The Visual Display
The Visual Display
The Visual Display
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Efiaard R.Tkfte<br />
<strong>The</strong> <strong>Visual</strong> <strong>Display</strong><br />
of Quantitative Information<br />
SECOND EDITION<br />
Graphics Prex Cheshire, Connecticut
T 44$3 ex,<br />
[\<br />
Coppight o 2oor by rdw:rd RolfTbfte<br />
?UBTISEED BY GRAPHTCS PRTSS ILC<br />
?osr OFFTCE Box 43o, CqtsgtRB, CoNNBcrrcur 06410<br />
rm nsha ro illtfitiotu dd b. Fed byldftd RolrTnftc. This worL n.y noi 6e @pieA EFodlccd, or hnihhn in vnoL or in<br />
I'd wirlDut wntu Fmi$id of i!. prujlnd, .cpt for bn€r*Gpts in cotucdion wilh Eviflj d lcnolIly :nt,5n. Us wirL ,,1 fod<br />
ofiDfo@arion *onse ra diml d@ic rlzpdid o. v!aEv*, cohpub sfMre, ft by sinnar u ;n{ihnr n€drodr now tnM<br />
edevdop€dinth€f!tueis&ridryfdn dd witnoft Eien Foni$ion '5ir!e !u61i5tE. A lubd of ilufatio4 e.epro&ccd !y<br />
Fni$iu; thF copyriglr+olds s d.diEd d Fge i9?.<br />
in the United States of Ansie4 Second edition, fowth yi ing,Januaryoo6
Contents<br />
PART I GRAPHICAL PRACTICE<br />
r Crcphiml Excellence 4<br />
z Gruphial Integriry J3<br />
3 So*tes oJ Ctaphical httegity afld Sophistiatiot 79<br />
PART II THEORY OF DATA GRAPHICS<br />
4 Data-Ink and Ctaphical Reiluigx 91<br />
5 Chattjunk: Vibrations, Crids, and Ducks 1o7<br />
6 Data-Ink Maxr ization and Craphical Design 721,<br />
7 Mtabijwctioning Graphical Elements r3g<br />
8 Dota Densiti and Suall Mubiples 161<br />
9 Aesthetics and Technique in Data Graphial Design 177<br />
Epikgue: Designs<br />
fu<br />
the Dkplay of Infornation r9r
Introdwction to the Second Edition<br />
This new edition provides high-resolution color reproductions of<br />
the many graphic.s ofWWifiam ?hyfair, adds color co other images<br />
*Lere appropriate, and includes all the changes a:rd correctiom<br />
zccumulated during the r7 printings ofthe fust edition.<br />
This book bega:r in 1975 when Dean Donald StoLes ofprinceton s<br />
Woo&ow Wilson School asLed me to teach statistics to a dozen<br />
i>umalists who llere vi)jiing rhrr ycar ro lerm some economjcr<br />
I annotated o collection ofre"dirgs. wirh r long seccion on<br />
*atistical graphics. <strong>The</strong> Jiterarue here was thin, roo often erimly<br />
devored to expl$ring use ofthe rulng oen and to p-..'E"ti"g<br />
"graphic<br />
standards" indiferent to the nature ofvisual evidence and<br />
qu:ntitative reasoning. Soon I wrote up some ideas. <strong>The</strong>nJohr<br />
TuLey, the pherromenal Princeton statistician, suggested that we<br />
give a series ofjoinr seminars. Since the mid-r96os, Tukey h:d<br />
opened up the field, as his brilliant tecLnical contributiom made<br />
it dear that the study ofstatistical graphics was intellecturlly<br />
tspecuble nd notjusr abour pic chars rnd rLrlirg pens.<br />
After moving to Yale University, Ifnished the maauscript in<br />
1982. A publisher was interested but planned to print only z,ooo<br />
copies and to charge a very high price, contrary to my hopes for<br />
a vride readcrship. I also sought to design the book so as to make<br />
n selJ-exenplfying that is, the physical object itselfwould reflect<br />
tf,c intellcctual principles advanced in the book. ?ublishers seemed<br />
.ppalled at the prospect that an author might govcm design,<br />
Consequendy I investigated sclEpublishing. Tlns required a frstrrte<br />
book desigocr, a lot ofmoney (at least for a young professor),<br />
and a large garage. I fotmd Howard Gralh who had designed<br />
mauy museum catalogs with great care and cnft. He was willirig to<br />
work closely with this diiicult ruthor who was f,lled with all sorts<br />
ofopiniorx about design and typogmphy. We spent the summer in
his studio laying out the book, page by page. We were able to<br />
integrate graphics right into the text, son:retimes into the middle<br />
ofa sentence, eliminating the usual separation oftext and imageone<br />
ofthe ideas Zrsai Drsplay advocated. To fnance the book<br />
I took out another mongage on my home. <strong>The</strong> bank oflicer said<br />
this was the second most unusual loan that she had ever made; 6rst<br />
place belonged to a lo:n to a circus to buy an elephantl<br />
My view on selipublishirlg was to go all out, to maLe the best<br />
rnd most elegant and wonderful booL possible, withour compromise.<br />
Otherwise, why do it?<br />
Moit ofall, the book, as a thing in itsef gave to me fresh new<br />
eyes for the intellectual and aestheticjoy ofvisual evidence, visual<br />
reasoning, and visual understanding.<br />
Cheshire, Conneetiett
Introduction<br />
Data graphics visually display mcasured quantities by means of<br />
the conbined use ofpoinrs, lnres, a coordinare system, numbcrs,<br />
symbols, words, shading, and color.<br />
<strong>The</strong> use ofabstract, uon-reprcscntational pictures to show numbers<br />
is a surprisingly rccent invention, perhaps because of the divcrsity<br />
of skills required-rhe visual-anisdc, empirical-statistical, and<br />
mathematical. It was not until r75o-r8oo that sratistical graphics_<br />
lengdr and area ro show quantity, time-series, scarterplots, and<br />
multivariate displays were invented, long after such triumphs of<br />
mathematical ingcnuity as logarithns, Cartesian coordinates, rLe<br />
calculus, and the basics ofprobabiLity theory. <strong>The</strong> remarkable<br />
William Playfair (1759 t84) developed or improved upon nearly<br />
:ll thc fundamental graphical designs, seeJ
<strong>The</strong> second part ofthe book provides a language for discussing<br />
graphics and a practical theory ofdata graphics. APplying to most<br />
visual displays of quantitative bformation, the theory l€ads to<br />
changc and improvements in design, suggests why some graphics<br />
might be better than othen, and generates new typer of graphics.<br />
TLe emphasis is on maximizing principles, empirical measures of<br />
graphical performance, and thc sequential improvement ofgraphics<br />
through revision and editing. Insights into graPhical design arc to<br />
be gained, I believe, &ord theories of whar males for excellencc<br />
in art, architecture, and prose.<br />
This is a book about the design of statistical graphics and, as such,<br />
it is concemed both with design and wirh statistics. But it is also<br />
about how to communicate infor:mation through the simultaneous<br />
presentation ofwords, numbers, and pictures. <strong>The</strong> design of statistical<br />
graphics is a universal rnatter-like mathemrticr-and is not<br />
tied to the unique Gatwcs of a particular language. <strong>The</strong> descriptive<br />
concep6 (a vocabulary for graphic$ and the principles advanced<br />
apply to most designs. I have at times provided evidence about fie<br />
scopc of these ideas, by showing how fiequently a pnnciple applics<br />
ro (a random sample of) news and scientfic graphics.<br />
Each year, the world over, somewhere between 9oo billion<br />
(9xro11) and u trillion (zlro12) images ofstatistical graphics arc<br />
printed. <strong>The</strong> principles of this booL apPly to most oftlose graphics.<br />
Some of thc suggested changes are small, but others are substantial,<br />
with consequcnces for hundreds of billions of printed peges<br />
But I hope also that the book has conseque.nces for the viewen and<br />
nukers of thot images-char they will never vies or creore 'tativ<br />
tical graphics the same way again. That is in patt because we are<br />
about to see, coliected here, so many wonderirl drawings, drose<br />
ofPlayfair, of Minard, of Marey, and, nowa&ys, of the computer.<br />
Most of all, then, this book is a celebration of data graphics.
Graphical Excellence<br />
Exccllence in statisrical graphics consists of complex idex<br />
communicated with clarity, prccision, and cficiency. Graphical<br />
displays should<br />
. show the data<br />
. induce the viewer to think about the substance rather than about<br />
methodology, graphic design, the tectrnology of graphic production,<br />
or something else<br />
. avoid distorting what the dara have to say<br />
, present many nunrben in a small space<br />
. make large data sets coherent<br />
. cncourage the eye to compare di{Ierent pieces ofdata<br />
. reveal thc data at several levels ofdetail, from a broad ovcrview<br />
to the 6ne structure<br />
. scrve a reasonably clear purpose: dcscription, exploration,<br />
tabulation, or decoration<br />
. be closeiy integrated with the statistical and verbal descriprions<br />
ofa data set.<br />
Graphics /eredl data. Indeed graphics can be more precise and<br />
revealing than conventional statistical cornputations. Consider<br />
Anscombe s quanet: all four ofthese data sets are described by<br />
exacdy the same linear model (at least until the residuals are exanined).<br />
10.0<br />
8.0<br />
13,0<br />
9.0<br />
11.0<br />
:14-0<br />
6_0<br />
12.0<br />
7.0<br />
5.0<br />
8_04 10,0<br />
6.95 8_0<br />
7.58 13_0<br />
8.81 9.0<br />
8.33 11,0<br />
9.96 14.0<br />
7,24 6.0<br />
4. 4.0<br />
'\0.84 72.0<br />
4.42 7,0<br />
5.68 5_0<br />
9.L4<br />
8.14<br />
4.74<br />
4.77<br />
9.26<br />
8.10<br />
6.13<br />
3.10<br />
9.13<br />
7.26<br />
10.0<br />
8.0<br />
13.0<br />
9.0<br />
11.0<br />
14.0<br />
6_0<br />
12.a<br />
7.0<br />
5.0<br />
7.46<br />
72.74<br />
7.L1<br />
7.41<br />
8.84<br />
6.08<br />
5.39<br />
8.15<br />
6.42<br />
5.73<br />
8.0<br />
8.0<br />
8.0<br />
8.0<br />
8.0<br />
6.58<br />
5.76<br />
7.71<br />
8.84<br />
4.47<br />
8.0 5,25<br />
19.0 12.50<br />
8.0 5.56<br />
8.0 7.91<br />
8_0 6.89<br />
N:11<br />
cquation of.egresioD line: Y : 3 + o5X<br />
standrd eno. ofestimte ofslolc : 0.118<br />
sun ofsquues X t:110.0<br />
rcgression suh of squa.es : 27.50<br />
rsidual sum of sq@res ofY : 13.7s<br />
corelatioD coediciot : .82
14 GRA?I]ICAL ?RACTICE<br />
Ard yet how they differ, as the grapLical display of the data<br />
makes vividly clear:<br />
Alrd likewise a graphic easily reveals point A, a wil&hot observation<br />
that will dominate standard statistical calculatiom. Note that<br />
point A hides in the marginal distribution but shows up x clearly<br />
exceptional in the bivariare scatter.<br />
F. J. Ansconbe, "cEpns h Statistical<br />
Aa lys," Amti.on Statittkiaf, z7<br />
Februrt r9j3),17 2t.<br />
stethe S. Brid md Stelhd !. Fidbss,<br />
"Re@t<br />
lconome.ric Modeling<br />
of Crime md P&isbDenr Suppoit for<br />
the Deten@@ Hypoth6is:" iD sElhd<br />
l. Fienbdg dd AlberJ. Reil3, Jr., cds.,<br />
btdkataB oJ Cnac afld Cnnivl JrniE:<br />
Qrd/titatilc st di6 (washinston, D.C.,<br />
1e8oJ, p. 89.
Of course, statistical graphics, just likc statistical calculations, are<br />
only as good as whar goes into them. An ill+pecifred or preposterous<br />
model or a puny dara scr cannot be rescued by a graphic<br />
(or by calculation), no matter how clever or fancy. A silly theory<br />
rneans a silly graphic:<br />
: -.004<br />
a<br />
€ -,*<br />
rlry au& sept, ocl Nd.<br />
so]A RrDAroN ND S]1h pRrG<br />
r1. New Yorl stftk pde (Bamt average). B. Sold R ctiarion, invqted,<br />
c- Lon.lon 3tocl pde, all by oonth!, 1929 (afrd cdia,Mara and<br />
u5 turn to the plactice ofgraphical cxcellence, rhe eficient<br />
mication of complex quantitative ideas. Excellcnce, nearly<br />
ofa muldvariare.orr. i" illLrsrrar.d herc lor frrndamenral<br />
and relational graphics. Thcse examples serve several<br />
providing a set ofhigh-qualiry graphics that can be<br />
d (and sometimes even redrawn) in constructing a theory<br />
graphics, helping to demorxtrate a descriptive terminology,<br />
eliry in briefabour the history ofgraphical developmenr.<br />
ofall, we will be able ro.eejuv hon good sratirtical<br />
b can be.<br />
CRA?HICALIXCXIIENCX 1J<br />
Edwlid R. Dewey and ndwin F. Dakin,<br />
cr.les: <strong>The</strong> s.ierc ofPrcnnian (New
16 GRA?HIcAT ?RACTICT<br />
Data Maps<br />
<strong>The</strong>se six maps report the age-adjusted dcath rate from various<br />
typcs ofcancer for the 3,o56 couaties ofthe United States. Each<br />
map portrays some 21,ooo numbers.l Only a picture can carry such<br />
a volume of data in such a small spacc. Furthermore, all that data,<br />
dranLs to the graphic, can be thought about in many different<br />
ways ar rnany dili'erent levels ofanalysis-ranging from the contemplation<br />
of general overall parterns to thc dctcction ofvery<br />
ine county-by-county detail. To takejust a Gw examples, look<br />
at the<br />
. high death rates 6om catcer in the northeast part ofdle country<br />
and around the Great Lakes<br />
, low rates in an east-west band across the rniddle ofthe courtry<br />
' higher rates for men than for women in the soutL, padicularly<br />
Louisiana (cancers probably causcd by occupatioml exposure,<br />
lrorn working with asbestos in shipyard$<br />
. unusual hot spots, including northern Min:resota and a few<br />
counties in Iowa and NebrasLr along the Mnsouri River<br />
. difi-erences in types ofcancer by region (for example, the lrigh<br />
rates of sromach cancer in the nonh-central part of the country<br />
probably the result of dre consumption of smoked Iish by<br />
Scandinavians)<br />
. rates in areas where you have lived.<br />
nre rnaps provide many lcads into drc causes-and avoidance<br />
of cancer. For example, the aurhors report:<br />
ln cerrain situations . . . the unusual orperience ofa county<br />
warrants fudrer investigation. For example, Salem County,<br />
NewJeney, leads the nation in bladdcr canccr mortality<br />
among whitc men. Wc anribute this exccs risk to occupational<br />
exposurcs, sincc about :5 percent ofthe employed persons in<br />
dns counry worL in thc chemical industry, particularly the<br />
manufacruring oforganic chemicals, wlnch rnay cause bladder<br />
tumors. After rhe frrding was communicated to NewJeney<br />
health oficials, a company in the area reported that at least 33o<br />
workers in a single plant had developed bladder canccr during<br />
the last 5o years. It is urgeDt that surveys ofcancer risk and<br />
programs in cancer control be initiated among worken and<br />
former worLers in this area.2<br />
ln,ch couty\ rate n loc.ted in tw.<br />
dnnensions dd, firrthei, at least fo6<br />
nuDbeB would be nece$r) ni re.onsdc<br />
he size ald sbape of.r.h cohty.<br />
This yields 7xl,016 entrics in. dru<br />
nra$ia sufi.i@t to rcproduce a nDp.<br />
Signi6candy hieh, but<br />
In highst de.ile, bui not<br />
Not signifE.ndy diferot<br />
Signiicanrly lower rhan<br />
,Robert Hoove., ThonlasJ. Manr,<br />
Innk w. McKay, od Joscph F. Fa!<br />
mmi,Jr., cncd br coDry: Ncw<br />
Resource for ntiologic Ches," S.ie,.o,<br />
189 (September 19, r97j), 1006.<br />
M^ps kom At16 oJ CM@t Moftalitr Jal<br />
U.s- Co!,r'6i r9Jei969, by Thonns J.<br />
MasoD, Ihnk W. Mchy, Roberr<br />
Hoover, wiluan J. Blot, dd Joeph F.<br />
Fnumdi,Jr. (wrshingron, D.c.: rublic<br />
Hmli\ Senice, Nadonal lNitutes of<br />
Hslth, ieTt. <strong>The</strong> six haps shoM hdc<br />
wete rcde$gled and ieilawn by<br />
Lawience Fahey and ldwdd Tufte.<br />
I<br />
I<br />
ffi<br />
n<br />
E
All rypes ofancer, white Gmales;<br />
age+djusted rate 6y county, 1g5o 1969<br />
AII types ofcancer, white males;<br />
age-rdjusted<br />
rate by coun.y, rssc-,+6s
Trachea, bronchus, and lung caacer;<br />
white femala; ageadjuted rate<br />
by county, 19Jo 1969<br />
Trachea, bronchuq aad lung cancer;<br />
white nales; age-adjusted rate<br />
by county, 19jo-1969
Stomach canccr, white f!ma1er;<br />
agc-adjustcd rate by county, 19jo 1969<br />
Sromach cancer, I'hite males;<br />
age-adjusted rate by counrl, 19Jo-1969
20 GR,{IEICAL IRACTICE<br />
<strong>The</strong> maps repay careful study. Notice how qui&ly and naturally<br />
our attention has been directed toward exploring the substantive<br />
content of the data rather than toward questions ofmethodology<br />
and technique. Nonerheles thc maps do have then flaws. <strong>The</strong>y<br />
wrongly equate the visual importance of each county wirh irs<br />
geographic area rather than with rhe number ofpeople living in<br />
the counry (or the number ofcancer deaths). Our visual impression<br />
of the data is entengled wirh rhe circumsrance of geographic<br />
boundaries, shapes, and areas the chronic problem aflicting shadedin-area<br />
designs ofsuch "blot rnaps" or "patch maps."<br />
A fiuther shortcoming, a defect of data rather than graphical<br />
composition, is that the maps are founded on a suspect aleta source,<br />
dearh cerrificate repom on the cause of death. <strong>The</strong>se reports fa1l<br />
under the influence ofdiagnostic fashions prevailing among doctors<br />
and coroners in particular places and times, a trcublesome<br />
adulterant of the evidence purporting to describe the already sometimes<br />
ambiguous mstter of the exact bodily site of the primary<br />
cancer. Thus part ofthe regional clustering seen on the maps, as<br />
well as some of the hot spots, may refect varying diagnostic<br />
cu.tom. and fads along wirh the acrual differ.nce' in canrer rare\<br />
Data maps have a curious history. It was noi until the seventeenth<br />
century that the combination ofcatogmphic and statistical skills<br />
required to consrruct the data map cane together, fully 5,ooo years<br />
after the 6nt geographic naps were dram on clay tablets. And<br />
many highly sophisticated geographic maps were produced centuries<br />
before the frrst map contining any statistical mateial was<br />
drawn.3 For example, a detailed map with a full grid was engraved<br />
during the cleventh century A.D. in China. <strong>The</strong> Yn Chi Thu (Map<br />
ofthe Tracks of Yii the creat) shown here is described by Joseph<br />
Needham as the<br />
. . . most remarkable cartographic work ofits age in any<br />
culture, carved in stone in +rr37 but probably dating Gom<br />
before +rroo. <strong>The</strong> scale ofthe grid is 1oo l, to the division.<br />
<strong>The</strong> coastal outline is relatively lirm and the precision ofthe<br />
network ofriversystems extraordinary. <strong>The</strong> size ofthe oiginal,<br />
which is now in the Pei Lin Museum et Sian, is about 3 feet<br />
square. <strong>The</strong> name ofthe geognpheris not known. . . . Anyone<br />
who compares this map with the contemponry productions<br />
of turopern relgious co.mogrrphy crnnor bur be Jmr/ed ar<br />
the evrenr ro wluch Cline'e geography wrs at th:r ume a\e:d<br />
of rhe West. . . . <strong>The</strong>re was nothing Like it in Europe till the<br />
Escorial MS. map ofabout +1JJo. . . ,4<br />
3Dab m?s are $uatly de$ribed as<br />
'lh€miic m.ps" in c,nogeph)'. For a<br />
thorough ,ccomr, sa Aitlff H. Robnsan,<br />
Eatl! <strong>The</strong>natn Mappikg in tlu<br />
Hitory a.f Cd asapb (Chi.aso,1982).<br />
On the history ofstatistical s.aphics, se<br />
H. cray FunLhotrsr. Histo.ical Develolmentof<br />
thecralhi€lRcpresdtation<br />
ofshtistical D,u," oJtn, 3 Novehbe(<br />
1917), 269-4o4i fd James R. Beniger<br />
md Dorothy r. Robyn, "Qunritarive<br />
Graphic in statislio: A BricfHistory,"<br />
Andnan Statituian, 32 (Fetl@ry s73),<br />
.Josph Needhan, S.ie,a zd ci,t,iJd-<br />
I'or l, Crird (Cantridge, 19i9), vol. 3,
cat4<br />
iill<br />
cRA?ItIC,{l EXCEf f!NCt 21<br />
I. Chavdnes, L€s Deu Plus Anciens<br />
Spdcimms de la Catogralhic Chinoise,"<br />
Brlletin de l *ole hon@ e de I'Exttne<br />
ot?tr,,3 Geoj). 1 lJ, Carr€ B.
