number ofpapers that are actually being worked with

The visual overviews that ZUIs offer for free byzooming out may seem like a solution to the load onhuman memory but in practice it does not becausevisual overviews of any complexity require significantvisual search in order to find anything If there are just asmall number of objects then the visual search task is nothard ndash but of course for a small number of objects youdo not need a ZUI to solve your organisationalproblems

In addition the visual richness of ZUIs is a double-edged sword It requires skill to design a complex spacewith documents of arbitrary size aspect ratio andcolour that people can comprehend and scan Alsopeople are not as good at scanning 2D designs as 1Dlayouts ndash unless the layout is highly structured Buthighly structured 2D layouts do not work well forvisual objects of arbitrary aspect ratios Designers areobligated to leave a lot of unused space scale down thelarge objects so they are unreadable or crop the largeobjects ndash thus losing much of their distinguishability

Finally it is crucial to consider the efficiency ofZUIs and their alternatives It is the nature ofzoomable structures to organise information hierarchi-cally Zooming in to an object shows you moreinformation about that object Of course hierarchicalstructures are also commonly exposed with a simpletree widget as is commonly found in most file systembrowsers The file system browsers exemplified byWindows Explorer are not only very efficient but alsomore easily let the user compare different areas in thehierarchy by simply opening and collapsing the desirednodes So in many cases the alternative to a ZUI is asimple tree widget that is simple compact easy tolearn and fast to use Thus in addition to the broaderchallenges of ZUIs we should always consider thealternatives that are also powerful

32 Design guidelines

So how should ZUIs be designed to best overcomethese obstacles Let us start with the big issues

321 Support the right tasks

ZUIs like most visualisations have strong potentialfor supporting users in understanding the big pictureidentifying trends patterns and outliers But theytypically are not good at helping users simply get thebest answer So be sure that some sort of overview-based design is called for in the first place

322 Be cautious about using zooming as the primaryinterface

Zooming navigation can be difficult and is uncom-mon Unless the data set is fundamentally spatial (egmaps) users probably do not want zooming as theirprimary interaction with the data Instead think abouthow people search and create data sets ndash which maythen be accessed partially by zooming ie createhybrid systems that combine mechanisms such asfaceted search and zooming [eg PhotoMesa (Beder-son 2001) the Hard Rock Memorabilia (2010) orMicrosoft Pivot (2010)]

323 Only use ZUIs when a small visual representa-tion of the data is available

Zooming works best when the objects can be recognisedwhen they are zoomed out So certain domains arebetter suited to ZUIs Photos are good purely textualdocuments are bad and audio recordings are terrible(unless there are associated images) Recent researchshows how a small visual snippet may be automaticallycreated from web pages (Teevan et al 2009)

324 Donrsquot limit yourself to fixed data sets

People rarely want to interact long with a fixed set ofdata But it is often technically difficult to dynamicallycreate and modify large zoomable spaces Be sure thatthe user interface does not get blocked or slowed downwhen new data is added to a zoomable space

And then there are several more specificsuggestions

325 Maintain aspect ratio

It is tempting to use semantic zooming to signifi-cantly change the visual representation of objects atdifferent sizes For example a visual chart may becondensed to a single value or a text document maybe replaced by its title when small However thesmall object must have the same aspect ratio ndash or atleast fit in the same box as the full object Otherwisenearby objects can end up overlapping when the userzooms out

862 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

326 Use meaningful layout

The problem of finding a specific object among a largenumberof themcanbe alleviated if theobjects are laidoutin a semantically meaningful way If the user can predictwhere an object will be based on attributes that they arelikely to be familiar with then they can trim the searchspace Traditional paper phones books with theiralphabetic ordering are a good example of this approach

327 Use consistent layout

Ideally once an object is placed in space it should notmove That way the user can build up an internal mapover time of where objects are Unfortunately thisideal is often in direct conflict with the previous one ofmeaningful layout If objects are to be laid outmeaningfully then the layout must change over timeas the data changes and as the organisationalmechanism changes PhotoMesa (Bederson 2001) isan example of a ZUI that went for a meaningful layoutbut had no consistency at all

328 Use scannable layout

Objects must be positioned in such a manner as to enablethe user to visually scan the space to look for an objectwithout missing or repeating objects Simple 1D lists ofobjects are easy to scanbecause the person just runs downthe list and the only state that must be kept is the currentitem and the direction of scanning In 2D however forany layout except for a grid there is the potential for amuch heavier load on short-term memory

329 Use breadth over depth

It is tempting to place objects in space at manydifferent sizes so more information can be discoveredby zooming in further ndash and in fact many of the earlierZUI applications did this However when objects areplaced at many different levels the user would notknow that the small objects exist (unless the space iscarefully designed to indicate them) and they can getlost Further even if the user knows they are there it istypically burdensome to zoom in and out repeatedlySpaceTree (Grosjean et al 2002) is an example of anapplication that was designed explicitly with breadth inmind Even with hundreds of thousands of objects allof them are at the same size

3210 Use small data sets

The problems that come with a large number of objectsin a zoomable space can be mitigated by not buildingapplications around large data sets ZUIs appear to

work better for users when they have no more thanabout 100 screens worth of information HoweverZUIs with 1000 or more screens worth of informationbecome problematic because of the issues raised in theprevious section (why ZUIs are challenging) This is afundamental problem since ZUIs are often designed towork for user generated data ndash which may not easily belimited So application designers may choose topartition the space into manageable subsets

3211 Use simple and consistent navigationinteraction

People must be able to discover how to navigate thespace with no training and it must be difficult orimpossible to get lost in Desert Fog Unfortunatelythere is significant inconsistency among applicationsThey vary between single and double clicks to zoom inhow much the screen is zoomed per action whetherzooms scale by a fixed percentage or whether the targetobject fills the screen And mechanisms are all over themap to zoom out ndash from right click left click on thebackground fixed on-screen buttons and pop-upbuttons And as with zoom in the actual amountthat is zoomed out is inconsistent across applicationsas well

4 Conclusion

Understanding ZUIs is difficult They have clearlybeen a success by traditional research measures(significant visibility publications and citations) Andelements of ZUIs have been broadly deployed incommonly used applications But the original broadvision of zooming as a basic information organisa-tional principle [such as Raskinrsquos vision of hospitalmanagement (Raskin 2000)] has not happened Ofcourse lack of commercial success does not mean anidea is not a good one (maybe the business reasons didnot line up or maybe the right tools got built but withthe wrong marketing strategy the timing may havebeen wrong or any number of other reasons mighthave stalled a successful business) But still the lack ofbroad success coupled with the fundamental issuesraised in this paper appears significant

I definitely would not rule out further success ofZUIs and I think that the successes that haveoccurred such as Google Maps Microsoft Word andApple iPhone are substantial However their applic-ability is much narrower than original envisioned Andthe design and technical challenges of building effectiveZUIs make it that much less likely that they will bebroadly deployed At the same time there are clearcases where zooming is valuable ndash such as withdocuments that are larger than screens (ie photos

Behaviour amp Information Technology 863

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

out The study varied whether or not users were shownoverviews and whether the map information wassingle or multi-level That is single level mapsdisplayed all textual information in a single font sizeso when the user zoomed out the text could not beread Multi-level maps displayed textual informationat different sizes Text referring to larger areas on themap was displayed at much larger fonts so that whenusers zoomed out they could read the text coveringlarger geographical areas