22 GRATHlCA! PRACTICI<br />
Z"l<br />
e g<br />
t&D Ecce fornr ulam,vfum,arq ue<br />
fi rLrClunm T.rbularum Ptolonrri,cum quibufdam locis,r'n<br />
quibus ftudiofus Geographic fc fatis excrcetc poreft.<br />
L$'dt<br />
SEPTENTRIO.<br />
N,yy,h<br />
6"U<br />
7'flnt<br />
3'<br />
( It<br />
Oft4S tl'..<br />
'1 ERID IE8.<br />
<strong>The</strong> rs+6 cdition of Co snograph;a by Pettrs Apianus contained<br />
examples ofmap design that show how very close EurcPean cartography<br />
by that time had come to achieving statistical graphicacy,<br />
even approaching the bivariate scatterplot. But, according to the<br />
historical record, no one had yet made the quantitative abstraction<br />
ofplacing r measured quantity on the map\ suface at the intersection<br />
ofthe two threads instead of the name ofa city, let alone<br />
rhe more diffrculr abstracdon ofrephcJng ladtude and longicude<br />
with some other dimensions, such as time and money. Indeed, it<br />
was Dot until 1786 that dre frrst cconomic time-series was ploned.<br />
0<br />
Ed
One of the fint data maps was Edmond Flalley's 16g6 chan<br />
showing trade winds and monsoons on a world map.s <strong>The</strong> detailed<br />
section below shows the cartographic symbolization; with, as<br />
Halley wrote, ". . . the sharp end ofeach litde snoaL pointing out<br />
that part ofthe Horizon, from whence the wind continuaily comesi<br />
and where Llere are Monsoons the rowr of.rroakr run alternarely<br />
backwa.ds and forwards. by wbich means rhey .rre rhicker ldenrerl<br />
than elsewhere."<br />
; ",fr a)\ TiI : $.P A<br />
-A'T'1 a-<br />
e-'.-r\-<br />
- c ,E'a!tr -' '<br />
a<br />
"jit:'" i to<br />
,.2-.>-":..<br />
't-/ a,<br />
':t:;t'{<br />
-/ /- .a<br />
-7<br />
r,cf
24 GRA?HrCAt ?RACTICt<br />
An early and most worthy use ofa map to chart patterns of<br />
disease was thc famous dot map ofDr. John Snow, who plotted<br />
the location ofdeaths from cholera in cmtral London for September<br />
r8J4. Deaths were matked by dots and, in addition, the<br />
area's eieven water pumps were located by crosses. Examining the<br />
scatter over rhe suface ofthe map, Snow observed that cholera<br />
occurred almost entirely among those who lived near (and drank<br />
from) the Broad Street water pump. He had the handle ofthe<br />
contaminated pump rcmoved, ending the neighborhood epidemic<br />
which had taken more than 5oo lives.6 <strong>The</strong> pump is located at the<br />
center of the map, just to the right of the o in rRoAD srrxrr. Of<br />
course the link between the pump and the disease might have beea<br />
reveeled by computation and analysis without graphics, with some<br />
good luck and hard work. But, herc at leest, graphical ana\sis<br />
testifies about the date far more efiiciendy than calculation.<br />
Charles Joseph Minard gave quantity as well as directioi,r to the<br />
data measures located on the world map in his pomayal ofthe<br />
1864 exports ofFrench wine:<br />
6 l. w. cilterr,<br />
"Pioneer<br />
Maps of Health<br />
dd Disede in Eng1and," C@j'd?ltai<br />
Jd,bdl, D4G9j8),17>183. Shom h(c<br />
is a re&.wiDg ofjolD Snow's map. For<br />
a r.prodetion and &tailcd dalysis oftlc<br />
originrl na!, seendwrd Tufte, r,,tal<br />
Ertlandtio$: Lnass aad Qrtunn\, E iden@<br />
fll Nd'drte (ch€sbne, com$icut,<br />
199?), chapr€r 2. Idealy, seJobn Snow,<br />
On the Mode oJ coM"niatiot o.f chaled
a<br />
rl<br />
ll<br />
tl<br />
l:<br />
i:! ;<br />
::. ;<br />
i ii :<br />
ji. i<br />
::: :<br />
;j. ;<br />
ii<br />
a:<br />
tii.:i<br />
i:r': t; :i:- !<br />
ti:!: Ft:<br />
iii;:<br />
i;::: t:i::<br />
i:":l<br />
j:::...<br />
"r^:n",,'<br />
'.rr,<br />
n-<br />
'/'r<br />
1.<br />
(.-<br />
!<br />
: l<br />
''1J 1,<br />
: : ! a<br />
i --< ;5.i<br />
i:=: !<br />
-i ;: ;
26 GRAPIIICAL ?RACTICT<br />
Computerized cartography and nodern photographic rechniques<br />
have increased thc dcnsity of informrtion some 5,ooo-fold in dre<br />
bcst ofcurrcnt data maps compared to Halley's pioneering cffort.<br />
This map shows the distribution ofr.3 million galaxies (including<br />
some overlaps) in the northern galactic hcmisphcre- <strong>The</strong> map<br />
dividcs the sky into t,oz4xz,zzz rcctangles. <strong>The</strong> number ofgalaxies<br />
counted in each of the 2,273,328 rectangles is represented by<br />
ten gray tones; the darker the rone, dre greater the number of<br />
galaxies counted. <strong>The</strong> north galactic pole is rt the center. <strong>The</strong><br />
sharp edge on thc left results from the earth blocking the view<br />
from the observatory. In the area near the perimeter of the map,<br />
the view is obscured by the interstellar dust of the gala-xy nr which<br />
we live (the Milky Way) as the line of sight passcs througL rhe<br />
flattened disk ofour gala:.y. <strong>The</strong> curious texture oflocal clusters<br />
ofg.laxe' seen in du. trul) ner vieq ofuhc univcrse wa. not<br />
anticipat€d by srudents ofgalaxies, rvho had, ofcourse, microscopically<br />
examined millions of photographs of galaxies before<br />
secing this macroscopic view. Although the clLrsters are clearly<br />
evidcnt (and accountcd for by a theory of galactic origin$, the<br />
seemingly random filamcnts may be happenstance. <strong>The</strong> producers<br />
of the rnap note the "stmng temptation to conclude that the galaxies<br />
are arranged in a rem:rkable {ilamentary pattem on scales<br />
ofapproxinmtely J' to 1J", but we caution drat this vnual impression<br />
may bc misleading beceu$ the eye tends to pick out linear<br />
parterns even in random noisc. Indecd, roughly sirnilar pttems<br />
rre seen on maps constructed ftom simulated catalogs whe'e no<br />
linerr structure has been built in. . . ."7<br />
<strong>The</strong> most extensive data maps, such as the cancer arlas and the<br />
count ofdrc galaxies, place millions ofbits ofinformation on a<br />
single page before our eyes. No other mcthod for the displiy of<br />
statistical information is so powerful.<br />
7 Michael Seldner,B.H. Sieles, ld$dd<br />
J. Grolh and P. Jancs E. Pccblcs,<br />
"New<br />
Reduction ofthe Lick Catalos of<br />
Ga]€iei Asttatunidl Jawidl, 82 C\ptil<br />
1977), 249-j r 4- See Gillian R. (napp,<br />
''Mining<br />
the Heavens: Thc SloanDigirrl<br />
Sky SuNey," Sll & T?leJ.or (Aqust<br />
ieeT), ao-a8i MarsaretJ. Geller andJohn<br />
P. Huchra, "Mapping the UniveBe.<br />
s&t A Tels.o?e (Ausust 199r, 1j1119.
clA?Htcal IxclrrIN c! 27
28 GRA?HIcAL ?RAcTIcT<br />
Tir[e-Series<br />
<strong>The</strong> time+eries plot is the most frequently used form ofgraphic<br />
design.s with one dimension marching along to the regular rhythm<br />
ofseconds, minutes, hours, days, wceks, nonrhs, ycars, centuries, or<br />
millennia, thc natural ordering ofthe tine scalc gives this dcsign a<br />
strcrgth and efiiciency of interpretation found in no other graphic<br />
arrangemeD!.<br />
This rcputcd tenth- (or posibly eleventh-) century illustration<br />
ofthe inclinations ofrhc planctary orbits as a function ofdme,<br />
apparentll parr of r text fbr monastery schools. is the oldesr known<br />
examplc ofan attcrnpt to show changing values griphically. It<br />
:rppears as a rnysterious and isolated wonder in the history ofdata<br />
g,ap\'' .. .rrcF rLe nexr e',r"nr graphc of " plotted Limc-crie.<br />
shows up some 8oo years later. According to Fulkhouser, the<br />
astronomcal contenr is confused and there are difricultics in rcconciling<br />
the graph and its accompanying texr with the actual movcments<br />
ofthe planets. Parricularly disconcerting n rhe wavy path<br />
ascribcd to thc sun.'An crasure and correction ofa curve occur<br />
..--. ,L^ -i,tJl^ ^f,t- ^--^l<br />
.ri4\t<br />
l \'PF<br />
!A random sam!]e of4,ooo g.rphcs<br />
dnm fron rj of tle sorld\ news<br />
!a!e6 d naBazins lublished fronr<br />
1974 Rr 1930 fou.d that more than 7t<br />
pcrccnt of all thc graphcs ptrblished<br />
wcrc tinc-serlca, Chaptcr 3 reporsnore<br />
,H. Gr,t lmkhouser, A Nore on a<br />
Tcnth Cedury Graph," Orin!, I<br />
(Ianudr 1916), 26e:62.
It was not unril rhe late 17oos thar dme-series charts began to<br />
appear in scienrific wrirings. Thrs drawing ofJohann Hcinrich<br />
Lambert, one ofa iong scries, shows the periodic variation in soil<br />
tempcrature in relation ro the dcprh undcr rhe surface. <strong>The</strong> greater<br />
the deprh, the greater the rime-1ag in temperature responsivencs.<br />
Modern graphc dcsigns showing time-series periodicities diFer<br />
little from those oflambert, although the data bases are far larger.<br />
a<br />
This plot of radio emissions from Jupiter is based on data collected<br />
by Voyagcr 2 in its pass ciose by the planct in July 1979. <strong>The</strong> rrdio<br />
inrensir;' increases and decrcases in a ren-hour cycle as Jupirer<br />
rolates. Maximum intensity occurs when rLeJovian north magneric<br />
polc is tipped roward fie spacecraft, indicating a norrhe<br />
hemisphere source. A sourhern source was dctected on July 7, as<br />
dre spacecraft neared the equatorial plane. <strong>The</strong> horizontal scale<br />
shows the disunce of the spacecraft from the planet measured in<br />
terms ofJupiter radii (R). Note the use ofdurl labels on the horizontal<br />
to indicatc both the d:te and disr:nce fromJupirer. <strong>The</strong><br />
entire bottom panel also serves to label the horizontal scale,<br />
describing the changing orientation ofthe spacecrafr relative to<br />
Jupiter as the planet is approached. <strong>The</strong> multiple time--serics<br />
erforce not only comparisons witLin each series over rime (as do<br />
all time-series plots) bul also comparisons between the tluee<br />
diferent sampled radio bands shown. Thn ricHy multivariare<br />
display is based on 453,6oo instrumenr samples ofcight bits cach.<br />
<strong>The</strong> resulting 3.6 million bits were reduced by pcak and average<br />
p(ocessing ro rhe 18,9oo points acually plotted on thc graphic.<br />
GRA?IIICAL IXCXTl!NCT 29<br />
J. H. I-anberr, 4'o'err -l3ertin, r7?9).<br />
D. A. clrneti, W. S. (u.th, and F. L.<br />
sc2ri "Plasnx wave obs.rvarioDs Ncar<br />
Jupite' Ihirial Rcsulrs f.om voy4er<br />
2, s.tdra:06 (Novchb{ 2t, 19?e),<br />
987 991; md Ietter from Dorold A.<br />
Guhett to Edward R. Tufre, Jue 27,
30 GRA?HrCAl ?RACTTCI<br />
Time+eries displays are at their best fbr big data sets with real<br />
variability. Why waste the power of data graphics ol simple linear<br />
changes,<br />
which can usually be better summarized in one or two numbers?<br />
Instead, graphics should be reserwed for the richer, more complex,<br />
more difiicult statistical material. This New York City wearher<br />
summary for r98o depicts 1,888 numbers. <strong>The</strong> daily high and<br />
low temperatures are shown i:r rclation to the long-run average.<br />
Thc path of tle normal temperatures also providcs a forccast of<br />
expccted change over the year; in thc rniddlc ofFcbruary, for<br />
instancc, New YorL City residents can look forward to warming<br />
ar the rete ofabour 1.J degrees pcr wcek all the way to Ju1y,<br />
the yearly peak. This distinguished graphic succesfully organizes<br />
a large collection ofnumbers, makcs compansons between di{Ierent<br />
parts ofthe data, and te11s a story.<br />
NEWYORK CITY'S WEATHER FOR 19AO<br />
Nd Yd! TiuelJanury 1r,1e&,p.32.
A design with similar strengdrs is Mareyt graphical train schedule<br />
for Paris to Lyon in rhe r88os. Arrivah aad deparnles from a<br />
sbtion are locared along the horizontal; length ofstop at a station<br />
is indicated by the length of the horizontal line. <strong>The</strong> stations are<br />
separated in proportion to their actual distance apart. <strong>The</strong> slope<br />
of$e line reffeca the speed ofthe trein: the moie nearly vertical<br />
the line, the faster the nain. <strong>The</strong> intersection of two lines locates<br />
the_time and place that trains going in opposite directions pass<br />
In 1981 a new €xpress nain from Paris to Lyon cut rhe rrip ro<br />
under chree hours. compared to more than njne hours when Marel<br />
pubJrshed tfie graphical train scheduJe. <strong>The</strong> path ofthe modem<br />
'tc\ (taifl d yafl.le uitesse) is sho*,n, overlaid on the schedule of<br />
1oo years b€fore:<br />
GRA?rlIC]lT EXCET,LENCI 31<br />
E. J. MNy, Ir n.thod. yathique P'n'<br />
r88r, p.20. Th. ne!}od is nkibured<br />
to the French eDgineer, !bry.
32 GnA?ItrCAL PR1\CITCE<br />
<strong>The</strong> two great inventors of modern graphical designs were J. H.<br />
L,zllJbert (1728-17j7), ^ Swiss-German scientisr and mathematician,<br />
and William Playfan (1759-1823), a Scottish political economisi.lo<br />
<strong>The</strong> fint known time-series using economic data was published in<br />
Pfayfairt remarkable book, <strong>The</strong> Connercial and Political Atlas (Inndon,<br />
1786). Note the graphical arithmetic, which shows the shifting<br />
balance of trade by the difference between the import and<br />
export time-series. Playfair contrasted his new graphical method<br />
with the tabular presentarion of dxta:<br />
Information, that is imperfectly acqoired, is generally as imperfectly<br />
rerained; and a fiun who has carefirlly investigated a<br />
printed table, finds, when done, that he has only a very faint<br />
and panial idea ofwhat he has read; aad that like a 6gure<br />
imprinted on sand, is soon totally eresed and defaced. <strong>The</strong><br />
amount ofmercantile transactio;s in money, and ofprofit or<br />
los, are capable of being as easily represenied in drawing. as<br />
any part ofspace, or as the face ofa country; though, till now,<br />
it has not been anempted. Upon that principle these Charts<br />
were made; and, while they give a sinple and distinct idea,<br />
rhey are as near perfrct accurery a< is :ny wey usefuJ. On<br />
rnspecong eny one ofthese Charts rcentively, a suftciendy<br />
dntinct impresion will be made, to remain unimpaired for a<br />
considerable time, and the idea which does remain will be<br />
simple and complete, at once including the duration and the<br />
amount. LPages 3 4j<br />
For Playfair, graphics were preferable to tables because graphics<br />
showed the shape ofthe data in a comperative persp€ctive. Time-<br />
IEIXT .'..2 tt. I P2AT'S ad EtPoRI'a b *z t" EnGaAnD<br />
7..m dz te //a? b n2 4 trlrqlai<br />
7L tt;tu,t.t t a'4"./,it fDlnt, N. t'br h 4-?'c^t".,4tuont,/?.t1rrt<br />
w.Lke'..Jry:4d<br />
-<br />
t?<br />
!ol:e TillinC, "E:!ly Expcrihstrl<br />
Gapht; B.itith Jordal Iot th. Hi'totr<br />
oIs.itu ,8 l19js), r93-2ra.
series plots did dis, and all but one of the 44 char* in the first<br />
ediaon of <strong>The</strong> Connereial atd Politiul Atla: were rime-series. <strong>The</strong>t<br />
one er.ception is the fiIst Lnown bar chart, which playfair invented<br />
beouse year-ro-1ear dara were missing ard he needed a design ro<br />
ponray tbe one-year dara rhrr w€rc avarlrble. Nonerhele.s hi was<br />
skeptical about his innovation:<br />
This Chan is different Gom the others in principle, as it does<br />
not comprehend any portion of cine. ard it is much bferior<br />
inrriLiry ro rhose thrr do: for rhough ir gire, the ex-renr ofrhe<br />
dill-erent braaches oftrade, it does not cJmpare the same<br />
branch ofcommerce with irselfat di{fercnt periods: nor do€s<br />
ir imprinr upon the mbd thar distircr idea,ln doine which.<br />
dre chiefadvantage ofcLarts consists: for as it wan-e the dimcnsion<br />
thar ir formed by durarion. rl'ere is no shape given<br />
to tnc qu.anflnes. lpage 1o1l<br />
He was right: small, noncomparative, higbly labeled data sets<br />
usualiy belong in tables.<br />
tu.r,i,....;?1e dz tu ?..,d,ahe 4.,, tu tL?. ttu 4.*d *.,ttae try*<br />
<strong>The</strong> chan does sloq, ar an) rare. Lhe imporn (cross-ha(c|ed<br />
) a:rd exports (solid lines) to and fiom Scodand in r78r for<br />
countriet, $.hich are ordered by volume of trade. <strong>The</strong> horizontal<br />
is at the top. posibll ro make ir more con!e en( (o see Ln<br />
ing the points by hand. Zero values are nice\ indieted both<br />
the absence ofa bar and by a "o." <strong>The</strong> horizontal scale misrepeats<br />
"2oo." In nearly all his chars, Pla/air placed the<br />
ls for the verical scale on the right side oftle page (suggesF<br />
tbat he plotted the &ta poin* using his left hand).<br />
Gla?ItIcal EXCEITENCE Jl
34 GRA?IlIC4L ?RACTTC!<br />
Playfair's last book addresed the question whether the price of<br />
wheat had increased relative to wages. tn his Iztter on ow agidlltural<br />
distrcsses, their eauses and renedies; auonpanied uith tables and<br />
copper-plate tharts shewing and coftFing the lices oI wheat, bread ad<br />
Iabow,ton 156j to 182r, Playfair wrote:<br />
You have before you, my Lordr and Gendemen, a chart of<br />
the prices ofwheat for zjo years, made from official retums;<br />
on ihe same plate I have traced a line representing, as nearly as<br />
I can, the wages of good mechanics, such as smiths, masons,<br />
and carpenters, in order to compare the proportion between<br />
them and the price ofwheat at every dif,erent period. . . . the<br />
main fact deserving ofconsideration is, that never at any former<br />
period was wheat so cheap, in prcportion to mechaaical labour,<br />
as it is at the present time. . . . [pager u9-3r]<br />
Here Playfair plotted three parallel time-*eries: prices, wages, and<br />
the reigns of Bitish kings and queens.<br />
<strong>The</strong> hirtory and genslogy ofroyalty was long a gr:phical frvorite.<br />
This superb construction of E. J. Marey brings together several<br />
sets of facts about English rulen inco a time-series that conveys a<br />
sense of the march ofhistory. Marcy (r83o-r9o4) also pioneered<br />
the development ofgraphical methods in human and animal<br />
physiology, including studies ofhorses moving ar different paces,
l<br />
(<br />
D p6ri( eP,<br />
GRA?IIICA! XXCtTTINCT 3J<br />
fie l<br />
E.I. Mrey, b Mdthode Qdphiqrc @^ni<br />
1385),<br />
!. 6,<br />
I.J. Mdey. Moy.Mr (todon, r89J).<br />
Segming wnh rhe @cks of tfie horse,<br />
the timHoies are from pages r91, 224,
36 GRA?HICAL ?RACTICE<br />
S*<br />
tl,e movemenr ofa starfrsh turning itselfover (read images fiom<br />
the bottom upwards),<br />
the undulations of thc dorsal Iin of a descendinq sea-horse,<br />
as wcll as the advance ofthe gccko.<br />
Marey's man in black velvct, photographed in stick-Iigure images,<br />
became thc time-series fbrerunner of Marcel Duchamp s Nlde<br />
Delcendiflg a Stihcnse.<br />
€_<br />
ro<br />
{
<strong>The</strong> problem with tirne-series is that the simple passage of time<br />
not a good cxplanatory variable: descriptive chronology is not<br />
expianation. <strong>The</strong>re arc occasional cxceptions, especially when<br />
is a clear mech:nism rh:t drives rhe Y-variable. This timedoes<br />
testiE rbout causality: the outgoing mail ofthe U.S.<br />
ofRepresentatives peaks every two years,jusr before the<br />
dav:<br />
grapbic is worth ar leasr 7oo wo'd:, the number used ir a<br />
report describing how incumbenr reprerenrariva exploit their<br />
mailing privileges to advance thei re-election campaiqns:<br />
TIITE<br />
YOIINff SflOIIIN<br />
::{-19iib..tg.9-!!, i! ry! "! !"tq<br />
--<br />
;s;:."ii;j-<br />
'"LH"l;"<br />
!rclon h: iis lta<br />
. ero"d n tr di k !$ 8,. roa<br />
!e h. bbb, ry o 4:.r b !i ui re<br />
GRAPTTTCAT EXCETLTNcx 37<br />
Ftulrlr tu! Er nm .Ef<br />
rq'q0rr. rw! bF:is-rio ir iodrddd dd j.. ilrJ -hi 5Hr6,<br />
!r-.ryqY. b- 13 blY --,1.-l!9,<br />
lron Gh: t+d hd, @iF &d a! id F p6d.4i;<br />
rls ss
105 r<br />
t00 I<br />
35r<br />
t 05t00..1<br />
s5.r<br />
38 GRA?IIICAT ?RAcTICE<br />
Time-series plots can be moved toward causal explanation by<br />
smuggling additional variables into the graphic design. For example,<br />
this decomposition ofeconomic data, a aying r,296 numbel\<br />
breaks out the top series into sexonal and trading-day fuctuations<br />
(which dominate short-terrn chrnged to reveal the long-run trend<br />
adjusted for inflation. (Note a signi{icant defect in the design,<br />
however: the vertical grid conceals the height ofthe December<br />
peaks.) <strong>The</strong> next step would be to bring in additional variables to<br />
explain the trarxformed rind improved series at the bo*om.11<br />
Srtendi. ond nEllb coryone s ofTohr Rdor so,er, untcdSb,es<br />
I<br />
----Y<br />
If tt fl<br />
l l<br />
t'<br />
1 t"f<br />
r<br />
111' 11"t' l'11 11 t' TT1'111111'TTl'<br />
,f<br />
^..-<br />
I<br />
]}"'ffi ,t,,t,,t ,t..t,,t. ,,t.,1.,1, ,t,,t.,t, t,,t.,t, ,t,,t,.t.<br />
ls tsr l97s l9l<br />
("<br />
f<br />
V<br />
/<br />
41<br />
40<br />
32<br />
21<br />
20<br />
1t05<br />
ll00<br />
-95<br />
I r20<br />
lt10<br />
Iro<br />
rs0<br />
,.ii"".<br />
'136<br />
lrz<br />
136<br />
l:r<br />
t36<br />
Jn<br />
!.ii..,<br />
12<br />
I<br />
128<br />
I ''121<br />
I<br />
lzo<br />
I See Willian s. clscland dd rrma J.<br />
Teipming,<br />
"Gralhical<br />
Methods foi<br />
Sssonal Adjstn@t," JoMzl d/tlo<br />
Aberi@ Shritriral ^btiatiaa 77 (M^rdl<br />
Julius slisLin, "M@suing cur@t E.G<br />
nomi. IiucturioDs," Sr,rirti.dl -Rd?o/ar<br />
0uly 1973), p.3.
Finally, a vivid design (with appropriate data) is the before-after<br />
gb.:a, "l..dt!o,l -?,.D!t; .isnd to.n: av\ q dttru^ont" ,a...d b\ a liq^d nt,^s.4<br />
tu44 .atit,in th. .tu\. fhp Jnap n n.ea.h au, tht"uEh ,r? .upa,r4u, r,p dc?aot<br />
.? td.cit:okd^. th. ruhp n th" to.D , ,ura.kadt i, iu! t," rt4h, n.Ehnui. b n. d<br />
rlnd before and after the collapse ofa bridge on the Rh6ne in r84o:<br />
ponl de Bou.g-SlAndtol sur- le Rh6ne.<br />
GnAPHTCAT<br />
!XCrrrtNcE 3g<br />
M. MitcicU Waldrop, "In Sedch of<br />
thc Maglciic Monopole,', S.i&.e<br />
oue a 1e82), p. ro87.<br />
Charls Joseph Minard, "De ta chure<br />
des PonB dus 1es grudes Crues,',<br />
(odober 2a, 1 3J6), Figue 3, in Mimrd,<br />
coll%tio' de k' bto.h"tet eans,1321-<br />
186e), held by thc Bibliorhlque dc<br />
I Ecol€ N,rionale dcaPotus cr Chau$ies.
40 GRAPIIICAL ?AACTICT<br />
Narrative Graphics of Space and Tihe<br />
Allr especially efective device for enhancing the explanatory power<br />
oftime-series displays is to add spatial dimensions to the desigr of<br />
the graphic, so thet the data are moving over space (in two or three<br />
&mensions) as well as over time. Thrce e.cellent space-time-story<br />
graphics illustiate here how multivaiate complexity can be subtly<br />
integrated into graphical architecture, irtegrated so gently and<br />
unobtrusivety that;et'ers are \arclly aware that they are looking<br />
into a world offour o1five dimensions. Occasionally graphics are<br />
Hligercntly multivariate, advertising the techdqne rather than<br />
the deta, 3ut not these three.<br />
<strong>The</strong> first is the classic of Cbarles loseph Minard (1?8r-i87o), the<br />
French engiaeer, which shows the tenible fate of Napoleon's army<br />
in Russia. Described by E. J- Marey as seeming to defi the pen of<br />
the historian by its brutal eloquence,l2 this conbination ofdata map<br />
and time-sedes, drawn in 1869, poftIa)'s a sequence ofdevastating losses<br />
suffered in Napoleont Russian campaign of 1812. Beginning at left<br />
on the Polish-Russian border near the Niemen tuver, th€ thicl tan<br />
flow-line shows the size of the Grand Army (4z2,ooo) as it invaded<br />
Russia in June 1812. <strong>The</strong> width ofthis band indicates the srze ofthe<br />
army at each place o11the map. In Septenber, the army reached<br />
Moscow, which was by then sacked and deserted, with roo,ooo<br />
men. <strong>The</strong> path ofNapoleon's retreat froni Moscow is delicted by<br />
the darker, lower band, which is linked to a temperatue scale and<br />
dates at the bottom ofthe chart- It was a bitterly cold winter, and<br />
many froze on the march out ofRussia. As the grapKc shows, the<br />
crossing ofthe Berezina River was a disaster; and the army finally<br />
struggled back into Poland with only 1 o,ooo men remaining. AIso<br />
shown are the movements ofauxiliary troops, as they sought to<br />
Fotect the rear and the fant ofthe advancing army. Minard's<br />
graphic tells a rich, coherent story with its multi.variate deta, far<br />
more enlightening tben just a single number bouncing along over<br />
time. Sirc variables are plotted: the size ofthe army, its locatiolr<br />
on a two-dimensional surface, direction ofthe army's movement,<br />
and temperature on vadous dates during the retreat from Moscow.<br />
At upper right we se€ Minard's French original, which was printed<br />
as a two-color lithograph in the form ofa small poster. And at<br />
lower rigbt, our Englirh translatioD.<br />
It may well be rbe besr strciscJcal graphic ever drawn.<br />
12 E.I. Mtey, ra nathnde gaphiqne<br />
(P!is, 188t, p.71. For more on Minard,<br />
see AJthu H. RobiNn, "<strong>The</strong> <strong>The</strong>matic<br />
I'{aps of Cbnl6Jseph Mimrd," Isdso<br />
Munii, ^ \1967), 95-ro8.<br />
Utper inage 6on chdl6Joeph MiD'd,<br />
Tobled* Gophiry\ et Con6 Fig*ati@s L<br />
M. Mitud, 1845 $6q,Blbhnthiete .le<br />
Pris, item 28 (62 by 2J cm, or 25 by 10 in).<br />
E glish t@slation by Dawn Finley ud<br />
redEving ty llaire Mosq cmpleted
.'--rr/i Jti4aizl:rt r", rN. ",,.,..1." rt.-.,- :. i Q. ,,.. . a,, ..". r.,,.t".:*f,.1. ,- . \,,..,i. rn|r rsr.,<br />
.r..,...,..r,.rri!,.!:.r.r,.-r...!.i....,i.{..,..,r'.,...,..,i.,,,,.,.<br />
.......,...,,i..-..,....i.......,i.....r.r,,.,-.,,,,....*:..r,"..,.,..,;".......,.,,,.,:.::,.1..t..:1..l-j.:.1:;'"<br />
,1<br />
.'<br />
I<br />
;tLJf,t ,'tili, iat f ., rri ?;t<br />
t,!i, a it ttr J N-at$.\t.*^ ,t an<br />
l:lLllJlL l.tatE "t :j,. nnpabe ?r rirnr o/ r/( Iii.unn. r;?q.d.r];.4,, ri<br />
I<br />
: .
42 CRA?IIICAI ?RACTICE<br />
Thc ncxt timc-space graphic, drawn by a computer, displays<br />
the levels ofthrcc air pollurants locaed over a twedimensionll<br />
surface (six countics ir southern Califomia) at four rimes during<br />
the day. Nitrogen oxides (top row) arc emitted by power plants,<br />
refneries, and vehicles. Refineries along the coast and Kaiscr Stceli<br />
Fonuna plant produce dre posr-midnight peaks shown in the first<br />
pancl; trafic and power plane (with rheir ireary daytime demand)<br />
send levels up during the day. Crrbon monoxide (second row) is<br />
1ow after nTidnight except out at the steel plant; morning tra{ic<br />
rhcn begins to generate each day's ocern ofcarbon monoxide,<br />
with the greatest concentrarion ar rhe convergence offive fieeways<br />
in downtown Los Angeles. Reactive hydrocarbons Ghird row),<br />
like nitrogcn oxides, comc from refineries after midnight and then<br />
increase with trafic during the day. Each of thc 12 time-spaccpollutant<br />
siices summarizes pollutanrs for z,4oo spatial locations<br />
(z,4oo squares five kilometers on a side). Thus z8,8oo pollurant<br />
readings are shown, except for those masked by peaks.<br />
Thc air pollution display is a snall n tiple. <strong>The</strong> same graphicrl<br />
design structure is repeated for each ofthe twelve slices or rnultiplcs.<br />
Small multiples are economical: once yiewers understand the<br />
dcsigl ofonc slice, they have immediate rcces to the dita in all<br />
the other slices. Thus, as dre eye moves ftom one slicc to the next,<br />
the constancy ofthe design allows the viewer to focus on chalrgcs<br />
in the data rather than on changes in graphical design.<br />
NITBOGEN OXIDES<br />
CAFBON MONOXIDE<br />
Las Anqels Tin$, lrly 22, 1979i b^ed<br />
on wor! of cregorr J. Mcxae, Califomia<br />
La*nuie of TechDologr.
Our third example of a space-time-srory grephic ingeniourly<br />
mixer space and time on the horizontal axis. This desien moves<br />
well beyond che conrcndonal rime-'eries becru.e ofiL" clever<br />
plotting field, with location relative to the eround surface on the<br />
verric:l axis and rime/spacc on rhe honzonial. Tle life cyde of<br />
the lapanese beetle is shown.<br />
More Abstract Designs: Relational GrapLics<br />
Ile invention ofdata graphics required replacing the laritudebngitude<br />
coordinates ofthe map wirh more abstrac measures rror<br />
f,ased on geographicai analogy. Moving from maps to statisticil<br />
pphics was a big step, and thousands ofyears pased bdore this<br />
rtep was taken by Lambert, Playfair, and others in the eishteenth<br />
. Even so, analogics to the physical world served as the<br />
ceptual basis for early time-series. Piayfair repeatedly compared<br />
charts to maps and, in the preface to the finr ed.Lt:Lon of <strong>The</strong><br />
nnercial a d Political Atlas, argued that his charrs corresponded<br />
a physical realization ofthe data:<br />
Suppose $e money we pay ir any<br />
_one rcar lo, 'he exp€nce<br />
of rhe Na!ry u_erc in gunca.. and rhat thesc guirea. wire lajd<br />
down upon,a lrrgc rable in a strarghr.line._and rouchirg each<br />
omer. and rnosf paid nexr )ear were latd down in anoiher<br />
GRAPHTCAI EXCETL!NCr 43<br />
L. Hugh Neman, Md, a/d hs.h<br />
(Lordon, 1e6J), p!. ro4-1oJ.
'-d. GRA?IIICAI PRACTICE<br />
srraieht line, rnd rhe same continued for a number ofyean:<br />
thesilines would be ofdifi'erent leneths, as there were fewer<br />
or more guineas; and they would make a shape, the dimensions<br />
ofwhich would agree exactly with the amount of the sums;<br />
and the value ofa guinea would be represented by the part of<br />
space which it covered. <strong>The</strong> Ch:rts are exactly this upon a<br />
small scale. and one division represents the breadth ot value of<br />
ten thousand or an hundred thousand guineas as marked, with<br />
the same exachess that a square inch upon a map may represent<br />
a square mile ofcountry. And they, therefore, arc a representation<br />
of the real money laid down in diFerent lines, as it was<br />
originally paid eway. lpages iii-ivl<br />
Fifteen years later in <strong>The</strong> Statistual Breviary,his most theoretical<br />
book about graphics, Playfeir broke free ofanalogies to the physical<br />
world and drew graphics as designs-in-themselves.<br />
One of four plates in <strong>The</strong> Statisdral Bteviary, rhis graphic is distingunhed<br />
by its multivariate data, the use ofarea to depict quantity,<br />
and the pie chart-in apparenrly the 6nt application of these<br />
devices. <strong>The</strong> circle represenrs dre area ofeach country; the line on<br />
the left, the population in rnillions read on the vertical scales; the<br />
line on the right, the revenue (taxes) collected in millions ofpounds<br />
sterling read also on the vertical scale; and the "doaed lines drawn<br />
between che population and revenue, are merely intended to connecr<br />
together the lines belonging to the same country. <strong>The</strong> ascent<br />
SIATISTICAI BREVIARY:<br />
B'FrLLl'aPL'fFlIL<br />
s-^/: /rr,"/,* npth " ".?-.,,--V* "-lt:*t 4;',",r.-"-.",' ,.,.,<br />
6<br />
xDx
ofthose Iines being ftom dght to left, or from left ro right, shews<br />
whether in Foportion to its population the country is burdened<br />
with heary ta-res or otherwise" (pages r3-r4). <strong>The</strong> slope ofthe<br />
dotted line is unbformative, since it is dependent on the diameter<br />
ofthe circle as wellas the heighr ofthe two v€rticals. Hov/evet,<br />
the sign ofthe slope does maLe sense. taking Plafan to his familiar<br />
point about whit he rcgarded as excessive taxation in Bdtain<br />
(sirtb circle from the ngbr. wich rhe dope runlrrjng opposire Lo<br />
mosr countriet. Playfair was enthusiastic about the multivariate<br />
arrangfment becaLise ir fos(ered comparisom:<br />
<strong>The</strong> autlor ofthis work applied the use of Lines to matters of<br />
commcrce and Erance abour filteen lears ago. -rrh grc,r<br />
succels. HJs mode wa" qenerallv approved ofas nor o'iiu fr.itir.rting,<br />
bur rcnderirg rlose srudie' rno'e cle,r, and rerjrned<br />
more e.srl) b) Lhe memo1. <strong>The</strong> pre'enr charL. rrc in like<br />
mamer inrerded ro aid rror'rjcal studies. bv shewine ro $e<br />
eye Lle 'izes ofdifferent courrrier rcpre.enrid bf imll,r form'.<br />
for where forms are not similar, the cye cannot iompare them<br />
easily nor accurately. From this circuirxtance it hapliens, thar<br />
we have a more accunte idea ofthe siza ofthe olaie*. whi.h<br />
are rphe,es. rlran of rhe narion. of Europe rn hich ". '* ."<br />
thc maps, all ofwhich are irregular forms in themselves as well<br />
as unlrke to each orher. Sizc. Populanon, and Revenue, are rhe<br />
rluee pnnopal objecrs ofarrenrion upon rbe gcncral scale of<br />
't.rti.rical srudie., wherhe, wc are acrurred bv"curiosirv or<br />
inrerev: I h:r c rherelore represcnred these cbrec objeci' in one<br />
r'lev/.... L?age 1Jl<br />
But here Playfair had a forerunner-and one who tJrought more<br />
clearly about the abstract probicms ofgraphical dcsign than did<br />
Playfair, who lacked mathematical skills. A most rem:rLable and<br />
explicit early theoretical statement advencing the gereral (nonanalogical)<br />
relational graphic was made byJ. H. I-ambert in 1765,<br />
3J y€ars before <strong>The</strong> Stutistical Breuiaryl<br />
We have in general two variable quantities, x, y, which will<br />
be collated with one another by observation, so rhat we can<br />
determine for each value ofx, which may be considered as an<br />
abscissa. rhe correspooding ordinare y. Were rhe expenments<br />
or observat;on\ comDletelv l(currle. the.e ordinateicould<br />
qire a number ofpoinrs throush *hich a straishr or curved<br />
Ine should be dr";n. B.,t ". dis s not 'o. theine deviarer ro<br />
a grcaler or lesse! exrent from the observational poinrs. ft must<br />
thercfore be drawn in such a way that it comes as near as<br />
possible ro is rrue posirJon dd goes. as ir u ere. tluough thc<br />
mrcldre ot the elven Dol]lts.',<br />
cRA?EaCAt TXCtLTENCE 4J<br />
!3johm Hetuich Lamhert, Bdlr,ile<br />
zen Cebtuucla da Mdthtuotik u"d d,th<br />
,r/dd!r8 @erlin, 1?6t, s quored iD<br />
Lam Til1iig. Earry nxlcjfteqral<br />
Gapb.," Btxbh Joanal ht the Histotr<br />
oJ S.iena,I (1s7l), 2o4-hs.
46 GRApETcAL ?RAcTrcr<br />
Lambert drew a graphical derivation ofthe evaporarion mte of<br />
water as a function oftemperature, according to Tilling. <strong>The</strong><br />
aralysis begim with two rime-seies: DrF, showing the decreasing<br />
height ofwater in a capillary tube as a fuaction of time, and alc,<br />
the temperature. <strong>The</strong> slope ofthe curve DrF is then aken (note the<br />
targent Dr c) at a number ofplaces, yielding the rate ofevaporation:<br />
To compiete the graphical calculus, the measured rate is plotted<br />
against the corresponding ternperarure in this relational graphic:<br />
Thus, by the early r8oos, graphical design was at last no longer<br />
dependent on direct analogy to the physical world-thanLs to the<br />
work ofl,ambert and Playfair. This meant, quite simply but quite<br />
profoundly, that any variable quantity could be placed in relationship<br />
to any other variable quantity, measured for the same<br />
units ofobseffation. Data graphics, because rhey were relational<br />
and not tied to geographic or time coordinates, became relevant<br />
J. H. &mbs!<br />
"Essi<br />
d'hygrom6rrie ou<br />
su li me@e de I'hlmidi4" Mlrotar<br />
de I'Aat#6ie Royle d?s S.ie&s d Bdl.r<br />
Lenes . . . i769 @*li^,177r), plare t<br />
facins !. 126; fron Tilingt artide.
to all quantitative inquiry. Indeed, in mod€rn scientific literature,<br />
about 4o percent ofpublished graphics have a reiational form,<br />
with two or more variables (none ofwhich are latitude, longitude,<br />
or time). This is no accidenr, since the relational eraphic in its<br />
barev lorm. (ne \crrrerplor and i15 \ari:nts is rle gieacesr of:ll<br />
graphical designs. It iinls at least rwo variablcs, ercouraging and<br />
even imploring the viewer to assess the possible causal relationship<br />
betwecn the plotted variables. [t confrons causal theories that X<br />
causes Y with empirical evidence as to the acrual relationslLip between<br />
X and Y, as in the case ofthe relationship berween hns<br />
cencer and smoking:<br />
CRUDE MALE DEATH RATE FOR LUNG CANCER<br />
IN T95O AND PER CAPITA CONSUMPTION OF<br />
CIGARETTES IN 1930 IN VARIOUS COUNTRIES,<br />
DEN'<br />
/j<br />
CELAND<br />
alFsweorr I<br />
./,<br />
ND<br />
7<br />
CI6ARETTE<br />
CONSUMPT]ON<br />
zt7,t2'a)72:r',4-.at:z,atzt,,/--<br />
Relorr of ihe Advis.ry Commirta to<br />
the Surgeon Generrl, S-oliry drd tledlr,t<br />
(War\iDgton, D.C., 1964), p. 176; based<br />
on R, Doll, "Etiotogy ofl-ug Cdcer,"<br />
Adudne' if Cdaat Raurh, 3 (rsi),
48 GRTTPEICAL ?RACTICB<br />
<strong>The</strong>sc small-mu1tiple relitional graphs show unemplolanent and<br />
inflarion over time in "Phillips curve" plo* for nine countries,<br />
demonstrating the collapse ofwhat was once thought to be an<br />
inverse relationship between the variables.<br />
ln8doo rd UMDlotnest Rds<br />
Unit€ds€t.3<br />
Patl, Mcct .ke, .t d., Towads FUI<br />
Enplornen ann ht@ Stabiw lPais,
<strong>The</strong>ory and measured observations diverge in the physical scialso.<br />
Here the relationship berween temperature and the<br />
conductivity ofcopper is asessed in a series ofmeasurefrom<br />
diferent laboratories. <strong>The</strong> connected points are from<br />
single publication, cited by an identfication number. <strong>The</strong> very<br />
answers reponed in the published literature result mainly<br />
impffities in the semples ofcopper. Note how effectively<br />
graphic organizes a vast amounr of alata, recording findings<br />
Lundreds ofstudies on a single page and, at the same time,<br />
ing comparisons of the varying results.<br />
GRAPlrrCAL rXCrr!xNCE 4g<br />
C. Y.Ho, R- \v. Poqell, mdP.n.riby,<br />
Thmal Condaxtiviry oJ th! Elenuk: A<br />
Coryeh^iue Reritu, snppletunt to.<br />
r, Joauat of Pfusnat ann ctukkdl<br />
Ref,reNe Data, 3 (914,1 2aa.