Before this study it was believed that overviewswere a good idea and that they improved subjectivesatisfaction (North and Shneiderman 2000) andefficiency (Beard and Walker 1990 Bauer 2006)However this study found a discrepancy betweenpreference and performance Study participants pre-ferred the overview condition but actually performedbetter in the no-overview condition especially whenusing the multi-level map

It turns out that switching between overview anddetail windows was correlated with higher taskcompletion time suggesting that integration of over-view and detail windows requires mental and motoreffort

This result is notable because a fundamentalconcern about ZUIs is that they require time to use(since the animations take time to occur) Worse thereis the potential that ZUIs tax human short-termmemory (STM) because users must integrate in theirheads the spatial layout of the information On theother hand there is the hope that the animatedtransitions are understood preconsciously by the hu-man visual system and thus may not in fact place asignificant load on STM

This study gives some evidence that the cost ofunderstanding an animated zoomable map of informa-tion is in fact less than the use of one with an overviewthat is one of the primary alternative interface designsOf course it does not say anything about the benefit orcost of a spatially organised information spacecompared to a non-spatially organised space It alsodoes not say anything about the benefit or cost ofanimation

242 Animation

A second study looked at simple animated navigationtransitions and found that these animations signifi-cantly increased performance ndash even including the timethe animations took These kinds of animations havelong been expected to provide benefit (Card et al1991) but evidence backing up this understanding hadbeen lacking Klein and Bederson (2005) looked at avery simple but highly controlled use of animationwhen scrolling while reading documents

Understanding animation precisely is important be-cause the potential benefit of animation is underminedby the fact that animation by its nature takes time Soeven if the animation provides some benefit there is acost as well and it is not obvious whether the benefitoutweighs the cost

Participants in this study read documents out loudand when they got to the end of a page they scrolled bypressing the down arrow key The animation speedvaried between 0 (ie unanimated) 100 300 and 500 msThe document type also varied between unformattedtext formatted text and abstract symbols (whereparticipants counted symbols instead of reading) Thereason for varying the document type was to understandhow visual landmarks interact with animation

The essential result was that animation did in factprovide a significant benefit even considering the timespent on the animation Reading errors were reducedby 54 with 500 ms animations Reading task timewas reduced by 3 and counting task time wasreduced by 24 for 300 ms animations The formatteddocuments had a higher overall performance (andlower improvement) implying that visual landmarksappear to be a good idea which is consistent with theHornbaeligk studyrsquos observation about multi-level maps

The lesson from this study is that when designedcorrectly the benefits of simple animated navigationaltransitions can outweigh their costs This supports thedesign decision to make zooming transitions use shortanimations This study is not definitive however since itis possible that the benefit that was seen while scrollingwould not occur while zooming However there is noobvious reason that the same benefits would not occurso this kind of animated transition continues to seemvaluable

243 Presentations

Two other studies looked at how people learned fromzoomable spaces (Boltman et al 2002 Good andBederson 2002) The outcomes here were more mixedwhere the zoomable presentation did not producebetter learning or memory of the presentation ndashalthough participants did remember the structure ofthe presentation better The first study looked at howchildren responded to a story that was presented withor without animated zooming The children with theanimated condition elaborated more during a discus-sion but did not otherwise recall the story better Thesecond looked at how college students responded to aslide presentation ndash again with or without animatedzooming The presentations were structured visually ina 2D hierarchical format with six sections arrangedin a circle where each section contained slides arrangedin a circle These students also did not recall the

860 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

presentation better but the ones who saw the animatedzooming condition did recall the structure of thepresentation better

Together these studies imply that zooming alone isprobably not enough to help people remember contentbetter However users of ZUIs may be more engagedand they may remember the spatial structure of thecontent better ndash even if they do not remember theactual content itself any better However the value ofremembering structure depends on the task If thestructure has value in and of itself then this is apositive value If on the other hand the visualstructure was added as a means to make the presenta-tion more visually interesting and had no intrinsicvalue then such improved retention would provide nobenefit

244 Zooming vs multiple windows

Two related papers (Plumlee and Ware 2002 Plumleeand Ware 2006) look more generally at modelling andthen studying how people perform visual comparisonsof multi-scale data using zoomable and multi-windowinterfaces Informed by an understanding of theextreme limitations of human visual working memorythey modelled how people can search using the twointerfaces The model showed that comparison tasksthat exceeded the capacity of visual working memory(which can hold roughly three visual objects) would bebetter served by adding windows Multiple windowsdecrease visual working memory load because differentviews can be directly compared In contrast ZUIsrequire the user to hold everything outside of thecurrent view in their memory ndash thus making theminappropriate for complex comparison tasks Thestudies that Plumlee and Ware performed demon-strated this essential characteristic and the trade-offbetween ZUIs and multiple windows closely matchedtheir models They also showed that users of ZUIs hadsignificantly higher error rates than users of multiple

windows The practical impact of these studies dependson the frequency and difficulty of performing zoomingcompared to creating and managing windows And italso applies only to visual comparison tasks

So again the lessons are interesting and importantto specific usage cases but there remains no studytelling us lsquowhether zooming is goodrsquo The real answer isthat it depends

3 Discussion

A trend in ZUI applications that appears to be fairlyclear over the years is that over time these applicationshave tended to get simpler and easier to use while atthe same time offering less creative control to endusers This is an important trend because researchersoften have a tendency to create tools that are rich andpowerful After all it is easier to conceptualise aninnovation as offering lsquomorersquo rather than offering lsquolessrsquo

But the marketplace is clear ndash success often requiressimplicity first and power second Of course marryingthe two where the power is under the hood is often thebest solution ndash but also very difficult to design Thesuccess of so many modern applications (eg searchtwitter photo sharing iPhone App Store) are examplesof this in that they can be used with almost no trainingin tiny little bursts

I used to think that only public kiosks had to havethese simplistic characteristics ndash but now it seems thateven applications that people use for hours a week andbecome true experts in follow the same rules If truethis implies that we as researchers have to stronglyconsider not only what new things technologies canhelp people do but also consider simplicity and speedas fundamental goals

31 Why ZUIs are challenging

As summarised in Table 2 the potential benefits ofZUIs are sometimes mirages ZUIs are generally

Table 2 Promise and reality of ZUIs

Promise Reality

Engaging If designed well The animated nature of ZUIs is sometimes referred to as lsquoeye candyrsquo implying that peoplecannot help but pay attention to the movement on the screen However this very movement is a double-edgedsword because if the user does not pay attention and misses the animation then the connection between thedifferent views is lost and the user is likely worse off than without a ZUI

Visually rich Can also make space more complex to navigate and understand ZUIs can have visual objects at any size and atany place ndash thus offering the possibility of placing significant content (scaled down) even in a tiny crackbetween two other objects This kind of structure is very difficult to perceive (since users cannot see the tinycontent in the first place) and conceptualise (since at any given view the vastly different sized objects can notbe seen simultaneously)

Simplicity Does not scale to very large data sets One of the attractions of ZUIs is that they appear conceptually simple(lsquojust zoom in for a closer lookrsquo) But in practice navigating large information spaces in ZUIs is often difficultto execute and requires significant short-term memory to navigate

Behaviour amp Information Technology 861

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

engaging (although they make some people feelphysically sick) and visually rich But the promise ofsimplicity falls short

While human visual perception does make it easy tosee where one is navigating the reality is that it places aheavy load on short-termmemory to remember where inspace you just were and where things are And therequirement of human memory to know how space isorganised means that ZUIs do not scale up very wellZUIs are oftenmotivated by the physical world and howpeople like laying papers out on their desk But no onewants all of their papers on their desk It is much morecommon to have only a relatively small number ofpapers that are actually being worked with

The visual overviews that ZUIs offer for free byzooming out may seem like a solution to the load onhuman memory but in practice it does not becausevisual overviews of any complexity require significantvisual search in order to find anything If there are just asmall number of objects then the visual search task is nothard ndash but of course for a small number of objects youdo not need a ZUI to solve your organisationalproblems

In addition the visual richness of ZUIs is a double-edged sword It requires skill to design a complex spacewith documents of arbitrary size aspect ratio andcolour that people can comprehend and scan Alsopeople are not as good at scanning 2D designs as 1Dlayouts ndash unless the layout is highly structured Buthighly structured 2D layouts do not work well forvisual objects of arbitrary aspect ratios Designers areobligated to leave a lot of unused space scale down thelarge objects so they are unreadable or crop the largeobjects ndash thus losing much of their distinguishability

Finally it is crucial to consider the efficiency ofZUIs and their alternatives It is the nature ofzoomable structures to organise information hierarchi-cally Zooming in to an object shows you moreinformation about that object Of course hierarchicalstructures are also commonly exposed with a simpletree widget as is commonly found in most file systembrowsers The file system browsers exemplified byWindows Explorer are not only very efficient but alsomore easily let the user compare different areas in thehierarchy by simply opening and collapsing the desirednodes So in many cases the alternative to a ZUI is asimple tree widget that is simple compact easy tolearn and fast to use Thus in addition to the broaderchallenges of ZUIs we should always consider thealternatives that are also powerful

32 Design guidelines

So how should ZUIs be designed to best overcomethese obstacles Let us start with the big issues

321 Support the right tasks

ZUIs like most visualisations have strong potentialfor supporting users in understanding the big pictureidentifying trends patterns and outliers But theytypically are not good at helping users simply get thebest answer So be sure that some sort of overview-based design is called for in the first place

322 Be cautious about using zooming as the primaryinterface

Zooming navigation can be difficult and is uncom-mon Unless the data set is fundamentally spatial (egmaps) users probably do not want zooming as theirprimary interaction with the data Instead think abouthow people search and create data sets ndash which maythen be accessed partially by zooming ie createhybrid systems that combine mechanisms such asfaceted search and zooming [eg PhotoMesa (Beder-son 2001) the Hard Rock Memorabilia (2010) orMicrosoft Pivot (2010)]

323 Only use ZUIs when a small visual representa-tion of the data is available

Zooming works best when the objects can be recognisedwhen they are zoomed out So certain domains arebetter suited to ZUIs Photos are good purely textualdocuments are bad and audio recordings are terrible(unless there are associated images) Recent researchshows how a small visual snippet may be automaticallycreated from web pages (Teevan et al 2009)

324 Donrsquot limit yourself to fixed data sets

People rarely want to interact long with a fixed set ofdata But it is often technically difficult to dynamicallycreate and modify large zoomable spaces Be sure thatthe user interface does not get blocked or slowed downwhen new data is added to a zoomable space

And then there are several more specificsuggestions

325 Maintain aspect ratio

It is tempting to use semantic zooming to signifi-cantly change the visual representation of objects atdifferent sizes For example a visual chart may becondensed to a single value or a text document maybe replaced by its title when small However thesmall object must have the same aspect ratio ndash or atleast fit in the same box as the full object Otherwisenearby objects can end up overlapping when the userzooms out

862 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

326 Use meaningful layout

The problem of finding a specific object among a largenumberof themcanbe alleviated if theobjects are laidoutin a semantically meaningful way If the user can predictwhere an object will be based on attributes that they arelikely to be familiar with then they can trim the searchspace Traditional paper phones books with theiralphabetic ordering are a good example of this approach

327 Use consistent layout

Ideally once an object is placed in space it should notmove That way the user can build up an internal mapover time of where objects are Unfortunately thisideal is often in direct conflict with the previous one ofmeaningful layout If objects are to be laid outmeaningfully then the layout must change over timeas the data changes and as the organisationalmechanism changes PhotoMesa (Bederson 2001) isan example of a ZUI that went for a meaningful layoutbut had no consistency at all

328 Use scannable layout

Objects must be positioned in such a manner as to enablethe user to visually scan the space to look for an objectwithout missing or repeating objects Simple 1D lists ofobjects are easy to scanbecause the person just runs downthe list and the only state that must be kept is the currentitem and the direction of scanning In 2D however forany layout except for a grid there is the potential for amuch heavier load on short-term memory

329 Use breadth over depth

It is tempting to place objects in space at manydifferent sizes so more information can be discoveredby zooming in further ndash and in fact many of the earlierZUI applications did this However when objects areplaced at many different levels the user would notknow that the small objects exist (unless the space iscarefully designed to indicate them) and they can getlost Further even if the user knows they are there it istypically burdensome to zoom in and out repeatedlySpaceTree (Grosjean et al 2002) is an example of anapplication that was designed explicitly with breadth inmind Even with hundreds of thousands of objects allof them are at the same size

3210 Use small data sets

The problems that come with a large number of objectsin a zoomable space can be mitigated by not buildingapplications around large data sets ZUIs appear to

work better for users when they have no more thanabout 100 screens worth of information HoweverZUIs with 1000 or more screens worth of informationbecome problematic because of the issues raised in theprevious section (why ZUIs are challenging) This is afundamental problem since ZUIs are often designed towork for user generated data ndash which may not easily belimited So application designers may choose topartition the space into manageable subsets

3211 Use simple and consistent navigationinteraction

People must be able to discover how to navigate thespace with no training and it must be difficult orimpossible to get lost in Desert Fog Unfortunatelythere is significant inconsistency among applicationsThey vary between single and double clicks to zoom inhow much the screen is zoomed per action whetherzooms scale by a fixed percentage or whether the targetobject fills the screen And mechanisms are all over themap to zoom out ndash from right click left click on thebackground fixed on-screen buttons and pop-upbuttons And as with zoom in the actual amountthat is zoomed out is inconsistent across applicationsas well

4 Conclusion

Understanding ZUIs is difficult They have clearlybeen a success by traditional research measures(significant visibility publications and citations) Andelements of ZUIs have been broadly deployed incommonly used applications But the original broadvision of zooming as a basic information organisa-tional principle [such as Raskinrsquos vision of hospitalmanagement (Raskin 2000)] has not happened Ofcourse lack of commercial success does not mean anidea is not a good one (maybe the business reasons didnot line up or maybe the right tools got built but withthe wrong marketing strategy the timing may havebeen wrong or any number of other reasons mighthave stalled a successful business) But still the lack ofbroad success coupled with the fundamental issuesraised in this paper appears significant

I definitely would not rule out further success ofZUIs and I think that the successes that haveoccurred such as Google Maps Microsoft Word andApple iPhone are substantial However their applic-ability is much narrower than original envisioned Andthe design and technical challenges of building effectiveZUIs make it that much less likely that they will bebroadly deployed At the same time there are clearcases where zooming is valuable ndash such as withdocuments that are larger than screens (ie photos

Behaviour amp Information Technology 863

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

presentation better but the ones who saw the animatedzooming condition did recall the structure of thepresentation better