JO GRAPEICAL PRACTICI<br />
Finally, two rclational designs ofa diferent sort-.rherein the<br />
data points are thernselves &ta. Here the e{fect oftwo variables<br />
intencting is portrayed by the fues on the plotting ield:<br />
And similarll tle varying sizes ofwhite pine seedlings after<br />
growing for one seson in sand containing di1rerent amounts of<br />
calcium, in paro per million in nutri;ntland ctltures:..<br />
E- C. Z@a!"<br />
"Cat<br />
itlophe Th@ry."<br />
S.i6tijc Z6eti.M, 2a1" ( prll 1976), 67i<br />
baFd @ K(md Z. Lor@, Kir{<br />
Sot M'rairy Nry Yqk, 19Jr).<br />
H. L. Mitddl, me Gtovth ann Ntfiinn<br />
oJ Whit Pire Seedling if Crlxtcs wih<br />
Vdthg Nittog.r" WtI'hM, Potasiut<br />
dd CdLiM, <strong>The</strong> Black Rock ror6t<br />
Suleti{ No. 9 (Cm*zll{nih+<br />
HuiL@, NN Yck, 1939), ?. zo.
Principles of Graphical Excellence<br />
Graphjcal cxcetlencc ;s the \\,c1l-designcd prescmtarjon of intcrcsring<br />
data-a maaer of rrrLrrarrrc, of srnrisrirs, and of r/r-.iqrr.<br />
Graphical crcellcnce consnts of comptex tdcas communlcatcd<br />
rith clariry, prccision, and eliiciency.<br />
Graphical cxcellence is that nhich givcs to thc viewcr fie grcatesr<br />
number ofideas in the shortesr limc widl thc least nrk rn thc<br />
smallcst space.<br />
tt<br />
rT-r<br />
Graphical cxcellence is nearl,v alu,ays rnultivariate.<br />
And gr:rphical<br />
excellence<br />
requircs<br />
telhng dre truth about the data.<br />
cIAPHJCA L !XCTLLINCt J1
As to the roptiery and justness of rcprcsenting suns oI noneL and tine,<br />
by pats of spaee, tho' very rcadily agreed to by no* net, yet a Jew seen<br />
to q?rchend that therc hlay possibly be sone deception in it, of which<br />
they are not awdle. , , .<br />
Willian Playfair, Tlz Coanerial dnd Palitial AtlB (Lonaan,<br />
t86)<br />
People said: " lvith the chd/t ot1 the uall, uith the j&utes pullished, let's<br />
emdate and nuse otr etthusiasn in prodution."<br />
St*e St.tistic.l Bureau ofthe Peo?lei Relublic of Chitu,<br />
Statitt 6l Wotk in the N.u China (Beijing, 1979)<br />
Get it right or let it alone.<br />
<strong>The</strong> coftlusion fou<br />
junp to nay be your own.<br />
Jdnd Thurber. F,rrra FdrLrrl o,' Ti,s (Ncw York. 1e56)
2 Graphical Integrity<br />
For many people the frrst word that comes to mind when they think<br />
about statistical char* is "lie." No doubt some graphic do distott<br />
the underlying data, making it hard for the viewer to leam the<br />
truth. But data graphics are no different from words in this regard,<br />
for any means ofcommunication can be used to deceive. <strong>The</strong>re<br />
is no reason to believe that graphics are especially vulnerable to<br />
exploitation by lian; in fact, most ofus have pretty good graphical<br />
lie detectors that help us see right through frauds.<br />
Much of twentieth-century thinLing about statistical graphics has<br />
been preoccupied with the quesrion ofhow some amateurish chart<br />
might fool a naive viewer. Other important issues, such as the<br />
use of graphics for serious data analysis, were hrgely ignored.<br />
At the core of the preoccuparion $drh deceptive graphics was the<br />
a$umprion that data gnphics were mainly devices for showing<br />
the obvious to rhe ignorant. It is hard to imagine any doctrine<br />
more likely to stifle inrellectual progres in a field. <strong>The</strong> assumption<br />
led down two fruidess paths in the graphically barren ye:n<br />
from r93o to r97o: First,<br />
"alive,"<br />
thrt graphics had<br />
"com-<br />
to be<br />
municatively dynamic," overdecorated and cxaggerated (otherwise<br />
all the dullards in the audience would fall rsleep in the face<br />
of rhose boriDg statistics). Second, rhat the main task ofgraphical<br />
analysis was to detect and denounce deception (the duliards could<br />
not protect themselves).<br />
<strong>The</strong>n, in the lrte r96os, John Tukey made statistical graphics<br />
respccbble, putting an end to the view that graphics were only<br />
for decorating a few numbers. For here was a world-class dara<br />
analyst spinning of half a dozcn new designs and, more importantly,<br />
using them effective]y to e]?lore complex data.1 Not a<br />
word abour deception; no tortured anempts to consEuct more<br />
"graphical<br />
standards" in a hopeless efort to end all distorions.<br />
Instead, graphics were used as instrumen* for reasoning about<br />
quantiredve information. With this good example, graphical work<br />
has come to flourish.<br />
Ofcourse false graphics are still with us. Deception musr always<br />
be confronted and demolished, even iflie detecrion is no lonqer<br />
ar rhe forefront ofresearch. Craphrcal excellence beg'ns wirh;lling<br />
rhe truth ebout the data.<br />
ljohn \v. Tutey rd Martin B. Wilk,<br />
"Dda<br />
Anllylis &d Sbristica: Tcchniqu6<br />
dd Appro.ch6," id Edward R. Tufte,<br />
ed., <strong>The</strong> Qudtit^tit. Anald' oJ Srict<br />
&otiem Ee.ding, M:s., r9?o), 3?o-<br />
39o; ud]ohr'w. Tukey, "Some<br />
Gr2phic &d S.migmphic Displ.ys."<br />
-l<br />
in<br />
. L. Bnda{t, .d., sldtndut P4?4s ln<br />
Honot oJc.otee w. stud{ot ( m6,
J4 GRAprrrCAl !RACTTCI<br />
Hcre are several gmphics that feil to tcll the truth First, the case<br />
of the dnrppcarnrg baselinc in the annual report ofa company that<br />
wouldjust as soon forget about r97o. A careful look ar the middle<br />
panel reveals a ncgative incomc in rgzo, which is disguised by<br />
having the bars begin at the bottom at apprcximately minus<br />
$4,2c1o,ooo:<br />
This pseudo-dccline was created by comparing six months'<br />
worth ofpaymcnts in 1978 to a full year's worrh in 1c76 and 1977'<br />
with the 1ie repeated four timcs over.<br />
D^y Mn\es, L1c., t97 Ann dl Repn,P.1,<br />
N.u Yrl T/n6, Augrst 8, 1978,<br />
!. c1.
And sometimes thc fact drat numbers have a magnitude as well<br />
as rr order is simply forgotten:<br />
compa€ttve Annurl (;6t pd c.pib fd @ of tnM. in<br />
Pftrb!4h crty Hom6 ud Penn yl*trt, stare H6pihh,<br />
,r47 arn $93<br />
\what is Distortion in a Data Graphic?<br />
A graphic does nor distort if .he visual representation ofrhe data<br />
is consistent with the numcrical representation. \Vhat then is the<br />
"visual<br />
represenration" ofthe data? As pLysically measured on the<br />
surface of the gnphic? Or the pereiled visual e6ecr? How do we<br />
know that the visual image represents rhe underlying numbers?<br />
One way to try ro answer these quesrions is to conduct cxperiments<br />
on the visual perception of graphics - Laving people look<br />
at lines ofvarying length, circles of different areas, and then<br />
rccording rheir a*e-mcnt. of rhe numeric:l qLranririe..<br />
I tlink I 3e tbat d@ B<br />
ir 3.ia tim6 biggq thd<br />
dq A. L rhat txfiec1?<br />
Such cxperiments have discovered vcry approximate power laws<br />
relating the numerical measure ro rhe reporred perceived measure.<br />
For exrmple, the perceived arca ofa circle probably grows somewhat<br />
morc slowly than the actual (physical, measured) arca:<br />
the reported perceived area : (actual area)*, where x : .g+.3,<br />
a discouraging result. Dilicrent people see the same areas somewhat<br />
CRAPIIICALINTTCRITY JJ<br />
Piilsburgh Civic Comni$ion, Re?or o,<br />
Expentlihr.' o.f the D?pathht rJchdtitiet<br />
(Pittsburgh, 191r), t. 7.
JO CRAIIIICAt PtACTTCt<br />
differently; perceptions change with experience; and perceptions<br />
are context-dependent.' Particuiarly disheartening is the securely<br />
established finding that the reported perception of something as<br />
clear and simple as line lengrh depends on the context and what<br />
other people have alrcady said about the 1ina.3<br />
Misperception and miscommunication are certainly not special<br />
but what is a poor designet to do? A different graphic for each<br />
perceiver in each context? Or designs that correct for the visual<br />
trensformations ofthe average perceiver participating in ihe average<br />
psychological experiment?<br />
One satisfactory answer to these questions is ro use a table to show<br />
the numben. Tables usually outperforn graphcs in reporting on<br />
small data sets ofzo numben or less. <strong>The</strong> special power ofgraphics<br />
comes in the display oflarge &ta sets.<br />
At any rate, given the perceptual dificulties, the best wc can<br />
hope for is some uniformiry in graphics (ifnot in the perceivers)<br />
and some assurance that perceivers have a fair chance ofgening<br />
the numben right. Two principles lead toward these goals and, in<br />
consequence, enbance graphical integrity:<br />
<strong>The</strong> representation ofnumbers, as physically<br />
mea$red on the surfacc ofthe gnphic itsell<br />
should be directly proportional to the numerical<br />
quantities rcpresented.<br />
Clear, detailed, and thorough labeling should be<br />
used co defeat graphical distortion and ambiguity.<br />
rvrite out explanetions ofthe data on the<br />
graphic itself. Labcl imporhnt eventi in the data.<br />
,Thc drmsivc litemtue is summdized<br />
in Michael Macdonald-Ros, How<br />
NuDters Are Shom: A ReYiew of<br />
R.smrch on the Plefnaiotr of Qlnrit<br />
tive D.b in Tds," /c/iGt.val<br />
Con^unndtian Rerieu, 25 i.1977), 359<br />
4@. In larticular, H. J. Meihoefe. fnds<br />
sr€at vriabilitr among p.(eivasi rc<br />
"<strong>The</strong> Utility ofthe Clrde as d E6ecnvc<br />
Canogrlphic Symbol," C6ddtar Crtott'phd,6<br />
11969), 1as-1lji nd "'lhe<br />
Visul Pe.ception of the Cnde in <strong>The</strong><br />
matic Ma!s: Expdihfttel Re$lts,<br />
ibid., 1o (1971), 63-84.<br />
3S. E. Ash, "Studi6 of Irdepdddcc<br />
rnd Submision to Group PresE. A<br />
Minority of One Against a Unmimou<br />
M^jotlty. pr.hologi&l Monostarhs<br />
DFwing !y CEM; colt'right 1961,
Violations ofthc first principlc constitute one form of graphic<br />
misreprcsentation, measured by the<br />
Lie Fa*or :<br />
size ofeffect shown in graphic<br />
size of effect in data<br />
If the Lie Factor is cqual to onc, rhen the graphic rnight be doing<br />
a reasonablejob ofaccurately representing the underlying numbers.<br />
Lie Factors gr$ter than 1.oJ or less rhan .9J indicatc suL<br />
stantiai distortion, far beyond minor inaccuracies in piotting.<br />
<strong>The</strong> logarithm ofthc Lie Facor can be taken in order ro compare<br />
overstating (iog LF > 0) with understating (log LF < 0) crrors. In<br />
practice almost all distorrions involvc overstaring, and Lie Feoors<br />
of two to Iive are nor uncommon.<br />
Here is an extreme example. A newsprper reported that the<br />
U.S. Congres and the Deparrment ofTransportation had set a<br />
series offuel economy standards to be mct by automobile manufacturers,<br />
beginning wirh r8 rniles per gallon in 1978 and moving<br />
in steps up to 27.j by 198J, an increase of J3 percenr:<br />
275-li!^<br />
roo - s.rc"<br />
lg.0<br />
<strong>The</strong>se strndards and the dates for their anainment were shown:<br />
This linc, rep.senxng 18 mil6 per<br />
8.Ion in 1978, is 0.6 inches long.<br />
This line, rep.senting ,7.J niles pa<br />
g.Uon in 198r, is J.l incles long.<br />
GRAPHICAT INTTGRITY J7<br />
Frcl Eonomy Standards<br />
lor Autos<br />
Sel by Cdrgrcs eid slDlrendted by ho TnnrDortario.<br />
o€larlno.l. h hiler rer g.llon.<br />
N., Yol& It'6, Augut 9, 1978. ?. 12.
J8 cRAPIIIcAL ?RAcTIcE<br />
<strong>The</strong> magnitude ofthe changc from 1978 to 198J is shown in the<br />
graph by thc rclative lcng.hs of thc two lires:<br />
51-n6 ^ lrn : 7819"<br />
Ll.l)<br />
Thus the numerical change of 53 percent is presented by some<br />
lines that changed 783 percent, yielding<br />
which is too big.<br />
LieFactor:E: r+.s<br />
53<br />
lJre dr,play J'o ha, 'everoJ peculrrnre' ofperrpecrire:<br />
' On most roads the future is in front ofus, toward the horizon,<br />
and the present is at our feet. This display revenes the convention<br />
so as ro enggerate the severity ofthe rniieage strndards.<br />
. Oddly cnough, the dates on the left remain a constant size on<br />
thc page cven as they rnove along widr the road toward the<br />
horizon.<br />
. <strong>The</strong> numbers on the right, as well as the width ofthc road<br />
itself, are shrinking because oftwo sinultancous effects: the<br />
change in the values portmved and the changc due to penpective.<br />
Viewers havc no cha:rce ofseparating the two.<br />
It is easy enough to dccoratc these data without lying:<br />
UAUIRXD FUtrL ICONOMY ST^NDATDS:<br />
NEw c^Rs BUrr rRoM 1978 ro 198J<br />
19?8 1979 1980 1981 1982 i98l 1984 198J
<strong>The</strong> non-lying version, in addicion, pu* the data in a contcxt by<br />
comparing the new car standards with the mileage achicved by<br />
the mix ofcars actually on the road. Also revealed n a side of the<br />
data disguised and mispresented in the original display: the fuel<br />
economy standards require gradual improvement at star-up,<br />
followed by a doubled rate from 1980 to 1983, and llatteting out<br />
after thrt.<br />
Sometimes decoration can help ediorializc abour rhe substance<br />
ofthe graphic. But it is wrong ro disrort rhe dara measures rhe<br />
ink locating values ofnumbcn-in order ro make an cdirorial<br />
comment or lit a decorative schemc. It is also a sure sign ofthc<br />
Graphical HacL at work. Hcre are many decorations bur no lies:<br />
CRA?IIIC4LINTTGRITY J9
OO GRAPIIICAL PRACTICE<br />
Design and Data Variation<br />
Each part ofa grrphic generates visual expecotions about its other<br />
parts and, in the economy of graphical percep tion, these expeciations<br />
often determine what the eye sees. Deception results from<br />
the incorrect extrapolation ofvisual expectations generated at onc<br />
place on the graphic to othcr placcs.<br />
A scalc moving in rcgular intcrvals, for cxample, is expccted<br />
to continue its march to the vcry end in a consistent faslion, without<br />
the muddling or trickery ofnon-uniform chrnges. Here an<br />
irregular scale is used to concoct a pseudo-declinc. <strong>The</strong> first sevcn<br />
increments on the horizontal scale are ten years long, masking<br />
the rightmost interval offour years. Consequendy the conspicuous<br />
feature ofthe graphic is the apparent fall ofcurvcs at thc right,<br />
particularly thc dcciinc in prizcs won by pcople Gom thc Unitcd<br />
States (thc hcavy, dark linc) in thc most rcccnt period. This cffcct<br />
results solely Gom dcsign variation. lt is a big lie, since in reality<br />
(rnd even in exrrapolation, scrling up each end-point by 2.J to<br />
take the four years worth ofdata up to a comparable decade),<br />
the U.S. curve turned sharply upward in the post-197o interval.<br />
A correction, with thc actual data for r97r-8o, is at the right:<br />
l{obel Prizos Awarded in Science,<br />
lor Seloclod Countrics, 1901-1974<br />
1921- 1931 r94r- 1951-<br />
1920 1930 1940 1950 1960<br />
30<br />
Nariomi Scida loudaiion, S.me<br />
.rd@to6, r97, Cwashinston, D.C.,<br />
l{obel Prizes Awarded in Science,<br />
lor Selected Counlries,lg0l-1980<br />
l90r l9]]- 1921. 1931 1941. 1951. 1961 197r.<br />
1910 1920 1930 1940 1950 1960 1970 1930
Thc confounding of lesigfl 1)anado with data vaidtion o'er the<br />
surface of a graphic leads to ambiguity and deception, for rhc eye<br />
may mix up changes in the design wnh changes in the data. A<br />
steady canvas makes for a clearer picture. <strong>The</strong> principle is, thenr<br />
Show dara variation, nor design vanaoon.<br />
Design variation co upts thn display:<br />
OPEC Oilhlces: Atter 18<br />
tlonth3 olslability, Pric$ Are *[;{<br />
DuetoftseAlnin -,.,r;<br />
Ooll.nprb.tr l<br />
Five diferent vertical scales show the price:<br />
During this time<br />
19?3 r9J8<br />
January March 1979<br />
Ap l June 1979<br />
Juiy September 1979<br />
October December 1979<br />
one vertical inch equels<br />
$8.oo<br />
$4.73<br />
$4.37<br />
$4.rr<br />
$3.92<br />
And two dilferent horizonkl scales show the passaqe of rimc:<br />
During this time<br />
1971-197E<br />
1979<br />
one horizontal inch equals<br />
3.8 years<br />
o.J7 years<br />
As the two scales shift simultaneously, rhe distorrion rakes on<br />
multiplicative force. On the left ofthe gmph, a price of$lo for<br />
one year is represented by o.3r square inches; on rhe righr side,<br />
by 4.69 square inches. Thus exacdy the same quanrity is 4.69/0.31<br />
:1i.1 times larger depending upon where it happens to fall on<br />
the surface ofthe graphic. Tlar is design variarron.<br />
*s<br />
Quarterly<br />
CRAPIIICALINTECRITY 61<br />
N./ Yo* Tffi.r, De.ehber 19, 1978,
62 cRA?HIcAl PRAcrrc!<br />
Dcsign variation infected similar graphics in other publications.<br />
Hcrc an increase of+s+ percent is depictcd as an increase of4,28o<br />
perccnt, for i Lie Factor of9.4:<br />
And an increase of7o8 percent is shown as 6,700 percent, for a<br />
Lic Factor of9.j:<br />
55i:if:5*.?f: til'-1;g;,trr,** OPEC Be clnr|:llt Pris<br />
=:-.5.=:.3;A:; :q:*-."":.:.::ii 197019;9<br />
*;:.eJ,'* '- --=,*r'-:.'.}!.f ' "::j.- aF<br />
Ali these rccounts ofoil priccs made a second error, by showing<br />
the price ofoil in inflated (currcnt) dollars. Thc 1972 dollar was<br />
worth much norc then the 1979 doilar. Thus in sweeping from<br />
Tine, Lptil9, 1979, p. s1.<br />
Wthirytan Po!t, M*ch 2a, rs7g, p.
left to right over the surface ofthe graphic, the vertical scale in<br />
efcct changcs-design variation-because the value of money<br />
changes over the years shown. <strong>The</strong> onlv way to think clearly about<br />
moneyovcr hme is to make comparisons udng inflation_adjused<br />
units ofmoney. Several distinguished graphc designers did ex_<br />
press the prjcc in real dollars-and they also avoided other sources<br />
ofdcsign variation. <strong>The</strong> stars were Brrinc$ Iteeh, the Sunday<br />
Tinrr (London), and <strong>The</strong> Econanist.<br />
700<br />
500<br />
500<br />
.100<br />
100<br />
:00<br />
pnce of crude<br />
oil<br />
I<br />
:100 |<br />
t\0[1lNAL<br />
In the graphic wc saw first, the rwo sources ofdesign variation<br />
covered up rn intriguing, non-obvrou. a"pecr ofrhi d-ar:: irr rhe<br />
rour years Frtor ro the 1979-1980 increases, rhe rcal pricc of oil had<br />
lecllnel. Busy with decoration, tLe graphic had mised the news.<br />
kddsdRy'fuh{],i<br />
hroft.&- . _. .,j-<br />
_-<br />
GRAPI{ICALiNTTCRITY 63<br />
s0ft hrufi<br />
Th. tutunist,D.cc1ltr 29, 19:/s, p. 41.<br />
Sdr'dr Tir.r (London), Dccenber 16,<br />
Brsitr s Week, Aprjl s, rs7g, p. g.