Together these studies imply that zooming alone isprobably not enough to help people remember contentbetter However users of ZUIs may be more engagedand they may remember the spatial structure of thecontent better ndash even if they do not remember theactual content itself any better However the value ofremembering structure depends on the task If thestructure has value in and of itself then this is apositive value If on the other hand the visualstructure was added as a means to make the presenta-tion more visually interesting and had no intrinsicvalue then such improved retention would provide nobenefit

244 Zooming vs multiple windows

Two related papers (Plumlee and Ware 2002 Plumleeand Ware 2006) look more generally at modelling andthen studying how people perform visual comparisonsof multi-scale data using zoomable and multi-windowinterfaces Informed by an understanding of theextreme limitations of human visual working memorythey modelled how people can search using the twointerfaces The model showed that comparison tasksthat exceeded the capacity of visual working memory(which can hold roughly three visual objects) would bebetter served by adding windows Multiple windowsdecrease visual working memory load because differentviews can be directly compared In contrast ZUIsrequire the user to hold everything outside of thecurrent view in their memory ndash thus making theminappropriate for complex comparison tasks Thestudies that Plumlee and Ware performed demon-strated this essential characteristic and the trade-offbetween ZUIs and multiple windows closely matchedtheir models They also showed that users of ZUIs hadsignificantly higher error rates than users of multiple

windows The practical impact of these studies dependson the frequency and difficulty of performing zoomingcompared to creating and managing windows And italso applies only to visual comparison tasks

So again the lessons are interesting and importantto specific usage cases but there remains no studytelling us lsquowhether zooming is goodrsquo The real answer isthat it depends

3 Discussion

A trend in ZUI applications that appears to be fairlyclear over the years is that over time these applicationshave tended to get simpler and easier to use while atthe same time offering less creative control to endusers This is an important trend because researchersoften have a tendency to create tools that are rich andpowerful After all it is easier to conceptualise aninnovation as offering lsquomorersquo rather than offering lsquolessrsquo

But the marketplace is clear ndash success often requiressimplicity first and power second Of course marryingthe two where the power is under the hood is often thebest solution ndash but also very difficult to design Thesuccess of so many modern applications (eg searchtwitter photo sharing iPhone App Store) are examplesof this in that they can be used with almost no trainingin tiny little bursts

I used to think that only public kiosks had to havethese simplistic characteristics ndash but now it seems thateven applications that people use for hours a week andbecome true experts in follow the same rules If truethis implies that we as researchers have to stronglyconsider not only what new things technologies canhelp people do but also consider simplicity and speedas fundamental goals

31 Why ZUIs are challenging

As summarised in Table 2 the potential benefits ofZUIs are sometimes mirages ZUIs are generally

Table 2 Promise and reality of ZUIs

Promise Reality

Engaging If designed well The animated nature of ZUIs is sometimes referred to as lsquoeye candyrsquo implying that peoplecannot help but pay attention to the movement on the screen However this very movement is a double-edgedsword because if the user does not pay attention and misses the animation then the connection between thedifferent views is lost and the user is likely worse off than without a ZUI

Visually rich Can also make space more complex to navigate and understand ZUIs can have visual objects at any size and atany place ndash thus offering the possibility of placing significant content (scaled down) even in a tiny crackbetween two other objects This kind of structure is very difficult to perceive (since users cannot see the tinycontent in the first place) and conceptualise (since at any given view the vastly different sized objects can notbe seen simultaneously)

Simplicity Does not scale to very large data sets One of the attractions of ZUIs is that they appear conceptually simple(lsquojust zoom in for a closer lookrsquo) But in practice navigating large information spaces in ZUIs is often difficultto execute and requires significant short-term memory to navigate

Behaviour amp Information Technology 861

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

engaging (although they make some people feelphysically sick) and visually rich But the promise ofsimplicity falls short

While human visual perception does make it easy tosee where one is navigating the reality is that it places aheavy load on short-termmemory to remember where inspace you just were and where things are And therequirement of human memory to know how space isorganised means that ZUIs do not scale up very wellZUIs are oftenmotivated by the physical world and howpeople like laying papers out on their desk But no onewants all of their papers on their desk It is much morecommon to have only a relatively small number ofpapers that are actually being worked with

The visual overviews that ZUIs offer for free byzooming out may seem like a solution to the load onhuman memory but in practice it does not becausevisual overviews of any complexity require significantvisual search in order to find anything If there are just asmall number of objects then the visual search task is nothard ndash but of course for a small number of objects youdo not need a ZUI to solve your organisationalproblems

In addition the visual richness of ZUIs is a double-edged sword It requires skill to design a complex spacewith documents of arbitrary size aspect ratio andcolour that people can comprehend and scan Alsopeople are not as good at scanning 2D designs as 1Dlayouts ndash unless the layout is highly structured Buthighly structured 2D layouts do not work well forvisual objects of arbitrary aspect ratios Designers areobligated to leave a lot of unused space scale down thelarge objects so they are unreadable or crop the largeobjects ndash thus losing much of their distinguishability

Finally it is crucial to consider the efficiency ofZUIs and their alternatives It is the nature ofzoomable structures to organise information hierarchi-cally Zooming in to an object shows you moreinformation about that object Of course hierarchicalstructures are also commonly exposed with a simpletree widget as is commonly found in most file systembrowsers The file system browsers exemplified byWindows Explorer are not only very efficient but alsomore easily let the user compare different areas in thehierarchy by simply opening and collapsing the desirednodes So in many cases the alternative to a ZUI is asimple tree widget that is simple compact easy tolearn and fast to use Thus in addition to the broaderchallenges of ZUIs we should always consider thealternatives that are also powerful

32 Design guidelines

So how should ZUIs be designed to best overcomethese obstacles Let us start with the big issues

321 Support the right tasks

ZUIs like most visualisations have strong potentialfor supporting users in understanding the big pictureidentifying trends patterns and outliers But theytypically are not good at helping users simply get thebest answer So be sure that some sort of overview-based design is called for in the first place

322 Be cautious about using zooming as the primaryinterface

Zooming navigation can be difficult and is uncom-mon Unless the data set is fundamentally spatial (egmaps) users probably do not want zooming as theirprimary interaction with the data Instead think abouthow people search and create data sets ndash which maythen be accessed partially by zooming ie createhybrid systems that combine mechanisms such asfaceted search and zooming [eg PhotoMesa (Beder-son 2001) the Hard Rock Memorabilia (2010) orMicrosoft Pivot (2010)]

323 Only use ZUIs when a small visual representa-tion of the data is available

Zooming works best when the objects can be recognisedwhen they are zoomed out So certain domains arebetter suited to ZUIs Photos are good purely textualdocuments are bad and audio recordings are terrible(unless there are associated images) Recent researchshows how a small visual snippet may be automaticallycreated from web pages (Teevan et al 2009)

324 Donrsquot limit yourself to fixed data sets

People rarely want to interact long with a fixed set ofdata But it is often technically difficult to dynamicallycreate and modify large zoomable spaces Be sure thatthe user interface does not get blocked or slowed downwhen new data is added to a zoomable space

And then there are several more specificsuggestions

325 Maintain aspect ratio

It is tempting to use semantic zooming to signifi-cantly change the visual representation of objects atdifferent sizes For example a visual chart may becondensed to a single value or a text document maybe replaced by its title when small However thesmall object must have the same aspect ratio ndash or atleast fit in the same box as the full object Otherwisenearby objects can end up overlapping when the userzooms out