71e Dislonr at the-Bolram are Ear+ kthar on the ]tuqLt hand Mdl)''
TLe Case ofSkyrocketing coveftment Spending<br />
Probably the most frequendy printed graphic, other rhan rhe daily<br />
weather map and stock-market trend line, is the display ofgovemment<br />
spending and debt over the years. <strong>The</strong>se arrays nearly<br />
always create the impression that spending and debt are r:pidly<br />
increasing.<br />
As usual, Playfair was the first, publishing this finely designed<br />
gnphic ir 1786. Accompanied by his polemic againsr the "ruinous<br />
folly" of the British govemment policy of financing is colonial<br />
wan through debr, it is surely rhe 6rst skyrockering govemment<br />
debt chart, beginning the now zoo-year history ofsuch displays.<br />
This is one of the few Playfairs that is taller then wide; less thaa<br />
one-tenth ofail his graphics (about 9o, drawn during 35 yeats of<br />
worl) are longer on the vertical. <strong>The</strong> tall shepe here serves to<br />
€mphasize the picture ofrapid growth. <strong>The</strong> money figures are<br />
not adjusted for inflation.<br />
But Playfair had the integriry to show an altemative venion a<br />
few pages later in Tlre Coulttlercial and Poltical Atl6. <strong>The</strong> inrerest on<br />
the national debt was plotted on a broad horizontal scale, diminishing<br />
the skyrocket effect. And, furthermore, "This is in real and<br />
not in nomiml millions" (page rz9):<br />
]ftd of dD NATIONAL DEBT f!!D .h€ R.Nl[tid.<br />
Tttz totun bu 6 r@t.ra* a4 bt,qib h@r'.r'tt w 9a "dazr.<br />
GRA?IIICA! INT!GRITY 6J
66 cRA?IIrcAr PRAcTrcr<br />
Although Playfair deflated money units over time in his work of<br />
1786, the matter has proved difrcult for many, eluding even modern<br />
scholars. This display helps its political point along by failing<br />
to discount for infatron and population growth and by r:sing a<br />
tall and thin shape (the area covered by the data is 2.7 times taller<br />
tharr wide):<br />
Let us look, in detail, at another graphic orr government spending:<br />
New lbd( State<br />
Toral Budget Expenditurcs and<br />
Aid to Localities<br />
Total Btrdger ---t<br />
t t<br />
Mafti. Fiot'tu, Congst: Kelttov of nE<br />
wthinStB BrablishMt G:tew H^aa\<br />
1917). p.92.<br />
Ntu Yotk Tin*,Feht@ry \ 9j6, p.
Despite the appearance created by the hlaeractive design, the state<br />
budget actually did not increase duing the last nine yean shown.<br />
To generate the thoroughly false impresion ofa substantial and<br />
continuous increase in spending, the chart deploys several visual<br />
and statistical tricks-all working in the same direction, to exaq_<br />
gerate rhe growrh ir rhe budger. <strong>The</strong>se graphical gimtlicks:<br />
This dusrer of rylc ehphasias &d<br />
sEerchcs our the low hlu for 1966-<br />
1967, a.ouaging thc impression tbai<br />
redt yem l,ve sttor u! ftom a shall,<br />
statle base. Horizontal do*s ?rovide<br />
Totat Budser --r P<br />
TorltAid to<br />
Lo.atitiss.<br />
Tlis squeerd-doM blo& of ryF<br />
contributes to an imse of small,<br />
sq@ered-dom rufuers b,ck in the<br />
Leaving behind the distortion in the charqjunk heap ax the lefr<br />
yields a calmer view:<br />
/|lill<br />
,rtlllillll<br />
CRA?IIICAL INTEGRITA 67<br />
<strong>The</strong>se tlJe par.lelepipcds }ave te@<br />
!la@d otr u olriei llme irto,t<br />
of the othd eight, dmring the imrge<br />
rlat the Dewer budgets towe. over rhe<br />
r 4 t 1 s 1 6 n<br />
i t<br />
Anows poinring sEaisht up emptasize<br />
r.ff, g'oMb. ComlaR wirh honzonut<br />
.row at left.<br />
ilrrffillll
68 cR,{?IIrcAL ?RAcTIcE<br />
Two statistical lapses also bias the chart. First, during the years<br />
shown, the state's population increaed by r.7 million people, or<br />
10 percent. Pert of the budget growth simply paralleled population<br />
growth. Second, the period was a trme ofsubstantial irflation;<br />
those goo& and services that cost state and local govemmcnts<br />
gr.oo to purchase in 1967 cosr $2.q in 1977. By not deflating, dre<br />
graphic rrixes up changes in dre value ofmoney with changes in<br />
tJre budget.<br />
Application of aithmetic makes it possible to take population<br />
and inflation into account. Computing expenditures tn constant<br />
Qeal) dollars yr capita tereals a quite different-and far more<br />
$360 -<br />
$34o -<br />
budg€i extdditurs,<br />
in .6tant dols<br />
1967 1968 1969 r97o 1971 1972 1973 19% 19jS 1y76 r97?<br />
Thus, in terms of real spending per capita, the state budget increased<br />
by about zo percent frorl.r 7967 to 7g7o and remained<br />
relatively corutant frorn r97o drough 1976. And the 1977 budget<br />
represents a substential ledine in expenditures. That is the real news<br />
story of these data, and it was conpletely missed by the craph of<br />
the Magical Parallelepipeds. Ofcourse no small set ofnumbers is<br />
going to capture the complexities ofa large budget-but, at any<br />
rate, why tell lies?<br />
<strong>The</strong> prbciple:<br />
In time-series displays ofmoney, deflated and<br />
standardized units of monetary measnremetrt ale<br />
nearly always better than nominal units.
<strong>Visual</strong> Area and Nurrerical Measqre<br />
Another way to confuse data variation with design variation is ro use<br />
areas to show one-dimcnsional abta:<br />
Acuoissemenl de nos<br />
etporlations d'aulos<br />
1927 - 1929<br />
And here is the incredibie shrinking doctor, with a Lie Factor of<br />
2.8, not counting the additional exaggeration from the ovcrlaid<br />
penpective and the incorrec horizontxl spacing ofthe data:<br />
TIIE SHRII{KIIIG FAIIII[Y DllCT(IR<br />
h Crllfornlr<br />
Psrcenhge 0l lhctors hvotsd Sohty io Famity practic€<br />
1:2,247RnnoDpopu,lro[<br />
8,023 DGlds<br />
GRA?IIlCAI INTTCRITY E)<br />
R. Satct, Iir G/a?t'iq,er eris, 1932).<br />
Iat Afg.le! Tin6, Angrsr s, 1979, p. 3.
70 GRA?HtC41 ?RACTTCI<br />
Many published e{Iorts using areas to show magnitudes make<br />
thc clcmentary nistake ofvarying both dimensions simulraneously<br />
in rcsponsc to changes in one-dimensional data. Typical is dre<br />
shrinking dollar fallacy. To depicr rhe rare ofinllation, graphs<br />
show currency shrinking on rwo dimensions, even though the<br />
value ofmoney is one-dimensioual. Here is one ofhundreds of<br />
TIt tt (irtrtqr), strlrslt ll,\r|'xrr!tr.!<br />
-'ffi<br />
TI lr l ,\r{cr ) !it,,!Tt:s o!.,\!}xu( -\<br />
rrlt (ft!E9{l!!E! rlt .!rLP.r!!J<br />
.r_<br />
"@ffiiil<br />
ilrj-:'IWWL-.*?<br />
'o t.s i<br />
Ifthe area ofthe dollar is accurately to reflect its purchasing power,<br />
then the 1978 dollar should be about twice as big as that shown.<br />
W$hitStan Pa{, O.tohet 2t, 1978. P. !.
<strong>The</strong>re are corxiderable ambiguities in how people perceive a twodimeruional<br />
surfaceand then converr rhar percepoon inro a onedimensional<br />
number. Clanges in physical area on rhe surface ofa<br />
graphic do not reliably produce appropriately proponional changes<br />
in perceivcd areas. <strong>The</strong> problem is all the worse when the arexs<br />
are tricked up into rhree dimensions:<br />
By surface area, the Lie Factor for this graphic is 9.4. Bur, ifone<br />
takes the barrel netaphor seritrusly and assumes that 6e uolune o{<br />
the barrels represents the price change, then the volume ftom 1973<br />
to 1979 ncteases 27,ooo percent compated to a data inqease of<br />
4J4 percent, for a Lie Factor of 59.4, which is a record.<br />
Similarly, a tltee-dimensional represenration pu6ng up<br />
oDe-{lmensloIr3t data:<br />
Conclusion: <strong>The</strong> use of wo (or three) varying dimensions to<br />
show one-dimensional data is a weaL ald ineficient rechdque,<br />
capable ofhandling only very small &ta sets, often v/irh error in<br />
desiga and ambiguity in perceprion. <strong>The</strong>se d€signs cause so many<br />
problems drat they should be avoided:<br />
<strong>The</strong> number of information-carrying (variable)<br />
dimensions depicted should not exceed the<br />
number ofdimensions in the &ta.<br />
GRAIHICAT INTTGRITY 71<br />
Ntu Yotk Ti66,Jetty 2j.1981,
I<br />
t ,<br />
CASSEPOSTALI<br />
DI RISPARMIO ITALIANE<br />
Numerodei<br />
Lrbretti, Libretto medro e Deposlto totale<br />
al frne di oqni mese<br />
t Err Ei9R L<br />
This multivariate history ofthe Italiaa post ofrce uses two dimcnsions<br />
in a way nearly consistenr with this principle, with the<br />
number ofposal savings books issued and the average size of<br />
deposia multiplying up to total deposits at the end ofeach montL<br />
froE 1876 ro 1881.<br />
l@.otuo G.bag\o, T.6ia CtuaL delle<br />
Sr.rirri.a (Mil2n. se.ord editior, 1888).
Bur Plavlair's circles, an earlv usc otarea to shoiv mrenirudc,<br />
arc nor consistcnr rvidr rhe pnnciplc. snrce thc txc-dimensioral<br />
data (ciry popularions) are reprcscnted bl an arcal dara ncasurc:<br />
Perhaps graphics that border on cartoons should bc crcmpt<br />
liorn thc prnrciplc. Wc certainlv would not rvant to forgo the<br />
.1,34o pound chickcn:<br />
I<br />
J 1,4<br />
i!<br />
tl<br />
i<br />
t<br />
cnA PHlc4L I N T!CRtT Y 7l<br />
sti4 if. A,rtuitt R?J,a,a Baok lNe\|<br />
t.
125<br />
74 GRAPHICA! !RACTICE<br />
context is Essential for Graphical Integrity<br />
To be truthful and revealing, data graphics must bear on the quesrion<br />
at the heat ofquantitarive thinking: "Compared to what?"<br />
TLe emaciated, data-thin design should always provoke suspicion,<br />
for graphics often lie by omission, leaving out data srt{icient for<br />
comparisons. <strong>The</strong> principle:<br />
Graphics mrrst not quote data out of context.<br />
Nearly all the important quesdons are left uranswered by tlis<br />
display:<br />
19JJ<br />
codecticut Traffc Deths,<br />
Before Gq::) ud After (1qi6)<br />
Strictd xnforcement by rhe Polie<br />
Against C$ Ex@cding Speed linit<br />
A few more data points :dd immensely to the account:<br />
125 Connaticut Tnfic Ded4<br />
1951 1959<br />
19Jt 1959
Imagine the very dif,erent interpretitions other possible timepaths<br />
surrounding the r955-r93e change would have:<br />
/\/\/\/\<br />
Comparisons with adjacent states give a still better conrext, revealing<br />
it was not only Connecticut that enjoyed a decline in trafiic<br />
fatalities in the year ofthe crackdown on speeding:<br />
i,<br />
Tnfiic Deaths per roo,ooo<br />
Persons in Comectidt,<br />
Msacluset*, Rho& hland,<br />
&d New York, 19J1 r9r9<br />
GNAPHICAL INTEGIIIY 75<br />
DoMld T. c,nPbeu ud H. L3umc<br />
Ros, "<strong>The</strong> Colmcltidt Ceckdom on<br />
Speding: Time Series D,rr in Q@si-<br />
E\perineMl Analysis," in ldwad R.<br />
Tnfte. eC,., me Q"mtxative Analy\it of<br />
So./l Prorldr (Rsding, Mas., 1970),
76 cRA?IlrCAI- ?RACTTCT<br />
Conclusion<br />
Lying graphics cheapen rhe graphical an everywhere. Since the<br />
lics ofren show up in news reports, millions ofimages are printed.<br />
When a chart on television 1ies, it lies tens of millions of times<br />
over; when a N?1, Yor! Trraes chart lics, it lics goo,ooo times over<br />
to a great nany important and influendal readers. <strong>The</strong> lies arc told<br />
about the major issues ofpublic policy-the government budgct,<br />
medical care, prices, and fuel economy standards, for example.<br />
<strong>The</strong> lies are systematic and quite predictable, nearly always exaggerating<br />
rhe rate of recent change.<br />
<strong>The</strong> main defense of the lying graphic is . . . "Well, at least it<br />
was approximately correct, wc wcrcjust t(ying to show the general<br />
dire*ion ofchange." But many of the deceltive displays we<br />
saw in rhis chapter involved fifteenfold lies, too large to be described<br />
as approxnnately correct. And in several cases the graphics<br />
were nor even approximately correct by the most lax ofstandards,<br />
since they falsfied the real news in the data. It is the special character<br />
ofnumbers that they have a magnitude as well as an order;<br />
nurnbers measure 4lartlty. Graphics can display the quantitative<br />
size of changes as well as their direction. <strong>The</strong> standard ofgcning<br />
only rhe direction and nor rhe magnitude right is the philosophy<br />
that infbrms the Pravda School ofordinal craphics. <strong>The</strong>re, every<br />
charr has a crystal clear direction coupled with fantasy magnitudes.<br />
Po.r npoAyhrrx nporrsmfuo(r U9l2 i:ll. Pr1!dd, M^t 24, 1982,p,2,
A second defense ofthe lying graphic is thaq although the desigl<br />
irselflies, the accual numbers are printed on the graphic for<br />
those picky folks who want to know the correct size ofthe ef'ects<br />
displayed. It n as ifnot lying in one placejustifred Iifteenfold lies<br />
elsewhere. Fcw writers would work under such a modest standard<br />
ofintegrity, and graphic designe$ should nor either.<br />
craphical integriry is more lilely to result ifthese six principles<br />
are followed :<br />
<strong>The</strong> representation ofnumbers, as physically measured on the<br />
surface ofthe graphic itself, should be directly proportional to<br />
thc numerical quantities represented.<br />
Clear, detailed, and thorough labcling shouid be used to deGat<br />
graphical distorrion and ambiguity. Write out explanations of<br />
the data on the gra?hic iaelf. Label irnportant evena in the data.<br />
Show data variation, nor design variarion.<br />
In time-scries displays ofmoney, dedated and stendardized unirs<br />
ofmonetary measuremcnt are neady always better than nominal<br />
units.<br />
<strong>The</strong> number of information-carrying (variable) dimensions<br />
depicted should not exceed the number ofdimensions in the<br />
data.<br />
Graphics must not quote dara out ofconrext.<br />
GRA?IIICATINT!GRITY 77
Sources of Graphical Integrity and Sophistication<br />
Why do artists draw graphi-qs drat lie? WWhy do rhe world's ma.jor<br />
newspapers and magazines publish them?l<br />
Although bias and stereoryping are the origin ofmore than a<br />
few graphical distortions, the primary causer ofinept graphical<br />
work are to be found in the skills, attitudes, and organizational<br />
structurc prevailing among those who design and edit statistical<br />
graPnrcs.<br />
LacL ofQuantitative Skills of Professional Artists<br />
Lurking belind the inept graphic is a lack ofjudgmenr about quanrirative<br />
evidence. Nearly all those who produce graphics for mas<br />
publication are trained exclusively in the ine arts and have had<br />
little experiencc wirh the analysis of data. Such experience is essential<br />
for achieving precision and grace in the presence of statistics,<br />
but everr textbooks ofgraphical design are silent on how to think<br />
about numbers. Illustrators too often see their work as an exclusively<br />
artistic e.ntef,prise-the words "creative," "concept," and<br />
'\ryle"<br />
combine regularly in all posible pernutations, a Big Think<br />
jargon for the sma1l task of constructing a time-series a few data<br />
points long. Those who get ahead are those who beaurify data,<br />
never mind statistical integrity.<br />
<strong>The</strong> Doccine That Statistical Data Are Boring<br />
Inept graphics also fourish because many graphic arcists believe<br />
rhat statistics are boring and tedious. It then follows that decorated<br />
graphics must pep up, animate, and a1l too often exaggerate whar<br />
evidence there is in rhe data. For example:<br />
. Tirre's fint full-time chart specialist, an art-sdrool graduate,<br />
says that in his work, "<strong>The</strong> challenge is to present statistics aj<br />
a visual idea rather than a tedious parade ofnumben.",<br />
. <strong>The</strong> opening sentence of the chapter on statistical char* in Jan<br />
\\ite's Graphn Idea Noteloo&: "\Vhy are statistics so boring?"<br />
Sample illustrations supposedly reveal "Dry starisrics tumed<br />
r "tu<br />
n diffi.dt to l owwhyrhBeeme<br />
etors ar€ bcing repeated. in Playfair's<br />
original work these kinds of mndes<br />
were tror hade; mor€ov€r, th€se cdo<br />
were nor as wides!.ead in the 193os as<br />
thet are now. Perhaps ihe reson is .n<br />
inqese in th. pdceived tred for en!!s<br />
. , , without a coffomitaht increase in<br />
thibing in !hei. coBruction- Evidence<br />
gathded by tbe commitre on ehphics<br />
oa the Amdi.d sr,ri$iol Asociation<br />
indietes that fornal $aining in gnphic<br />
lr$dtnion bas had, marked d(line<br />
ai at1 lev.h ofedlcation over the lasr<br />
-Mat_<br />
few d.-.ad€s." Howard wair€r,<br />
ing Newspa!€r cnlhs Fit to Pritrt," in<br />
Paul A- Kol€a, et al., eds. Pida$ihg o/<br />
,/i{iile rd,gd!.2 (New Yo.!, r93o),<br />
2 Tine. Febtury r!, 1980, p- 3-
80 caA?HIcar PRAcrrct<br />
into symbolic graphici' and "Plain statistics embellishcd or<br />
humanized wirh pictures."3<br />
. Alinebook on graphics,Herdeg's Gnphis f Diagrans,isdcscribed<br />
by its publisher: "An intemational review demonstrating convincingly<br />
that statistical and diagrammatic graphics do not<br />
necesarily have to be dull."a<br />
<strong>The</strong> doctrine ofboring dara serves political ends, helping to<br />
advance ccrtain interests over odrers in bureaucmtic struggles for<br />
control ofa publication's resources. For ifthe numbers are dull<br />
dull dull, then an artist, indced many artists, indeed an Art Department<br />
and an Art Director are required to animate the data,<br />
lest the eyes ofthe audience glaze over. Thus the doctrine encourages<br />
placing data graphics under control of artists rather than in<br />
the hands of those who write the words and Lnow the substancc.<br />
As the art bureaucracy grows, style replaces content. And the word<br />
people, having lost space in the publication to data decorators,<br />
console thcmselves wirh thoughts thar statistics are really rather<br />
lf the statistics are boring, then you'vc got the wrong numbers.<br />
Finding the right numbers requircs as much specialized skillstatistical<br />
skill-and hard work as creating a beautiful design or<br />
covering a comPlex news slory.<br />
<strong>The</strong> Doctrine That Graphics Are only for the<br />
Unsophisticated Reader<br />
Many believe that graphical displays should divet and entertain<br />
rhose in the audience who 6nd the words in the text too dificult.<br />
For example:<br />
. Consuner Reports describes the design of their new consumer<br />
magazine for children: "For the 6rst test issue, CUi professional<br />
stafproduced an article about sugar that was longer on<br />
graphics dran on information. We had Gared children might be<br />
overwhelmed by too many facrs."s<br />
. An art ditector with overall responsibility for the design of<br />
some 3,ooo data graphics each year (yieiding 2.5 billion printed<br />
inages) said that graphics ale irtended more to lure the reader's<br />
attention away from the advertising than to explain the news<br />
in any detail. "Unlile the advertisements," he said, "at least<br />
we dont put naked women in our graphics."6<br />
tl^n \ . \tt\ne, Ctuphn Il1ed Notebaok<br />
(Ncw York, 1980), P!. 1a3, 16i-<br />
awaher Hedea, ed., qathitlDiaCtud<br />
@tticlr. re76).<br />
sco,emd R.?r,J, 4j 0u1y t98o), 4o8.<br />
6louis Silvdstein, Gnphics ar rhe Nd,<br />
Yorl ?ider," lresenbiion at the Fnst<br />
ccneral coddence on Social Ghphi6,<br />
Lecsbug, Vnginia, Odober 21, 1978.