862 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

326 Use meaningful layout

The problem of finding a specific object among a largenumberof themcanbe alleviated if theobjects are laidoutin a semantically meaningful way If the user can predictwhere an object will be based on attributes that they arelikely to be familiar with then they can trim the searchspace Traditional paper phones books with theiralphabetic ordering are a good example of this approach

327 Use consistent layout

Ideally once an object is placed in space it should notmove That way the user can build up an internal mapover time of where objects are Unfortunately thisideal is often in direct conflict with the previous one ofmeaningful layout If objects are to be laid outmeaningfully then the layout must change over timeas the data changes and as the organisationalmechanism changes PhotoMesa (Bederson 2001) isan example of a ZUI that went for a meaningful layoutbut had no consistency at all

328 Use scannable layout

Objects must be positioned in such a manner as to enablethe user to visually scan the space to look for an objectwithout missing or repeating objects Simple 1D lists ofobjects are easy to scanbecause the person just runs downthe list and the only state that must be kept is the currentitem and the direction of scanning In 2D however forany layout except for a grid there is the potential for amuch heavier load on short-term memory

329 Use breadth over depth

It is tempting to place objects in space at manydifferent sizes so more information can be discoveredby zooming in further ndash and in fact many of the earlierZUI applications did this However when objects areplaced at many different levels the user would notknow that the small objects exist (unless the space iscarefully designed to indicate them) and they can getlost Further even if the user knows they are there it istypically burdensome to zoom in and out repeatedlySpaceTree (Grosjean et al 2002) is an example of anapplication that was designed explicitly with breadth inmind Even with hundreds of thousands of objects allof them are at the same size

3210 Use small data sets

The problems that come with a large number of objectsin a zoomable space can be mitigated by not buildingapplications around large data sets ZUIs appear to

work better for users when they have no more thanabout 100 screens worth of information HoweverZUIs with 1000 or more screens worth of informationbecome problematic because of the issues raised in theprevious section (why ZUIs are challenging) This is afundamental problem since ZUIs are often designed towork for user generated data ndash which may not easily belimited So application designers may choose topartition the space into manageable subsets

3211 Use simple and consistent navigationinteraction

People must be able to discover how to navigate thespace with no training and it must be difficult orimpossible to get lost in Desert Fog Unfortunatelythere is significant inconsistency among applicationsThey vary between single and double clicks to zoom inhow much the screen is zoomed per action whetherzooms scale by a fixed percentage or whether the targetobject fills the screen And mechanisms are all over themap to zoom out ndash from right click left click on thebackground fixed on-screen buttons and pop-upbuttons And as with zoom in the actual amountthat is zoomed out is inconsistent across applicationsas well

4 Conclusion

Understanding ZUIs is difficult They have clearlybeen a success by traditional research measures(significant visibility publications and citations) Andelements of ZUIs have been broadly deployed incommonly used applications But the original broadvision of zooming as a basic information organisa-tional principle [such as Raskinrsquos vision of hospitalmanagement (Raskin 2000)] has not happened Ofcourse lack of commercial success does not mean anidea is not a good one (maybe the business reasons didnot line up or maybe the right tools got built but withthe wrong marketing strategy the timing may havebeen wrong or any number of other reasons mighthave stalled a successful business) But still the lack ofbroad success coupled with the fundamental issuesraised in this paper appears significant

I definitely would not rule out further success ofZUIs and I think that the successes that haveoccurred such as Google Maps Microsoft Word andApple iPhone are substantial However their applic-ability is much narrower than original envisioned Andthe design and technical challenges of building effectiveZUIs make it that much less likely that they will bebroadly deployed At the same time there are clearcases where zooming is valuable ndash such as withdocuments that are larger than screens (ie photos

Behaviour amp Information Technology 863

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

engaging (although they make some people feelphysically sick) and visually rich But the promise ofsimplicity falls short

While human visual perception does make it easy tosee where one is navigating the reality is that it places aheavy load on short-termmemory to remember where inspace you just were and where things are And therequirement of human memory to know how space isorganised means that ZUIs do not scale up very wellZUIs are oftenmotivated by the physical world and howpeople like laying papers out on their desk But no onewants all of their papers on their desk It is much morecommon to have only a relatively small number ofpapers that are actually being worked with

The visual overviews that ZUIs offer for free byzooming out may seem like a solution to the load onhuman memory but in practice it does not becausevisual overviews of any complexity require significantvisual search in order to find anything If there are just asmall number of objects then the visual search task is nothard ndash but of course for a small number of objects youdo not need a ZUI to solve your organisationalproblems

In addition the visual richness of ZUIs is a double-edged sword It requires skill to design a complex spacewith documents of arbitrary size aspect ratio andcolour that people can comprehend and scan Alsopeople are not as good at scanning 2D designs as 1Dlayouts ndash unless the layout is highly structured Buthighly structured 2D layouts do not work well forvisual objects of arbitrary aspect ratios Designers areobligated to leave a lot of unused space scale down thelarge objects so they are unreadable or crop the largeobjects ndash thus losing much of their distinguishability

Finally it is crucial to consider the efficiency ofZUIs and their alternatives It is the nature ofzoomable structures to organise information hierarchi-cally Zooming in to an object shows you moreinformation about that object Of course hierarchicalstructures are also commonly exposed with a simpletree widget as is commonly found in most file systembrowsers The file system browsers exemplified byWindows Explorer are not only very efficient but alsomore easily let the user compare different areas in thehierarchy by simply opening and collapsing the desirednodes So in many cases the alternative to a ZUI is asimple tree widget that is simple compact easy tolearn and fast to use Thus in addition to the broaderchallenges of ZUIs we should always consider thealternatives that are also powerful

32 Design guidelines

So how should ZUIs be designed to best overcomethese obstacles Let us start with the big issues

321 Support the right tasks

ZUIs like most visualisations have strong potentialfor supporting users in understanding the big pictureidentifying trends patterns and outliers But theytypically are not good at helping users simply get thebest answer So be sure that some sort of overview-based design is called for in the first place

322 Be cautious about using zooming as the primaryinterface

Zooming navigation can be difficult and is uncom-mon Unless the data set is fundamentally spatial (egmaps) users probably do not want zooming as theirprimary interaction with the data Instead think abouthow people search and create data sets ndash which maythen be accessed partially by zooming ie createhybrid systems that combine mechanisms such asfaceted search and zooming [eg PhotoMesa (Beder-son 2001) the Hard Rock Memorabilia (2010) orMicrosoft Pivot (2010)]

323 Only use ZUIs when a small visual representa-tion of the data is available

Zooming works best when the objects can be recognisedwhen they are zoomed out So certain domains arebetter suited to ZUIs Photos are good purely textualdocuments are bad and audio recordings are terrible(unless there are associated images) Recent researchshows how a small visual snippet may be automaticallycreated from web pages (Teevan et al 2009)

324 Donrsquot limit yourself to fixed data sets

People rarely want to interact long with a fixed set ofdata But it is often technically difficult to dynamicallycreate and modify large zoomable spaces Be sure thatthe user interface does not get blocked or slowed downwhen new data is added to a zoomable space

And then there are several more specificsuggestions

325 Maintain aspect ratio

It is tempting to use semantic zooming to signifi-cantly change the visual representation of objects atdifferent sizes For example a visual chart may becondensed to a single value or a text document maybe replaced by its title when small However thesmall object must have the same aspect ratio ndash or atleast fit in the same box as the full object Otherwisenearby objects can end up overlapping when the userzooms out