. A news director at a national television nerwork said that<br />
graphics must be instandy u:rdentandable: "Ifyou have to<br />
explain it, don't us€ it."7<br />
This kind ofgraphical thinking lead-r to<br />
Th€ Compsy Car€t6rt. oae uaerl by I Our d tO<br />
<strong>The</strong> Consequences<br />
Emdoeees dutnE the Fisc.l Y.d tg49<br />
What E. B. White said ofwriting is also true ofstatistical graphics:<br />
''No<br />
one c:n wrire decendy who is distrr.rstful of the reader's<br />
intelligence, or whose attitude is parronizing."3 Contempt fot<br />
graphics and their audience, along with the lack of quantitative<br />
skills among illustrators, has deadiy consequenca for graphical<br />
work: over-decorated and simplistic designs, tiny data ses, and<br />
big lies.<br />
Like censo*hip, tlese constraints on graphical design lead to<br />
elliptical aad eccentric commuaication. In seeking to avoid<br />
the subdeties ofthe scatterplot, arrists drew up these convoluced<br />
specimens, forcing bivariate data into a univariate design:<br />
souRcts oI INT!GRITY AND SOPIIISTICATION 81<br />
Tlnr€rview vith aurhor, Jlny 1980.<br />
Mery Eldor Slqr CratirS Srdrltr6<br />
(New Yort, 19J2), p. j, eho alpre<br />
lriateiy dGdibcs rhb as u "lmc@$ary<br />
3ID W ia Stru.I,J., ed !. B. whjre,<br />
Tr. Elmar d/Stlc (New Yo!k, r9j9),<br />
Np Yorl Ttr6, Jue 16. r98o, p. ar 3.<br />
AIm D. Manvel, "Takrion dd Ece<br />
noDic Gowt\," Tedtion with R.?r.nution<br />
Ntu]et|d,9 Awe 1990), p. 3.
82 cRAlIIrcAL ?R,{crrc!<br />
But beyond reviewing a few examples, let us look more systematically<br />
at the level of graphical sophistication prevailing at<br />
differenr pub|c,rons. ln order to makc comparisont ,mong a<br />
variety ofnewspapers, rnagazines, scientfic joumals, and books,<br />
I have compiled a rough measure of graphical sophistication-the<br />
share ofa publication's graphics tlnt are /eldrirtlal. Such a design<br />
Iinl.s rwo or more v:riables but is rot a time-serie! or: map.<br />
Relational graptrics are.essential to competcnt statistical analysis<br />
since they confront statements about cauie and effect with evidence,<br />
showing how one variable afecs another. <strong>The</strong> design idea<br />
is a simlle one, although not quite as simple as the bar chart, timeseries<br />
plot, or data map. Relational graphics have been used since<br />
1765 rnd are printed billions of times and ways every year; and<br />
there is evidence that twelve-year-old children understand the<br />
design.'<br />
All these graphics count as sophisticated by our hardly<br />
demanding measureJo<br />
tl<br />
lfi<br />
1E0<br />
160<br />
110<br />
'\20<br />
100<br />
'rt - ri 7*<br />
, eil*<br />
It*<br />
1 **Agt )t€l{48tr0e6t/rr5t<br />
7.9,<br />
12<br />
g: I t!t-)t z<br />
-16 18 20 32 X4%<br />
22 21 26 2A 30<br />
#1f,tPfn+<br />
<strong>The</strong> frequency ofuse ofrelational designs was count€d for<br />
randomly selected issues from 1974 to 1980 of each of r5 news<br />
publications. A toul of about 4,ooo graphics were examined itr<br />
sampled issues. Scaling up the observed data by the frequenry and<br />
circulation ofthe publication hdicates that th€ sampl€ repreffnts a<br />
population ofz5o to 3oo billion printed graphical images.<br />
"j<br />
eclara lrancis BamberCd, hterlrF<br />
tation ol Gr.phs at the lleh@bry<br />
Sdtrool Le\e1," Cdthaln Uiiu*nq af<br />
An e ica E dt car ian dl Re i. dnh M onagtup h s,<br />
1l (May 19a2). Ad&rioul daa from<br />
tqtbooks ed smdudted tsts de<br />
1oA vdiety ofhesues ofcnphial<br />
inteUig@.e md complexity are lossible<br />
dd dother, data dmsity, is ditused in<br />
p"". ot citv r.itrffiJ<br />
2.8 Feet per Second Faster<br />
Neu York Tines, Fehtvry 29,1926,<br />
P 46.<br />
Isao Sato sd Miyohei Shidolara, N./<br />
Politkt tud E@lania (rokro, tn4), p.<br />
113; rJlpdese ligh school textbook.
SOURCES OF INTEGRITY AND<br />
Table r shows the raults, ranling the rJ news publications by<br />
graphical sophistication. Seven of the papcrs, fr oi\ pneda to tl.re<br />
WaIl Steet Jownal, prodt:ced no relational graphics among rLose<br />
sampled and usually limited themrelves to time-series. Other papers<br />
published more advanced graphics: theJapanese ,4:ahi (a masi circulanon<br />
daily), 4bahata t Red Fhg. a Communist prrty paper<br />
that appears, from the data, to have employed a sophisticared and<br />
talerrted graphical designer in,r97e), and Atton Keizai (a, financial<br />
daily), as well as Da Spiegel and, <strong>The</strong> hononisr. Although none<br />
reached the level ofsophistication found in displays ofscientfic<br />
daa (a random sample ofzzo graphics ftom Saerce t978-t9lohad,<br />
42 p€rcent ofrelational design), it is clear that some graphical<br />
intelligence is possible in news work, 3t leasr in Japan and at a few<br />
European weeklies.<br />
Table I<br />
ciaphi@l soplisti@tion, Wdld ptes, 19?4 19so<br />
,4!a,'dtd ('Red flag ) (Ja!d, d'ily,<br />
,ardl,t Sli,6!r (Japd, d.ity, 8,0Cr0.000)<br />
D4 S?i.gel (cmmy, weclly, 1,000,000)<br />
fie EadmiJr (Briain, weehy, 170,000)<br />
Nihoi Keizai shibb't (I^p^, dail.l<br />
findcial paper, 1,700,000)<br />
ae Mo,/e (lrdch, daly, a.40,000)<br />
-B,ri,e$ itl.st (U.S., welly, 800,Ooo)<br />
Ne/ Y"r* "irs (U-s., daiiy, 900,000;<br />
Suday, 1,500,000)<br />
Pd,ld (rJssR, .laily, 10,s00.000)<br />
Fto*Iwtet Allseneke (Gd|lhy, datr,<br />
300,000)<br />
fie Tt,.J (Brirain, daily, 400,0qt)<br />
irl6nt,Jrof PoJl (U.S., dailr, 600,000;<br />
Suday, 800,000)<br />
Itma (U.S., wcctly, 4,300,000)<br />
Die 2.'/ (c€rnmy, wekly, 300,000)<br />
Wdll siee t Jarnal lU.S., d^ily, 2,U]/J,O|NJ)<br />
Pe.corag€ of sratisri@l<br />
ghphics based on more<br />
rhD one qnabb, bur not<br />
93%<br />
7.6%<br />
s.7%<br />
2.4%<br />
1.7%<br />
o.5%<br />
0.U/o<br />
0.4%<br />
0.U/o<br />
0.0%<br />
0.0%<br />
o.o%<br />
0.s%<br />
202<br />
1t9<br />
454<br />
342<br />
297<br />
L14<br />
422<br />
54<br />
93<br />
107<br />
727<br />
213<br />
449<br />
soPHrsTrcATrON<br />
83
84 GRAPHIcAL PRAClIC!<br />
Japanese graphical distinction is consistent with that country's<br />
heavy use ofstatistical teduriques in the workplace and extensive<br />
quantitanve training, cven in the early years ofschool:<br />
. . . no nation rank' higher in irs collecrive pa'sion for rtaristics<br />
hJapan. srarisrics are rhe subjrcr ofr,holiday, local and uarional<br />
conventions, award' ceremonies and narionwide statistical<br />
collection and graph-drawing contests.<br />
"This year," said<br />
"we<br />
Yoshiharu Takaha:hi, a Govemment statistician, had<br />
almost 3o,ooo entiies. Actually, we had 29,836."<br />
Entries In ihe lchildren sl sunsrical graph contesr were<br />
'creened rhree uie. b1 judges. who givifirst prize this year<br />
to the work oflive 7-year-olds. <strong>The</strong>ir graph creation, titled<br />
"Mom,<br />
play with us more often," was the result ofa survey<br />
of3z claismates on the frequenry that mothers play with their<br />
off.pring and Lhe reasons giren for not dorng so. . . . Orhcr<br />
childrerl' *orL examined Lhe f,equency of{,mily phone urage<br />
and correlated the day's temperature with cicada singing.l'<br />
Note the reletional design of the last cJrildren's graphic mentioned.<br />
<strong>The</strong> five U.S. publications examined rank toward the bottom<br />
of the world lnt, along wlrh Puoda and t tew European papers.<br />
Note, in Table r, the complete dominance ofnon-relational<br />
designs ac thc lower-ranled newspapen and magazines. This is<br />
unfortunate because the relational graphic, unlike the simpler<br />
designs, is an explanatary graphic- surely a :ratural for news reporting<br />
and anal)sis.<br />
<strong>The</strong> statistical graphics found in collcge and even high sclool<br />
textbooks are more sophisticared than those in news publications.<br />
Indeed, grade school children may expcrience a greater density of<br />
relational graphics than someone who reads only B siless W?ek,<br />
the New Ya& Tines, Tine, the Wall Street Jounal, and the Washington<br />
Posr. Tables z and 3 record the graphical sophistication of<br />
textbooks and ofa varieq' ofstandardized educational tests.<br />
A comparison bctwccn these data and Table r suggests rhat most<br />
news publications ou*ide ofJapan operate at a pre-adLrlt levcl of<br />
intelligence in graphical design.t'<br />
llAndrew H. Malcoln\ "Data-Loling<br />
Japme€ Rejoice on Statistics Day, Ne/<br />
Iorl TiueJ, October 28, 1977, p, ^-1,<br />
1,R€ader of nNs publications, ?anidlult<br />
the elite !re$, have considehble<br />
educatiomland profssional att.inments,<br />
wiih tbe rc$hing sblhi.al skills- About<br />
30 percent of the 1.j million rcades of<br />
the suday Ne Yd! Tira attended<br />
coll€ge, according to a r98o liueJ market<br />
surev. <strong>The</strong> audidce for st.tisncal<br />
gaphics is shan6 tn,n mmy insfrtos
Table z<br />
caphi€l Sophntietion, Coleg€ and HigL schoot Textbook,<br />
M€dicine ed lutlic h€alrh: 11 eniclg in<br />
Judith Teu, er a1., Srdrtriai .4 Gridc ro<br />
th? Uik"ous<br />
Int/odudia9 to Plr'hot1gy, by Eta6t<br />
H ged, cr al.<br />
Gaetul Ch.nbtrt, by LttN. Pulng<br />
ry'. o/ Edl,, by Edmrd wit on, er,l.<br />
Weath6, astronoey, engiD€Eing: 7<br />
dtides iD Tdu, Statlrid: ,1 c,id.<br />
ComnuiorioD, vork, edu€rio4<br />
eoDonigr 20 anicl4 in Tmu,<br />
Stdtisti6: A C\ide to the Ufksogi<br />
Polidcal Behauiu a.f the Anentun<br />
Elnlorak, !y William H. Fluigd<br />
ed Nmcy H, Zinede<br />
Eedri.J. by Pail Samuebon<br />
Dehodaq in An.tita, by R&eft A. Datl<br />
Anencar Coumhdt, by J^rcs Q. WiJsan<br />
cr.zirr'l .ryi".?Li, by william Masrenon<br />
Th. Plojert Phrlns Coure, bt rtstd<br />
Nu Politis @d &ononks,hy Isrc S.to<br />
ed Miyohei shinolda 0.pe6e)<br />
Bia@ical Scien.e: An E@lagi@l Aw.oach,<br />
Biologicd Scio@s Clriculh Study<br />
TlE A36itatr E@@6y,by P.oy I.<br />
!$.iotou: <strong>The</strong> sttldf of Itunn<br />
R?lddoar,irr, by Ilvetu Tho6<br />
Neu Ehks dd social S.i6@, by<br />
Yokichi Yajim, et al. 0apoe)<br />
Rie o.f &e Andkan Na!ia., by<br />
I€wiJ Paul Todd ed Male Cuti<br />
Masqdd's Ak.n.aa Cortuent, rdised.<br />
ty Wiliu Mccitughu<br />
Pccentage of st.tistical<br />
gBlhics basd on more<br />
ths one mriatle, lur Dor<br />
82%<br />
68%<br />
66%<br />
4-,t%<br />
44%<br />
43%<br />
42%<br />
16%<br />
8%<br />
o%<br />
n%<br />
4a%<br />
2Vk<br />
|a%<br />
591)<br />
glo<br />
uh<br />
V.<br />
vh<br />
souRcls oF INTBCRITY AND SO?FTSTICATION 8J<br />
l7<br />
g2<br />
53<br />
59<br />
9<br />
43<br />
25<br />
39<br />
u<br />
33<br />
22<br />
2a<br />
132<br />
3<br />
5<br />
39<br />
70
86 cRA?FrcAr ?RAcrrcE<br />
Table 3<br />
Ghphi@l Sophiltication, Eduational Tesis<br />
N,tioMl uiversity dtuce *min,!ions.<br />
Review materials, Law School AdmissioD<br />
Test, UDiFd States 1975<br />
Standaidized tests for sbde shool, hish<br />
schoot, and colesei United States, 1970s:*<br />
Arithmctic, mrthm,tics.,lgehra, od<br />
analytic gmmedyt 21 ie*s<br />
Social $udics, hi$ory, dd<br />
govehrnenr; r+ tefs<br />
cmml ,bllity, 5 tests<br />
snlhics bAed on nore<br />
than one mnabb, bur not Nuhb€r of<br />
a timseiies or a na! smlhics<br />
I$qo<br />
67%<br />
41.%<br />
2470<br />
21%<br />
* Graphics colected ii Janes R. Bdiger, conpiler, Seia&'1 Stantradi2e'| Tat Itetu<br />
thdt Metrte Abilitt with CftprB (Wxshington, D.C.: Bueau of Soci.l Science<br />
And so, just as there is a double standard ofinregrity at a good<br />
many news publications-one {br words, another for graphics<br />
so there is a double standard ofsophistication. <strong>The</strong> statistical<br />
graphics are stupid; $e prose is often serious and sometimes even<br />
demanding ofexpertise, as can be seen in these sentences from a<br />
single issue of the Ner Yo&'fines.<br />
Rc.y.ling pero.lollrrs Dray po$lone rhe day ofreckoning. bur irs e11'e.rs would<br />
$on bccone intoldable without a steady de!recirtiotr in tleir puchsing<br />
power. rioxting stes ofexchrnge .mot restore ev€n r sembldce of<br />
Numerous facets of the performmcc secn decidedly uDfashiomble ifnot<br />
doMdght eccenFic: the squu+toed iNrunfr$l phasing lnd rhe fteqnot<br />
plodding tenpos in the arias, the Ronunricized tratm@t ofthe choFles, n\e<br />
sendous retards at every cadence, the often intrusiydy drboFre c@muo<br />
improvir.tions dd an inconsistent attitude tovard exprc$ion whnh rdg*<br />
fron leaving Malterid emthses to mechdical lircnlisn.<br />
<strong>The</strong> Cou.t shows tro dgn ofrelr€ting from irs view thar a state sovemmmt<br />
is protecied by sover€isn immuity ,g,inst .otrr ord{s to lay ietroactive<br />
dahages for past violations.<br />
And Dr. Gdi! crahrm, a nedical dnedor wirh Smirhkline Corp., nakes of<br />
Thohzine, noied thar neuroleptiG produ@ no cuploria, dd are rberefore<br />
mrikely to be ab$d by patieft wnh a hiliory of drug or alcohol dep@d@ce.<br />
"<strong>The</strong>y are, ifanything, dyslhorosenic," Dr. Gmnam eid.<br />
49<br />
47
Conclusion<br />
<strong>The</strong> conditions under which many data graphics are producedthe<br />
lack ofsubstantive and quantitative skills ofthe illusrraron,<br />
dislike ofquantitative evidence, and contempt for the intelligence<br />
of the audience-guarantee graphic mediocrity. <strong>The</strong>se condirions<br />
engender graphics drat (r) lie; (z) employ only the simplest designs,<br />
often unstandardized tine-series based on a small handful of<br />
dara points; and (3) miss the real news actually in the data.<br />
It wastes the tremendous conrmrmicativc power ofgraphics to<br />
use them merely to decorate a few numbers. Moreover, much of<br />
the world these days is observed and assessed qurtrtiratively-and<br />
well-designed graphics are far more effective than worib in<br />
showing such observations.<br />
How can graphic mediocrity be remedied?<br />
Surely there is something to be said for rejecting once:nd for<br />
all the doctrincs that dara graphics are for the unintelligent and<br />
thar.ror.rics are bonng. <strong>The</strong>se docnines blame the vrcoms lrhe<br />
audience and rhe data) rarher than the perperrators.<br />
Graphical competence demands three quite diferent skilis: the<br />
substantive, statistical, and a*istic. Yet now most graphical work,<br />
pafticularly at news publications, is under the diredion ofbut a<br />
single expertisc-the artistic. Allowing artisr-illustrators to control<br />
the design and content of statisticri graphics is almost like allowing<br />
typogmphen to control the content, style, and editing ofprose.<br />
Substantive and quantitative expcrrise must also parricipate in the<br />
design of data graphics, at leasr if staiistical integrity and graphical<br />
sophistication are to be achieved.<br />
SOURClS O! INT!GRITY AND soPIIrsTrcaTIoN 87
PART II<br />
<strong>The</strong>ory of Data Graphics
Evuyone spoke of an inJotnatiofl oeEioad, but uhdt theft uas in Jai<br />
was a on-iflfotuatiotl otte oail,<br />
Richa.d saui wur@, whdt-If, Co{A-Be (Philalelptt , re76)
4 Data-Ink and Craphical Redesign<br />
Data gtaphics should drew rhe vieweri attention to the sense and<br />
substance ofthe dao, not to something else. <strong>The</strong> data graphical<br />
form should present the quanticative contents. Occasionally artful_<br />
ness of design males a graphic worthy of the Museum of Modem<br />
A.t, but essenrielly staristical graphics are instruments to help<br />
peopie reason about quanritative information.<br />
Playfairt very first charc devoted too much oftheir ink to<br />
graphical appararus, wirh elaboraregrid lines anddetailed labels. This<br />
tme-series, engraved in Augusr 1785, is Gom the early paga of<br />
<strong>The</strong> Connetcial and Political Atlas:<br />
CzA.Er {Lu?onrs a2?z<br />
Ext"ozzs gf INGLAND at ,t rd/ran aZ,Mon?E LaLaru<br />
,fran d! y.a/ r'/7o at /782 4/ p, "/g/,.ti<br />
?u.Bddzn tr;u 6 Arila/ ;z.tr l;n/! tlr 4ib-nad tr.:u ,.b EmtnED Tzo.rJ,1*.D ronrr!<br />
zrttbtd * *t;d aa,4a! ryr.