862 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

326 Use meaningful layout

The problem of finding a specific object among a largenumberof themcanbe alleviated if theobjects are laidoutin a semantically meaningful way If the user can predictwhere an object will be based on attributes that they arelikely to be familiar with then they can trim the searchspace Traditional paper phones books with theiralphabetic ordering are a good example of this approach

327 Use consistent layout

Ideally once an object is placed in space it should notmove That way the user can build up an internal mapover time of where objects are Unfortunately thisideal is often in direct conflict with the previous one ofmeaningful layout If objects are to be laid outmeaningfully then the layout must change over timeas the data changes and as the organisationalmechanism changes PhotoMesa (Bederson 2001) isan example of a ZUI that went for a meaningful layoutbut had no consistency at all

328 Use scannable layout

Objects must be positioned in such a manner as to enablethe user to visually scan the space to look for an objectwithout missing or repeating objects Simple 1D lists ofobjects are easy to scanbecause the person just runs downthe list and the only state that must be kept is the currentitem and the direction of scanning In 2D however forany layout except for a grid there is the potential for amuch heavier load on short-term memory

329 Use breadth over depth

It is tempting to place objects in space at manydifferent sizes so more information can be discoveredby zooming in further ndash and in fact many of the earlierZUI applications did this However when objects areplaced at many different levels the user would notknow that the small objects exist (unless the space iscarefully designed to indicate them) and they can getlost Further even if the user knows they are there it istypically burdensome to zoom in and out repeatedlySpaceTree (Grosjean et al 2002) is an example of anapplication that was designed explicitly with breadth inmind Even with hundreds of thousands of objects allof them are at the same size

3210 Use small data sets

The problems that come with a large number of objectsin a zoomable space can be mitigated by not buildingapplications around large data sets ZUIs appear to

work better for users when they have no more thanabout 100 screens worth of information HoweverZUIs with 1000 or more screens worth of informationbecome problematic because of the issues raised in theprevious section (why ZUIs are challenging) This is afundamental problem since ZUIs are often designed towork for user generated data ndash which may not easily belimited So application designers may choose topartition the space into manageable subsets

3211 Use simple and consistent navigationinteraction

People must be able to discover how to navigate thespace with no training and it must be difficult orimpossible to get lost in Desert Fog Unfortunatelythere is significant inconsistency among applicationsThey vary between single and double clicks to zoom inhow much the screen is zoomed per action whetherzooms scale by a fixed percentage or whether the targetobject fills the screen And mechanisms are all over themap to zoom out ndash from right click left click on thebackground fixed on-screen buttons and pop-upbuttons And as with zoom in the actual amountthat is zoomed out is inconsistent across applicationsas well

4 Conclusion

Understanding ZUIs is difficult They have clearlybeen a success by traditional research measures(significant visibility publications and citations) Andelements of ZUIs have been broadly deployed incommonly used applications But the original broadvision of zooming as a basic information organisa-tional principle [such as Raskinrsquos vision of hospitalmanagement (Raskin 2000)] has not happened Ofcourse lack of commercial success does not mean anidea is not a good one (maybe the business reasons didnot line up or maybe the right tools got built but withthe wrong marketing strategy the timing may havebeen wrong or any number of other reasons mighthave stalled a successful business) But still the lack ofbroad success coupled with the fundamental issuesraised in this paper appears significant

I definitely would not rule out further success ofZUIs and I think that the successes that haveoccurred such as Google Maps Microsoft Word andApple iPhone are substantial However their applic-ability is much narrower than original envisioned Andthe design and technical challenges of building effectiveZUIs make it that much less likely that they will bebroadly deployed At the same time there are clearcases where zooming is valuable ndash such as withdocuments that are larger than screens (ie photos

Behaviour amp Information Technology 863

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

326 Use meaningful layout

The problem of finding a specific object among a largenumberof themcanbe alleviated if theobjects are laidoutin a semantically meaningful way If the user can predictwhere an object will be based on attributes that they arelikely to be familiar with then they can trim the searchspace Traditional paper phones books with theiralphabetic ordering are a good example of this approach

327 Use consistent layout

Ideally once an object is placed in space it should notmove That way the user can build up an internal mapover time of where objects are Unfortunately thisideal is often in direct conflict with the previous one ofmeaningful layout If objects are to be laid outmeaningfully then the layout must change over timeas the data changes and as the organisationalmechanism changes PhotoMesa (Bederson 2001) isan example of a ZUI that went for a meaningful layoutbut had no consistency at all

328 Use scannable layout

Objects must be positioned in such a manner as to enablethe user to visually scan the space to look for an objectwithout missing or repeating objects Simple 1D lists ofobjects are easy to scanbecause the person just runs downthe list and the only state that must be kept is the currentitem and the direction of scanning In 2D however forany layout except for a grid there is the potential for amuch heavier load on short-term memory

329 Use breadth over depth

It is tempting to place objects in space at manydifferent sizes so more information can be discoveredby zooming in further ndash and in fact many of the earlierZUI applications did this However when objects areplaced at many different levels the user would notknow that the small objects exist (unless the space iscarefully designed to indicate them) and they can getlost Further even if the user knows they are there it istypically burdensome to zoom in and out repeatedlySpaceTree (Grosjean et al 2002) is an example of anapplication that was designed explicitly with breadth inmind Even with hundreds of thousands of objects allof them are at the same size

3210 Use small data sets

The problems that come with a large number of objectsin a zoomable space can be mitigated by not buildingapplications around large data sets ZUIs appear to

work better for users when they have no more thanabout 100 screens worth of information HoweverZUIs with 1000 or more screens worth of informationbecome problematic because of the issues raised in theprevious section (why ZUIs are challenging) This is afundamental problem since ZUIs are often designed towork for user generated data ndash which may not easily belimited So application designers may choose topartition the space into manageable subsets

3211 Use simple and consistent navigationinteraction

People must be able to discover how to navigate thespace with no training and it must be difficult orimpossible to get lost in Desert Fog Unfortunatelythere is significant inconsistency among applicationsThey vary between single and double clicks to zoom inhow much the screen is zoomed per action whetherzooms scale by a fixed percentage or whether the targetobject fills the screen And mechanisms are all over themap to zoom out ndash from right click left click on thebackground fixed on-screen buttons and pop-upbuttons And as with zoom in the actual amountthat is zoomed out is inconsistent across applicationsas well

4 Conclusion

Understanding ZUIs is difficult They have clearlybeen a success by traditional research measures(significant visibility publications and citations) Andelements of ZUIs have been broadly deployed incommonly used applications But the original broadvision of zooming as a basic information organisa-tional principle [such as Raskinrsquos vision of hospitalmanagement (Raskin 2000)] has not happened Ofcourse lack of commercial success does not mean anidea is not a good one (maybe the business reasons didnot line up or maybe the right tools got built but withthe wrong marketing strategy the timing may havebeen wrong or any number of other reasons mighthave stalled a successful business) But still the lack ofbroad success coupled with the fundamental issuesraised in this paper appears significant

I definitely would not rule out further success ofZUIs and I think that the successes that haveoccurred such as Google Maps Microsoft Word andApple iPhone are substantial However their applic-ability is much narrower than original envisioned Andthe design and technical challenges of building effectiveZUIs make it that much less likely that they will bebroadly deployed At the same time there are clearcases where zooming is valuable ndash such as withdocuments that are larger than screens (ie photos

Behaviour amp Information Technology 863

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

web pages and maps) And the fact that they arevisually engaging cannot be ignored