92 TIIEORY OF DATA CRA?IJICS<br />
Within a year Playfair had eliminated much of the noniata<br />
detaii in favor ofcleaner design that focused attention on dre<br />
time-series itself He then began worLing wirh a new engraver<br />
and war soon producing clear and elegant displays:<br />
no.po.r6 sd ImFons ro std Aon DEI{MARI( & NOICIVAY fi.@ v@<br />
This improvement ir gnphical design illustrates the fimdemental<br />
principle of good statistical graphics:<br />
Above all else show the data.<br />
<strong>The</strong> principle is the baris for a theory ofdau graplucs.<br />
th. nAb h@td L4. ittu ZlSooo a.]z<br />
1o rzaq
Data-Ink<br />
A large share ofink on a graphic should present dara-information,<br />
the ink changing as the daa change. Data-ink is the non-era$ble<br />
core of a graphic, the non-redundant ink arranged in rerponse<br />
to variation in the numbe.s represented. <strong>The</strong>n,<br />
Data-ink ratio -<br />
&ta-ink<br />
total ibl used to print the graphic<br />
: proponion ofa graphici ink devoted to the<br />
non-redundant display of data-information<br />
: r.o - proportion of a graphic that can be erased<br />
without loss of data-tuformarion.<br />
A Gw graphics use every drop oftheir ink to convey measured<br />
quantities. Nothing can be erased without losing informarior in<br />
there continuous eight tracks ofan electroencephalogram. <strong>The</strong> data<br />
change from background activiry to a series ofpolyspiLe busts.<br />
Note the scale in the bottom block, lower right:<br />
DATA-INT 93<br />
I(etueiJr A. Kcoi, FMlanentals oJ EleF<br />
ttuc$e alogturhy g.tev Y6k, t9j1),
94 TIITORY OT D,{T,{ GNA}IIICS<br />
Most of the ink in this graphic is data-ink (rhe dots and labels<br />
on the diagonal), with perhaps ro-zo percenr non-data-ink<br />
(the grid ticl6 and the frame):<br />
tu-;.<br />
E *q-i'<br />
'E**<br />
.,' Y .<br />
*1'.f":i:<br />
*-<br />
.:";.;"i*r'"" n"** -'<br />
*uT'.'.i?,F..."<br />
[e169dor3'ol4im4$.lidcorFFduc<br />
Pbldo'a1o@j6.nc<br />
In this display with nearly all its inl devoted to mattcrs other<br />
than data, the grid sea overwhelms the numbers (the faint points<br />
scattered about the diagonal):<br />
Jotn \ylet Bffie\ size abd crle: An<br />
Esdf ofl th. Sttectur .f Bialosr (PtiIce
Anorher p&lished vetion of the same data drcve the shate of<br />
daa-ink up to about o.7 an improvement:<br />
R.r{ioaship ot Acrut Racs of Rccinnrion io predictcd Ra!6 oo4 cities t960).<br />
But a third reprint publication ofthe same figure forgot to plot<br />
the points and simply retraced the grid lines Gom the original,<br />
including the exces strip ofgrid along the top aad right margins.<br />
<strong>The</strong> resulting frgure achievc a graphical absolute zero, a null dataink<br />
ratio:<br />
DATA-INK 9J<br />
<strong>The</strong> thre smphics were lullisbed iD,<br />
resFctively, Stanley Kelley, J.., Richud<br />
l A)rs, ed willi.n c. Bowen, 'RegisthrioD<br />
md voring: Putting Firsr<br />
mings list," An{nM Politl.al Sderc<br />
Re,i@, 61 G96?),371; th@ ieprint€d<br />
in Edw,rd R. Tufie, ed-, fte Qdrirr'tive<br />
halysis of S'ridl I'toblem! Frudils,<br />
M.$. 1e7o), p. 262; md repiinred<br />
€ain in Wiliam J. Crofty, ed., P,,rlir<br />
opiniar ed Politi6: A Redlt F.leq
96 TgroRY OF DATA CRA?HrCS<br />
Maxibiziag the Share ofData-IDk<br />
<strong>The</strong> larger the share ofa graphic! ink devoted to data, the better<br />
(other relevant manen being equal):<br />
Maximize the data-ink ratio, wirhin reason.<br />
Every bit ofink on a graphic require.s a reason. And nearly always<br />
that reason should be that the ink presents new information.<br />
<strong>The</strong> principle has a great inany consequences for graphical editing<br />
and design. <strong>The</strong> principle makes good sense and generates reason'<br />
able graphical advice-for perhaps two-thitds ofall statistical<br />
graphics. For the others, the ratio is ill-defined or isjurt not appropriate.<br />
Most important, howcver, is that other principles bearing on<br />
graphical design follow from rhe idea of maximizing the shrre of<br />
data-ink.<br />
Two Erasing Pdnciples<br />
<strong>The</strong> othcr side ofincreasing the proportion of data-ink is an<br />
erasing principle:<br />
Erase non-dara-ink, within rceson.<br />
Ink that fails to depict statistic.l information does not have much<br />
interest to the viewer ofa graphic; in fact, sometimes such nondata-ink<br />
clutters up the data, es in the case ofa a\i& mesh ofgrid<br />
lines. While it is true that this boring ink sometimes helps set the<br />
stage for the dara action, it is surprising, as we shall see in Chapter<br />
7, how often the datr themselves can serve ar their own stage.<br />
Redundant data-irk depicts the same number over alrd over. <strong>The</strong><br />
labeled, shaded bar of dre bar chart, for example,<br />
unambiguoudy locates the altitude in six separate ways (any five<br />
of dre six can be erased and the sixth will still indicate the height):<br />
as the (r) height of the left line, (z) height ofshading, (r) height<br />
ofright line, (+) position oftop ho;zontal line, (5) position (not<br />
conrent) ofnumber at bart top, a]ld (6) th€ number itself. That is
more ways than are nccded. Gratuitous dccoration and reinlbrcernent<br />
ofthe data neasurcs qenerare much redundanr data-ink:<br />
Bilatcral symmetry ofdata measures also creatcs redundancy,<br />
as in the box ploq thc open bar, :nd Chernofffaces:<br />
@ n @<br />
t\<br />
Half-faces carry the same information as full faces. Halves may<br />
be easict to sort (by matching the right half ofan unsorted facc<br />
to thc lcft halfofa sorred face) than full f:rces. Or clse an<br />
asymmetrical full face can be uscd to repoft addirional variables.l<br />
Bilateral symmetry doublcs the space consumed by rhe design<br />
in a graphic, without adding new in(brmation. <strong>The</strong> fi:w studies<br />
done on the perception of symmetrical dcsigns indicatc that "whcn<br />
looking at a vase, for instance, a su\ect would examine one ofits<br />
symmctric halves, glance at the other halfand, sccing drat it was<br />
identical, cease his explorarions. . . . <strong>The</strong> enjoymenr of symmetry<br />
. . . lies not with rhe physical properties ofthc {igure. At lcast eye<br />
movements suggest anything bur symmdry, balance, or rest,"2<br />
DATA-IN]l 97<br />
1Bemh.rd nu.r a Hos Riedvyl,<br />
''Cnphic.l<br />
Ret'.esenration of Multi<br />
vrrirtc Data bt Meals ofAsyDnl-trical<br />
F{es," lautnat oI the Antntu sl4njtndt<br />
,lso.iar'of, 76 (DeceDrber 1s81), 7J7<br />
,Leonrrd Zrsne, yisual P{ettidk o.f<br />
Fd (l.{ew Yo.k, 1970),<br />
!!. 2j6 2j7.
g8 TIIEOAY Or DArA GRA?IIrCS<br />
Redundancy, upon occasion, has its uses: giving a context and<br />
order to complexity, facilihting comperisons over various par* of<br />
the data, perhaps creating an acsth€tic balance. In cyclical timeseries,<br />
for example, parts of the cycle should be repeated so that<br />
dre eye can track any pan ofthe cycle without having to jump<br />
bacL to the beginning. Such rcdundancy possibly inproves Marey's<br />
r88o train schedule. Those pcopJe leaving Pads or Lyon in the<br />
cvening Iind that the! trains run of,'the right-hand edge ofthe<br />
chart, to be picked up on the left agin:<br />
Attaching an extra half rycle makes every train in the first 24<br />
hours of the schedule a continuous line (as would mourting the<br />
original on a cylinder):
And, similarly, instead of once atound the world in rhis display<br />
ofsurface ocean current", one rnd rwo-thirds rimes around i. bimer:<br />
DATA-INK 99<br />
Kirk Brym ud Michael D. Cof,, Tle<br />
Cncubtion ofthc Wo.ld oc€dr: A<br />
Nundicrl Study. Prt 1, A Homogenous<br />
Modet," Jaundt aJ P\skal oeeaE
1OO THEORY OI DATA CRA?IIICS<br />
Most data representetloff, however, are ofa single, uncomplicated<br />
number, and little graphical repetition is needed. unles redundanry<br />
has a distincdy worthy purpose, the second ensing prhciple applies:<br />
Fra.e redundanL dau-ink, within rea.on.<br />
Application of the Principles in Editing and Redesign<br />
Ju:t as a good editor ofprose rudJexly prunes out unncccssary words,<br />
so a designer ofstatistical graphics should prune out ink rhat fails<br />
to present fresh drta-information. Although nothing can replace<br />
a good graphical idca applied to an interesting set ofnumbers,<br />
ediring arrd revision are as essential to sound graphical design work<br />
as they are to writing. T. S. Elior emphasized thc "capital importance<br />
ofcriticisrn in the work ofcreation irself. Probably, indccd,<br />
the larger part ofthe labour ofan author in composing his work<br />
is critical labour; the labour ofsifting, combining, constructing,<br />
expunging, corrcciog, testing: this frightful toil is as much criticai<br />
Consider this display, which comparcs cach long bar with the<br />
adjacent short bar to show the viewer that, urder the vario<br />
experimental conditions, the long bar is long€r:<br />
rT. S. Eliot, <strong>The</strong> F@dim ofcriticlsm,<br />
in &1e.te'1 E$ayt ry+1q2<br />
Olew York, 1932), !.18.<br />
Jan6 T. KuaicH dd N. Bnce Mccurchcon,<br />
'croseEnltmccndt<br />
of the<br />
Sour T,se oD Sinele HutrD Taste<br />
Papil\^e." Jorndl of Eapdineatal Psr<br />
thology: Ceaddl, 1os (1979), 76.
Vigorous pruning improves rhe graphic immensely, while still<br />
retaining all the data of the original. It is remarkable that erasinq<br />
alone can work such a ransformation:<br />
"ll,'r<br />
;L ! L] L]<br />
''lrl<br />
LlLtjb<br />
-lh<br />
:'<br />
ILr,,n-<br />
<strong>The</strong> horizontals indicate the paired compadsons and would change<br />
ifthc experimental design changed-so they counr ar informationcarrying.<br />
AII the asterisks are out since every paired comparison<br />
was statistically signifrcant, a point that the caption can note. Here<br />
is the mix ofnon-data-inL and redundant data-ink thatwas erased.<br />
about 65 percent ofthe original:<br />
-lr<br />
.I<br />
'll r<br />
lli'<br />
li<br />
tl<br />
f:Ir<br />
LLL<br />
LL,<br />
DATA-INK 1O1
]02 TII!ORY OI DATA GRA?IiICS<br />
<strong>The</strong> drta graphical arithmetic looks like this the origiral design<br />
equals the erased part plus the good part:<br />
,Ll I'ljrlrl<br />
'll'[ I']t'l<br />
'"'<br />
I r<br />
LLLI<br />
" l l ' l<br />
:llrllll<br />
ltuLil<br />
' l 1 r<br />
LJ L Ll lr<br />
,:11<br />
'".<br />
1,bLt-<br />
<strong>The</strong> nexr graphic, drawn by the distinguished science illustrator<br />
Rogcr Hayward, shows the periodicity of properties of chemical<br />
elements, exemplified by atomic volume as a function of atomic<br />
number. <strong>The</strong> data-ink ratio is les than o.6, lowered because the<br />
76 data points and the reGrence curves are obscured by rhe 63 darL<br />
grid marks arrayed over rhe data plane like a precision marching<br />
band of 63 mosquitoes:<br />
E '<br />
_l<br />
E<br />
50 70<br />
Atomic Number<br />
i<br />
i<br />
+<br />
Linu Pauh€, G?ntrdl Clenntry (Sm<br />
FrDcisco,194?), p- 64.
,ilY.1-:.5::-"'" :rth rhe csenrial informarion of the gr:phic.<br />
rhe penods and the cmpirical obr.*"rlonr.<br />
":T^i.,11.."'. rnarks 2nd p111 qfrh. fiamc can be safeiy erased,<br />
removed trom cbe denomiraror of Lhe d,ata_rn_k rado:<br />
Tlc-unclurtered displ:y bring. our:norher aspecr or rhe data:<br />
lii iiii "'il*.-:". do no, fir rhe smoodi ,r,_*".,, .*".,<br />
llrthei w.ell. <strong>The</strong> dJrJ_jnl rario has:ncreased<br />
only<br />
,. "U.u,.q,<br />
lhe frame<br />
;;ri<br />
ljnes rernarjung a! pur€ non_rDtormrlion:<br />
I<br />
DATA-INI{ 1OJ
1O4 TEEORY Ol DATA GRA?HrCS<br />
<strong>The</strong> reGrence curves prove essential for organizing the date to<br />
show the periodicity. <strong>The</strong> curves create a srmcturc, giving an<br />
ordering, a hierarchy, to the flow of infomation from the Page:<br />
d 9<br />
F-<br />
I<br />
Jg<br />
E<br />
60 70<br />
Atomic Numb€r<br />
Restoring the grid fails co orgadze the data. <strong>The</strong> ticks are too<br />
powerful, and drey also add a disconcerting visual vibration to the<br />
graphic. With the ticks, the reGrence curves become all the more<br />
necessary, since the eye needs some guidance through the rneze of<br />
dots and crosses:<br />
E '<br />
f<br />
SB<br />
E
<strong>The</strong> space opened up by erasing can be effectively used. Labels for<br />
the initial elemens ofeach period, an alkali, show the beginning<br />
ofeach cycle in the periodic table of elements-and in the graphic.<br />
<strong>The</strong> unusual rare-earths are indicated. In addition, the label and<br />
numbers on the veftical axis are mmed to read Gom left to rieht<br />
raLher *un boLrom to rop. maLirg the graphic slightly more<br />
accessible, a linle more friendly:<br />
Conclusion<br />
I<br />
I<br />
Na ll !<br />
"i ii<br />
' l r I<br />
, r<br />
,, |<br />
.,J<br />
.i lY 1 ;<br />
I<br />
I<br />
Five principles in the theory ofdata graphics produce substantial<br />
changes in graphical design. <strong>The</strong> principles apply to many graphics<br />
and yield a series ofdesign options through cycles ofgraphical<br />
revision and editing.<br />
Above all else show the data.<br />
Maximize the data-ink ratio.<br />
Erase non-data-ink.<br />
Erase redundant data-inL.<br />
Revise and edit.<br />
I<br />
DATA-rNK 1O5
With sauge piawes jll then gaps<br />
And o'er u abitable dowtrs<br />
Plate ebphants for want of touns.<br />
Jonatlo SwifCs indictnent of rTth
5 Chartjwnk: Vibrations, Crids, and Ducks<br />
<strong>The</strong> interior decoration ofgraphics generates a lot ofinl that does<br />
not tell the viewet anyrhing new. <strong>The</strong> purpose ofdecoration varies<br />
to make the graphic appear more scienti{ic and precise, to enliven<br />
the display, to give the designer an opportunity to exercise artistic<br />
slills. Regardles ofis cause, it is ail non-&ta-ink or redundant<br />
data-ink, and it is often charlunk. craphical decoration, which<br />
prospers in technical publications ar well as in commercial and<br />
media graphics, comes cheaper than the hard work required to<br />
produce intriguing mrmbers and secure evidence.<br />
Sometimes rhe decoration is thought to refect rhe artist's fundamental<br />
dcsign contribution, capturing rhe essential spirit ofthe<br />
data and so on. Thus principles ofartistic integriry and creativiry<br />
are invoked to defend-even to advance-the cause ofcharqjunk.<br />
<strong>The</strong>re are bctter ways to portray spirirs and essences than to get<br />
rhem rl'rangled up wrrh sLari.rical graphics.<br />
Fortunately most char{unl does not involve artistic considerations.<br />
It is simply conventional graphical paraphemalia rourinely<br />
added to every display that passes by: over-busy gdd lines and<br />
excess ticks, reduadant representations ofthe sinplest data, the<br />
debris ofcomputer plottiag, aad many ofthe devices generaring<br />
derign variation.<br />
Like weeds, many varieties ofcharlunl flourish. Here three<br />
widespread types found in scicntific and technical research<br />
work are catalogued unintentional optical art, the dreaded grid,<br />
and thc self-promoting graphical duck. A hundred charlunLy<br />
examples from commerciai and media graphics have been forgone<br />
so as to demonstrate the relevance ofthe critique to the professional<br />
scientific production ofdatr graphics<br />
Unintentioml Optical Art<br />
Contemporary optical art relies on moird effects, in which the<br />
design interacts with the physiological tremor ofthe eye to produce<br />
the distracting appearance ofvibration and movemmt.