Furthermore as described in this paper there is arange of potential benefits For some users and tasksthose benefits will outweigh the costs The problemthough is that the precise balance is not well under-stood Furthermore the full range of zoomable designsis surely not fully explored This leads to the followingkey areas for further research

41 Definitive understanding of benefit

What specifically are the benefits of zooming transi-tions and under what circumstances are those benefitsfound Do different users have different performancecapabilities and preferences when using ZUIs

42 Standardised navigation

ZUIs need standardised navigation mechanisms Somany approaches have been used that they need to becatalogued so that it becomes easier to see what aregood candidates for standards And any companiesthat make more than one ZUI application must ensurethat they at least have a company-wide standard

43 Design space exploration

Despite all the effort over the years envisioninginteraction design and application domains forZUIs it is a very large space design There are verylikely still new approaches and applications to bediscovered I expect that creative design effort willcontinue to extend the richness and reach of ZUIapplications

44 Reflections

In retrospect I believe that we as researchers some-times let ourselves get too caught up in the excitementof something new and the coolness of something soengaging When we could not figure out what the killerapp in a zoomable environment was perhaps weshould not have been so fast to think that others wouldfigure it out if only we solved the technologicalproblems Perhaps one or two platforms were a goodidea but were five necessary

The idea that we should only develop technologywhen we know what it is for is dangerous because it isthe nature of basic research that you do not alwaysknow what a new technology will enable Transistorsand lasers come to mind as technologies that theinventors had no idea what they were good for

But I believe that in humanndashcomputer interaction(HCI) which is user facing by its nature it should be

much more apparent what the value to users is whendeveloping technology An example of HCI effort thatsaw more success only when researchers focused onuser needs more than underlying technology is soft-ware engineering (Myers and Ko 2009) We are notdeveloping technology for its own sake we aredeveloping technology to address user needs and ifwe do not start by addressing those needs we areputting our effort at risk This is a point that seems tooobvious to make except that many of us (especiallycomputer scientists) sometimes miss it

Acknowledgments

I have worked with many people on ZUIs over the past 18years since Ken Perlinrsquos first paper on Pad I canrsquot thankthem all but want to acknowledge the people I worked mostclosely with on ZUIs especially in the early days Thisincludes of course Ken Perlin who had the originalconceptualisation of ZUIs immediately followed by closecollaborations on Pad thorn thorn with me Jim Hollan Jon Meyerand Allison Druin I also appreciate the enthusiasm andcommercial efforts of Franklin Servan-Schreiber who Iworked with when he was at Sony I worked closely withstudents Lance Good and Bongwon Suh and staff JesseGrosjean and Aaron Clamage Then my later work withMary Czerwinski and George Robertson at Microsoft andJohn SanGiovanni first at Microsoft and then at Zumobihelped me to understand more closely what it really takes tounderstand an interface polish a product and make it workfor a large audience And finally my colleagues at the HCIL(especially Ben Shneiderman and Catherine Plaisant) havebeen instrumental in being rigorous and ruthless in helpingme think about the real costs and benefits of the technologieswe build

Note

1 People commonly say lsquozoomable user interfacesrsquo orlsquozooming user interfacesrsquo I prefer the former since itfocuses on users being in control However there isneither a substantive difference between these terms norpredominant usage of one over the other

References

Apple iPhone [online] Available from wwwapplecomiphone [Accessed 7 January 2010]

Apple OS X Dock [online] Available from wwwapplecommacosxwhat-is-macosxdock-and-finderhtml [Accessed7 January 2010]

Bauer DS 2006 The cognitive ecology of dynapad amultiscale workspace for managing personal digitalcollections Thesis (PhD) UMI Order NumberAAI3222983 University of California at San Diego

Beard DB and Walker JQ 1990 Navigationaltechniques to improve the display of large two-dimen-sional spaces Behavior and Information Technology 9 (6)451ndash466

Bederson BB 1993 PA3D early zoomable user interface(ZUI) [online] Available from wwwyoutubecomwatchvfrac145JzaEUJ7IbE [Accessed 7 January 2010]

864 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Bederson BB 2001 PhotoMesa a zoomable imagebrowser using quantum treemaps and bubblemaps InProceedings of the ACM Symposium on User interfaceSoftware and Technology UIST 2001 New York NYACM 71ndash80 DOI httpdoiacmorg101145502348502359

Bederson BB et al 2004a DateLens a fisheye calendarinterface for PDAs ACM Transactions on ComputerndashHuman Interaction 11 (1) 90ndash119 DOI httpdoiacmorg101145972648972652

Bederson BB Grosjean J and Meyer J 2004b Toolkitdesign for interactive structured graphics IEEE Transac-tions on Software Engineering 30 (8) 535ndash546 DOIhttpdxdoiorg101109TSE200444

Bederson BB and Hollan JD 1994 Padthornthorn a zoominggraphical interface for exploring alternate interfacephysics In Proceedings of the 7th Annual ACMSymposium on User interface Software and TechnologyUIST 1994 New York NY ACM 17ndash26 DOI httpdoiacmorg101145192426192435

Bederson BB et al 1996 Local tools an alternative totool palettes In Proceedings of the 9th Annual ACMSymposium on User interface Software and TechnologyUIST 1996 New York NY ACM 169ndash170 DOIhttpdoiacmorg101145237091237116

Bederson BB et al 1996 Padthornthorn a zoomable graphicalsketchpad for exploring alternate interface physicsJournal of Visual Languages and Computing 7 3ndash31

Bederson BB and Meyer J 1998 Implementing a zoominguser interface experience building Padthornthorn SoftwarePractice and Experience 28 (10) 1101ndash1135

Bederson BB Meyer J and Good L 2000 Jazz anextensible zoomable user interface graphics toolkit inJava In Proceedings of the 13th Annual ACM Sympo-sium on User interface Software and Technology UIST2000 New York NY ACM 171ndash180 DOI httpdoiacmorg101145354401354754

Boltman A et al 2002 The nature of childrenrsquosstorytelling with and without technology InProceedings of American Educational Research Associa-tion (AERA) Conference New Orleans WashingtonDC AREA

Card SK Moran TP and Newell A 1986 Thepsychology of humanndashcomputer interaction MahwahNJ Lawrence Earlbaum Associates

Card SK Robertson GG and Mackinlay JD 1991The information visualizer an information workspaceIn Proceedings of the SIGCHI Conference on HumanFactors in Computing Systems CHI 1991 New YorkNY ACM 181ndash186 DOI httpdoiacmorg101145108844108874

Cockburn A Karlson A and Bederson BB 2008 Areview of overview thorn detail zooming and focus thorn con-text interfaces ACM Computer Surveys 41 (1) 1ndash31DOI httpdoiacmorg10114514566501456652

Cockburn A Savage J and Wallace A 2005 Tuning andtesting scrolling interfaces that automatically zoom InProceedings of the SIGCHI Conference on Human Factorsin Computing Systems CHI 2005 New York NY ACM71ndash80 DOI httpdoiacmorg10114510549721054983

Combs TT and Bederson BB 1999 Does zoomingimprove image browsing In Proceedings of the FourthACM Conference on Digital Libraries DL 1999 NewYork NY ACM 130ndash137 DOI httpdoiacmorg101145313238313286

Druin A et al 1997 KidPad a design collaborationbetween children technologists and educators InProceedings of the SIGCHI Conference on Human Factorsin Computing Systems CHI 1997 New York NY ACM463ndash470 DOI httpdoiacmorg101145258549258866

Furnas GW 1986 Generalized fisheye views ACMSIGCHI Bulletin 17 (4) 16ndash23 DOI httpdoiacmorg1011452233922342

Furnas GW and Bederson BB 1995 Space-scalediagrams understanding multiscale interfaces In Pro-ceedings of the SIGCHI Conference on Human Factors inComputing Systems CHI 1995 New York NY ACMPressAddison-Wesley 234ndash241 DOI httpdoiacmorg101145223904223934

Good L and Bederson BB 2002 Zoomable user interfacesas a medium for slide show presentations InformationVisualization 1 (1) 35ndash49 DOI httpdxdoiorg101057palgraveivs9500004

Grosjean J Plaisant C and Bederson BB 2002 Space-Tree design evolution of a node link tree browser InProceedings of Information Visualization Symposium(InfoVis 2002) New York IEEE 57ndash64

Hard Rock Memorabilia [online] Available from httpmemorabiliahardrockcom [Accessed 7 January2010]

Hightower RR et al 1998 PadPrints graphical multiscaleWeb histories In Proceedings of the 11th Annual ACMSymposium on User interface Software and TechnologyUIST 1998 New York NY ACM 121ndash122 DOIhttpdoiacmorg101145288392288582

Hillcrest Labs [online] Available from wwwhillcrestlabscom [Accessed 7 January 2010]

Hornbaeligk K Bederson BB and Plaisant C 2002Navigation patterns and usability of zoomable userinterfaces with and without an overview ACM Transac-tions ComputerndashHuman Interaction 9 (4) 362ndash389 DOIhttpdoiacmorg101145586081586086

Igarashi T and Hinckley K 2000 Speed-dependentautomatic zooming for browsing large documents InProceedings of the ACM Symposium on User interfaceSoftware and Technology UIST 2000 New York NYACM 139ndash148 DOI httpdoiacmorg101145354401354435

Ishii H and Ullmer B 1997 Tangible bits towardsseamless interfaces between people bits and atoms InProceedings of the SIGCHI Conference on Human Factorsin Computing Systems CHI 1997 New York ACM234ndash241 DOI httpdoiacmorg101145258549258715

Jul S and Furnas GW 1998 Critical zones in desertfog aids to multiscale navigation In Proceedings ofthe 11th Annual ACM Symposium on User interfaceSoftware and Technology UIST 1998 New York NYACM 97ndash106 DOI httpdoiacmorg101145288392288578

Karlson AK Bederson BB and SanGiovanni J 2005AppLens and launchTile two designs for one-handedthumb use on small devices In Proceedings of theSIGCHI Conference on Human Factors in ComputingSystems CHI 2005 New York NY ACM 201ndash210DOI httpdoiacmorg10114510549721055001

Klein C and Bederson BB 2005 Benefits of animatedscrolling In Extended Abstracts on Human Factors inComputing Systms CHI 2005 New York ACM 1965ndash1968 DOI httpdoiacmorg10114510568081057068

Behaviour amp Information Technology 865

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11

Kunkel P 1999 Digital dreams the work of the Sony designcenter New York Universe Publishing

Lichtschlag L Karrer T and Borchers J 2009 Fly a toolto author planar presentations In Proceedings of the27th International Conference on Human Factors inComputing Systems CHI 2009 New York NY ACM547ndash556 DOI httpdoiacmorg10114515187011518786

Microsoft Canvas for OneNote [online] Available fromwwwofficelabscomprojects canvasforonenote [Ac-cessed 7 January 2010]

Microsoft Pivot [online] Available from httpresearchmi-crosoftcomen-usdownloadsdd4a479f-92d6-496f-867d-666c87fbaadadefault [Accessed 7 January 2010]

Microsoft pptPlex [online] Available from wwwofficelabscomprojectspptPlex [Accessed 7 January 2010]

Microsoft Seadragon [online] Available from wwwseadragoncom [Accessed 7 January 2010]

Myers B and Ko AJ 2009 The past present and future ofprogramming in HCI In HumanndashComputer InteractionConsortium (HCIC 2009) Winter Park CO FebruaryAvailable from httpwwwcscmuedunatprogpapersMyersKoHCIC09pdf

North C and Shneiderman B 2000 Snap-togethervisualization evaluating coordination usage and con-struction International Journal of Human ComputerStudies 53 (5) 715ndash739

Perlin K and Fox D 1993 Pad an alternative approach tothe computer interface In Proceedings of the 20th AnnualConference on Computer Graphics and interactive Tech-niques SIGGRAPH 1993 New York NY ACM 57ndash64DOI httpdoiacmorg101145166117166125

Piccolo2D [online] Available from wwwpiccolo2dorg[Accessed 7 January 2010]

Plumlee M and Ware C 2002 Zooming multiplewindows and visual working memory In M DeMarsico S Levialdi and E Panizzi eds Proceedingsof the Working Conference on Advanced Visual Interfaces(AVI lsquo02) New York NY ACM 59ndash68 DOI httpdoiacmorg10114515562621556270

Plumlee MD and Ware C 2006 Zooming versus multiplewindow interfaces cognitive costs of visual comparisonsACM Transactions on ComputerndashHuman Interaction 13(2) 179ndash209 DOI httpdoiacmorg10114511657341165736

Prezi [online] Available from wwwprezicom [Accessed 7January 2010]

Rao R and Card SK 1994 The table lens merginggraphical and symbolic representations in an interactivefocus thorn context visualization for tabular information InProceedings of the SIGCHI Conference on Human Factorsin Computing Systems CHI 1994 New York NY ACM318ndash322 DOI httpdoiacmorg101145191666191776

Raskin J 2000 The humane interface Upper Saddle RiverNJ Addison-Wesley Professional

Rekimoto J 2002 SmartSkin an infrastructure for free-hand manipulation on interactive surfaces In Proceed-ings of the SIGCHI Conference on Human Factors inComputing Systems CHI 2002 New York ACM 113ndash120 DOI httpdoiacmorg101145503376503397

Robertson G et al 2000 The task gallery a 3D windowmanager In Proceedings of the SIGCHI Conference onHuman Factors in Computing Systems CHI 2000 NewYork ACM 494ndash501 DOI httpdoiacmorg101145332040332482

Schematic [online] Available from wwwschematiccom[Accessed 7 January 2010]

Sun L and Guimbretiere F 2005 Flipper a new methodof digital document navigation In Extended Abstracts onHuman Factors in Computing Systems CHI 2005 NewYork NY ACM 2001ndash2004 DOI httpdoiacmorg10114510568081057077

Teevan J et al 2009 Visual snippets summarizing webpages for search and revisitation In InternationalConference on Human Factors in Computing SystemsCHI 2009 New York NY ACM 2023ndash2032 DOIhttpdoiacmorg10114515187011519008

van Wijk J and Nuij W 2004 A model for smooth viewingand navigation of large 2D information spaces IEEETransactions on Visualization and Computer Graphics 10(4) 447ndash458

Ware C 2004 Information visualization 2nd ed WalthamMA Morgan Kaufmann

Zoomorama [online] Available from wwwzoomoramacom[Accessed 7 January 2010]

Zumobi [online] Available from wwwzumobicom [Ac-cessed 7 January 2010]

866 BB Bederson

Dow

nloa

ded

by [

Uni

vers

ity O

f M

aryl

and]

at 0

513

13

Dec

embe

r 20

11