1O8 TITEORY OF DATA GRAPIiICS<br />
<strong>The</strong> efect extcnds beyond the ink of the design to the lvhole Page.<br />
When exploited by the experts, such as Bridget Riley and Victor<br />
Vasarcly, op art effects are undoubtedly eye-catching.<br />
But statistical graphics are also oftcn drawn up so as to shimmer.<br />
This moiri vibration, probably the most common form ofgraphical<br />
clumer, rs inevitabiy bad art rnd bad data graphics <strong>The</strong> noise<br />
clouds the flow ofnrformation as these examples from teclurical and<br />
scientiic publicarions illustrate :<br />
[--5<br />
4u.fu"fii;il<br />
n::#g:'ss<br />
TECIDOS DTALGODAO<br />
( coToNNADES )<br />
E6N\\\<br />
NNN<br />
lliluk kffi-ffi<br />
iir- -- rii<br />
-<br />
( COITON IEXTILTs ]<br />
In$iruto de Expdsso Commcrcirl,<br />
Btd\il: Ctdphnat E.anoni.avEndtnti1s<br />
(Rio dcJeeno, 1929), p.15-<br />
@!<br />
L ]<br />
i;ndm trc;* 1<br />
pR00uciAlrDEricrD05<br />
ffiffiff<br />
t( /ip(rRwD"'l<br />
I TmPoRTAQA! |<br />
ffitutu-ffih -m<br />
-
t I 1.5<br />
TE<br />
.9 .^<br />
I<br />
0.0<br />
On chis page, what should have been simple tables are tumed into<br />
bad graphics pubtished in major scienrifc journals. Above a duck<br />
moir6 with an unintentional Necker Illusion. as the two back olanes<br />
optically flip to the front. Some prrrmids concerl others; and one<br />
va.iable (stacked depth ofthe stupid pyramids) has no labet or scale.<br />
Below, we learn very litrte about data, but do discov€r that moir6<br />
vibration may well be at a ma-rimum for equally spaced bars:<br />
CIIARTJ UNK<br />
Months after Operation<br />
Figur€ 2. Senal Echocardiogrsphic ass€ssmenrs ot rhe seve y ot Reguqirarion in he puhonary Aitoglaff in 3t pari€nrs.<br />
Th6 numedcal gradss werc assigned according ro iho sovsrfty ot rcgurgitarion, as toltovrs: o, nono; O.g, rdv6r: 1.0 !o 1.5, mitd:2_0,<br />
moderato: and 3.0, ssv€re,<br />
:l<br />
s<br />
!<br />
: 10<br />
Nichola T. Koucholtos, er ai.,<br />
''Replaemdr ofrhe Aorti. Roor with<br />
a Pulmonary Autograft in Chndrm od<br />
Youg Addts with Aonic-Valye Disqsq '<br />
me Ntu EntlardJastual a.f Medkine, 3p<br />
J.he, T. Knicki dd N. Bre<br />
Cutcheon,<br />
Me<br />
'CroleEnllu@mdt<br />
of rhc<br />
Sou T.ste on Single H1lmD Tasb<br />
Pepinae," Junat<br />
.tbtogf: Cd-al,<br />
oJ E lennent.t P\yros<br />
(le?s). j6-<br />
Eain M. CoDford md Maii€ n. Huot,<br />
''Glucose Thsfer from Male to Female<br />
Scbistoson6," S.idc, 2r3 (Septdbd
110 TIITORY OI DATA GRAP1IICS<br />
And, finaliy, from the stylc sheet once providedby the Joumal<br />
of the Anetican Staritical lssoc;ation, a graphic described as<br />
"an<br />
example ofa 6gurc prepared in rhe proper form :<br />
A. Avetage Prcbabilltles ol W lton Nl1.1l<br />
wnhn=10<br />
Ar'ERAGE PROSABILITY<br />
<strong>The</strong> display required r3r line-strokes and r5 digits to cominunicate<br />
its simple informrtion. <strong>The</strong> vibrating lines are poorly drawn,<br />
unevenly spaced, and misaligned with the vertical axis.<br />
Vibradng char{unk even frequents the graphics of major<br />
scientiic j oumals:<br />
Tle t@ most lrequdtly citcd<br />
(fooEbted) scidiifrc jomal: rsdom<br />
enple ofissrs lublishcd 1980 1982<br />
Jo,ndt<br />
6.f Blolosicdl Cheflnty<br />
Jounat aJ the Ahdldr Chehnal sochtr<br />
Joutuat of chenndt Pltt'i6<br />
Bio.hini'a et Biophtsi.a Add<br />
Proreedihgt oJ the Ndtiondl Auden! .J<br />
Neu Engtdnd Jottual o_f Medlci"e<br />
2%<br />
2%<br />
3%<br />
a%<br />
t't%<br />
12%<br />
t5%<br />
21%<br />
568<br />
565<br />
327<br />
432<br />
225<br />
438<br />
364<br />
3tt<br />
338<br />
"JASA<br />
sryte sheet," Joutudt o.f ttE Anonan<br />
Stdtktudl Asociatiai. 7! (Maeb
Moir€ effects have proliferated with computer graphics (in<br />
programs such as Excel). Such unfortunate panerns were once<br />
generated by means ofthin plastic tranGr sheers; now the<br />
computer produces instant chartjurL. Shown here are a few of<br />
the many vibrating possibilities. Crosrhatching should be replaced<br />
with tint screens ofshades ofgray. Specfic areas on a graphic<br />
should be labeled with wot& rather than encoded with hatchine.<br />
l:31:!:!f:l a a a a o t r'r'o'o'r'o'o'r'r.r<br />
a a a a a ...................<br />
NNNNNNNNW NRN./////////t'/m<br />
ffiHffi liiiiiffi:iiiiiiiiiii i:i::iii::iiiliii i:iii:iiiii: ::::::::::::::<br />
))))))))))))})Dl))l))) :<br />
CIIARTJI'NR 111<br />
a a a a a o a a<br />
a a a a a a a (<br />
ryfiw lEii*iEiEiE!;: ililililffiililffiililfl il[ilil1t ffi I ffi ililillll<br />
ililillililtii1iltt<br />
illtititl<br />
ffi i11:ii1:i:il1:ii illlllll:l-E ffiffi ilffi<br />
fil.ill]Tff[ffn i]fidrffiiifitdfiiliitfilfifit Ffiililil i ffi i : : i i i i i i : i i i : i i<br />
| | I il lllllllll 1ililililrililr ililililililil|ilililt ilililil1ililiii iiiiiiiliiiilit ilititi ililtl<br />
,m ilililIililIilililtililililililtilil | | | | | I | | |<br />
ffi ffix* ilmffi ,ttttllltl|ltl1tl||1ttlll >)>>>)))>"<br />
1+++H# N\N\N &ttt9tt mm$ NSTNSSS \\\'*<br />
This form of clurlunl is a twentieth-century innovation, and<br />
computer graphics are multiplying it more than ever. <strong>The</strong> handbooks<br />
and textbooks ofstatistical graphics, along with user's manuals<br />
for computer graphics programs, are 6lled up with vibrating<br />
graphics, presented as exemplars of design. Note the high
112 TAEONY OF DATA GRAPIIICS<br />
proportior ofchariunky graphics in the morc recent publicarions.<br />
Computer graphics are particularly acrive:<br />
Tertbmtr rd hadbooks of sbtistnal gblhnsi Perentage of<br />
dd mo,l5 for comluter smlhics programs grxlbics rith Total nuhbet<br />
(ordmd by date ofpublicarion) mond vjbhdon ofsmDhics<br />
wilttdc. Bti an, Gaphi. ^4ekatuJt<br />
A@riry Fdra (New Yorl, 1914)<br />
R. S.tet, t"r G'd?1t4,6 (Pr.is, 1932)<br />
Hdb€rr Arlin dd PiyDrond R. Colton, c/a?rJi<br />
H.'o tn Mdke ad ue m (New Yo'k, 1936)<br />
Mrry Elemor sped, C/rarmg srdr6,i6<br />
(New York, 1952)<br />
Atu c. I\.gea, Ctafin chdn' Handbook<br />
(W.rhington, D.C , 1961)<br />
F. J. M@khose and H. R. vllkinson, Ma?r<br />
dd Dtdj',,J (rondon, rhird €dition. 1971)<br />
c.rvin F. Schnid dd Stanron E. Sclxnid,<br />
HMdboak a.f Ctofin Ptes atr, (NewYork,<br />
A.J- M,ccregor, C/dprnJ Sir?/rried<br />
<strong>The</strong> usli ndut for a widdy disrlibured<br />
conrputer smphic !ac*ase: s.1s/cR 4PH Uei'<br />
clir'. (c!)'' Nordr cfotin . 1980)<br />
<strong>The</strong> nmud for a very er@sive conlurer<br />
grrhis proerrnr: TeI-.4-c td_l Ute" ' Mafl"al<br />
(Sd Diego, 1!]81)<br />
'127a<br />
29%<br />
17%<br />
32%<br />
14%<br />
220<br />
14%<br />
6a%<br />
53%<br />
Can optical art effects ever produce a bener graphic? Bertin<br />
exhons: "h is the designer's dury to make the rrrosc ofthis variation;<br />
to obtain the resonance [of moiri vibration] without provoking<br />
an uncomfortable scnsation: to firt with ambiguiq. without<br />
succumbing to it."1 But can sratistical graphics successfully<br />
"flin<br />
with ambiguiry"? k is a clever idea, but no good examples<br />
are to be found. <strong>The</strong> key difiiculty remains: moiri vibration<br />
s an undisdpliaed znbiguiry, with an illusive, eyc-straining quality<br />
thar contamimtes the entire graphic. lt has no place in data<br />
gmphical design.<br />
<strong>The</strong> Grid<br />
One of the more sedate graphical elements, the grid should urually<br />
be muted or completely suppressed so that its presence is only<br />
impliot-lest it compete with the data. Grids are mosrly for the<br />
initial plotting of data at home or ofice rarher than lbr putting<br />
I<br />
5<br />
95<br />
't34<br />
399<br />
459<br />
I<br />
J^.q\es Beii'n, S.thialos aJQaphi.s:<br />
Diagrd',r j Net,,/eJ, MoF (Madison,<br />
wisconsnr, 198j, rnnrlated by wilian J.<br />
!ers), !. 80 i this book is rhc Euslidl<br />
tradationof Bcrtin\importdtwo.k,<br />
sln ialasie Etu?h4"e (Pa^, 1967) -
into print. Dark grid lines are chariunk. <strong>The</strong>y carry no informrtion,<br />
clutter up the graphic, and generatc graphic activiry unrelared<br />
to dara information. This grid camoullages the profile ofthe d:ta in<br />
the age-sex pyramid of the populadon ofFrance in 1967:<br />
A revision quiets the grid and gives emphasis to the data:<br />
cHATTJUNK rrl<br />
BAed on dali in LTdiur Narional de 1,<br />
Sradtiqu. et des Erudes Economiqus,<br />
Aintone ndtudEt de la FrMe, ry68<br />
earis, 1968), p!. 3a3ji redram in<br />
Henry S. Shryock ddJacob S. Siesd,<br />
<strong>The</strong> Methads md Mateidls oJ Denoga?hy<br />
(wnhington, D.C., 197r, vol. 1, 24a_
E '.r<br />
tsT<br />
114 TIITORY OI DATA GRA?HICS<br />
Thc space occupied by the doubled grid lines consumes 18 percent<br />
of the area of this otherwise most ingenious design, a "multiwindow<br />
plot." Optical white dots appear at the intersections of<br />
the gnd lnes. (<strong>The</strong> plot shows the following: <strong>The</strong> large square<br />
contains X1, X, scatterplo* for the indicated levels ofx, and X,.<br />
<strong>The</strong> marginal plots on dre right are conditioned on X, and the<br />
plots al the top on X,. <strong>The</strong> upper right comer shows the unconditional<br />
X,, X, scatter.) Redrawing eliminates the vibration:<br />
3 t.i I<br />
trtrHHtr'<br />
PLOT OF PARTICLE PHYSICS MOMENTL]M DATA<br />
- 11<br />
ffi -GrT;<br />
l,ritl<br />
r_l<br />
f.7;:al<br />
Ell<br />
r--.-:-l<br />
E]<br />
t-tl<br />
l.+!<br />
t.r<br />
|'1': I<br />
ft<br />
ti:4ti I<br />
Paut A. Tlrey and John W. Tlkey,<br />
"Data-DrivenView<br />
Seletioni Agglom'<br />
eration dd Sharp@jng, ' id Vic Bamett,<br />
eC.., Iflteryrering Mthi"dndft Datd \Chichester,<br />
Engl,rd, 1981), 231-232.<br />
ffil<br />
$i<br />
:,;l'.<br />
r;}*
<strong>The</strong> grid in the classic Marey train schedule is very active:<br />
Thinning the grid lines helps a little bit:<br />
CIIARTJUNK 11J
"l!;tr*;<br />
116 TIIXORY OF DATA CRA?IIICS<br />
A better treamrent, howerer, is z gray grid:<br />
When a graphic serves as a look-up able, then a grid rnay help<br />
in reading alrrd interpolating. But even in this case the gridr shouid<br />
be muted relative to the dam. A gray grid works well and, with a<br />
delicate 1ine, may promote more accurate data reconstruchon<br />
than a dark grid.<br />
Most ready-made graph paper comes with a darkly printed grid.<br />
<strong>The</strong> reverse (unprinted) side should be used, for then the lines<br />
show through frintly rnd do not clutrer the &ta. Ifthe paper is<br />
heavily gridded on both sides, throw it oul<br />
Self-Proneoting Graphics: <strong>The</strong> Duck<br />
When a graphic is tahen over by decorative forms or compurer<br />
debris, when the dara measures and srructures become Design<br />
Elements, when the overall design purveys Graphical Style rather<br />
than quantitative infomation, then that graphic may be called a<br />
lac! in honor ofthe duck-form store,<br />
"Big Duck." For this building<br />
the whole structure is itselfdecoration, just as in che duck data<br />
grzpbrc. In Learning fron Las Vegas, Robert Yenturi, Denise Scoft
Brown, and Steven Izenour writc about the ducks ofmodem<br />
architecture-and their thoughts are relevant to the design ofdata<br />
graphics as well:<br />
When Modem architects righteously abandoned omament on<br />
buildings, they unconscioudy designed buildings rhat werc<br />
ornament. In pmmoting Space and Articularion over symbolnm<br />
and omament, they distorred the wholc building inro<br />
a duck. <strong>The</strong>y substituted for the innocent and inexpensfue<br />
practice ofapplied decoration on a conventional shid rhe<br />
rather cynical and expensive distortion ofprogram and structure<br />
to promote a duck. . .. k is now rime ro reevaluate the<br />
oncc-horrifying staremcnr ofJoLn Ruskin that archirecrure is<br />
the decoration ofconstrucrion, but we should append the<br />
warning ofPugin: It is:ll righr to decorate consrrucion bur<br />
never construcr dccoranon-.<br />
cHARTJUNK 117<br />
,Robet Verturi, Dcnisc Scoft B.oM,<br />
.nd SteYer lzenou, I4nirgl'o,, l,r<br />
/eJdr (Cambridsc, rcvised editi.n,<br />
1977), p- 16t. <strong>The</strong> rniti"l sratencnt of<br />
the duck concept n foud on pp.37 1ol<br />
a& D!.1, Flandes, New Yorkr phore<br />
snph by Edward Tuftc, Jury 2o@.
11d THEORY OF DATA GRA?HICS<br />
<strong>The</strong> addition ofa fake perspective to the data structure clutters<br />
many graphics. This variety ofchar{unl, now at high fashion in<br />
the world of Boutique Data Graphics, abounds in corporate annual<br />
r€ports, the phony statistical studies presented in advertisements,<br />
the mas media, and the more muddled sorts ofsocial science<br />
A series ofweird three-dimensional displays appearing in the<br />
magezine Anetiean Edurarion in the r97os delighted connoisseurs<br />
of the graphically preposterous. Here 6ve colon report, almost by<br />
happenstance, only frve pieces ofdao (since the division within<br />
each year adds to roo percent). This may well be the worst graphic<br />
ever to frnd its wav into Drinti
Thcre are rome superbly produccd ducksr<br />
ffi<br />
h<br />
r<br />
^sEo-<br />
CHARTJUNK 119<br />
witlian L. ($.1, et,t., 7t. cduo,,id<br />
ttdtl trlar (Sa.hnreDto, r973, r979),<br />
Applied lrigation Water<br />
19t12<br />
tttbb<br />
E^!q<br />
Eh<br />
\<br />
tE*tu<br />
Eb<br />
!b<br />
tE'4tu<br />
l@tu<br />
5.furd<br />
t+ar"'6aG-
12O TIIEORY OF DATA CiA?IIICS<br />
Occasionally designers seem to seel credit merely for possessing<br />
a new tecinology, rather than using it to make beaer designs.<br />
Computers and their aftliated apparatus can do powedil things<br />
graphically, in part by turning out the hun&eds ofplots necessary<br />
for good data analysis. But at least a few computer graphics only<br />
evoLe the response<br />
"Isn't it remarkable tJrat the computer can be<br />
programmed to draw like that?" instead of "My, what interesting<br />
darz."<br />
a0<br />
55<br />
o5!<br />
drs<br />
=r0<br />
-!5<br />
3t0<br />
E2s<br />
*20<br />
! rs<br />
I to<br />
llrf 5i0il cnliE colf colf<br />
utEi inqE aovl?or ttrEig<br />
tlluE Sitis<br />
@<br />
@<br />
a<br />
a<br />
Z<br />
N!<br />
NI<br />
N<br />
NI<br />
lliflAtIOx<br />
(I.ql5l<br />
tlxErtlotiE|rl (xrl00l<br />
3||0irn0Et (L.Ett<br />
mCt (1.10!l<br />
CilliE lI-l2ll<br />
<strong>The</strong> symptoms of the We-Used-A-Computer-To-Build-A-Du&<br />
Sy1&ome appear in this display liom a professionaljournal: the<br />
thin substance; the clotted, crinkly lettering all in upper-case sans<br />
serif; the poindessly ordered cros-hatching; the labels written<br />
in computer abbrwiations; the optical vibration-all these the<br />
by-products ofthe technology ofgraphic fabrication. <strong>The</strong> overly<br />
busy vetical scaling shows more percentag€ marLen and labels<br />
than there are actual &ta points. <strong>The</strong> observed values ofthe<br />
percentages should be printed instead. Since the inforrnation conrirts<br />
ofa few numbers :nd a good m:ny words. it is b€sr ro pass<br />
up the computerized graphics capability this time aad tell the<br />
story with a table:<br />
ColT. ?qIEi (I.lsU<br />
Colft0EiCE (t.2!ll<br />
mtEionlE (x-s!?l<br />
CoI?E!EicE tx-!221<br />
Anhu I!- Mils, Edic N. Goldenberg,<br />
ud Lutz E$ring, "Tyleset Politi6:<br />
Itup.ct of NewspaFa on Plblic Cofrden@,"<br />
Ahqidr Politk|l Stide<br />
R.!iee, 7 3 (19j9), 67-84.
Contdt dd tone of Iront-page<br />
articls in 94 U-S. newspapes,<br />
Odober &d Novernbd, 1974<br />
Wate.gate: defddnts ed lrosotoB, 5t<br />
!ord: ladon ofNixon<br />
Inflarion, ligh cost ofliving 415<br />
Goverm€nt conpetence: costs, $ality, nn<br />
elari€s of public chployees<br />
Co.ndence in govemncnr: powe. of 6<br />
sp€ial rersrs, t]usr in political<br />
lqdes, dilhonsry in lolirjd<br />
GovemeDt !owd: regdation of busine*, 154<br />
seOecy, control of ciA ed Flr<br />
Crihe 123<br />
Sttonages: energy, food<br />
Conclusion<br />
103<br />
10u<br />
68<br />
Pdccnt ofarticlc! wirh<br />
sle.ific pds@ or lolicy<br />
2a%<br />
3W<br />
52%<br />
4<br />
30%<br />
25%<br />
L3%<br />
r6%<br />
Charqjunk does not achieve thc goals ofits propagators. <strong>The</strong><br />
overwhelming fact of data graphics is that they stand or fa1l on<br />
their content, gracefi:1ly displayed. Graphics do not become<br />
attractive and intercsting through the addition of omamental<br />
hatching and false pcrspective to a few bars. Charqjurk can tum<br />
bores into disasters, but it carl ncver rescue a thin data set. <strong>The</strong><br />
best designs (for examplc, Mimrd on NapolcoD in Russia, Marey's<br />
graphical train schcdule, the cancer maps, the Ti,res weather hiF<br />
tory of New York City, the chronicle of the annual adventures<br />
of the Japancse beetle, the ncw view of rhe gala\ies) arc inbiguing<br />
and aniosity-proroleing, drawing dre viewer into thc wonder of the<br />
data, sometimes by narrative power, somctimes by immense detail,<br />
and sometime-s by elegant prcsentarion of simple but interesting<br />
data. Iut no infomatiof, no sense of discovcry, no wonder, no<br />
substance is gencrated by chardunk.<br />
Forgo ctrarqjunlc, including<br />
moiri vibration,<br />
the grid, and the ducL.<br />
CIIARTJUNK 121
Painting is special, seyrate, a nattel of neditatian and nntenplatioti,<br />
Jor ne, no physical action ot sorial spot- As nuth onyiausness as<br />
pssible. Clatity, conpleteness, quintessence, qltiet. Na noke, flo s.hnutz,<br />
na s(hn ,, o Iauw s&uanefti. PlrIedion, pa*i,rt1pts. cansananc.<br />
consunt ateness. Na palpitations, no gesticulatiotl, no lrotesqtlerie.<br />
Sphitraliry, sercnitl, absoluteness, ohefet1ce. No autonatisn,<br />
fla auideflt, no anxiet/, tro uttharcis, no cha ce. Detachneflt,<br />
disi te|estedfless, tho ghtfulness, transrcndmce. No hunbuging,<br />
no bunon-holing- nt e:plo ation. no ni,ing &hgt up.<br />
Ad Reinlardt, statement<br />
6r tle caraloguc<br />
of rhe exhilirioD, "Thc New Decade:<br />
3J Anericd Painters and Sariptors," Whitfty Mlsfln ofAncrid AJt,
o Data-Ink Maximization and Graphical Design<br />
So far the principles of maximizing data-ink and erasing have<br />
helped to generate a series ofchoices in the proces ofgraphical<br />
revision. This is an imporant result, but can the ideas reach beyond<br />
the dctails and parriculadtie! ofeditirg? Is it possible to do<br />
what a theory of graphics is supposed ro do, that is, to derive new<br />
graphical forms? In this chapter the principles are applied to many<br />
graphical designs, baric and advanced, including box plots, bar<br />
charts, histograms, and scatrerplors. New designs result.<br />
Redesign ofthe Box Plot<br />
Mary Eleanor Spear's<br />
"range bar"<br />
|.-Rong€<br />
andJoh Tuleyi "box plot"<br />
from lo*est to highestomouni---------l<br />
l* Inierouo.ril€ Ro nEeJ<br />
Mary llanor S|eu, Cltrttj Stdtntid<br />
(NewYork,1e52), P.166; and]ohn Ww.<br />
Tukey, E ?ldfurory Ddia /ralpJ (Reading,
124 THTORY OF DATA CRAIHTCS<br />
crn be mostly erased withour loss of information:<br />
<strong>The</strong> revised desigl, a 4adl,ile p/ot, shows the same live numbers. It<br />
n casy to draw by lund or computer and, most importandy, can<br />
replace the conventional scatterplot frame. <strong>The</strong> straightcdge need<br />
only be placed on the paper once to draw thc quartile plot, compared<br />
to six separare placings for the box plor. An altemative is<br />
but this design will not work cllcctively to ftame a scanerplot.<br />
Nor does it look very good.<br />
Pcrlups spccial emph:sis should bc given to the middle half of<br />
rhe distribution, however, as in thc box plot. This can be done by<br />
changing line weights<br />
or. even oc,rer. b1 olt.e*ing r\c niddle \aJf'<br />
This iarrer design is thc prcfcrtcd form of the quartile plot. It uses<br />
the ink efi'ectively and looks good.<br />
tn drese revisions ofthe box plot, the principJe of maximizing<br />
data-ink has suggested a variety ofdesigns, but the choice of the<br />
best overall arrangcment naturally also resn on statistical ard<br />
aesthetic criteria-in other wor&, the procedure is onc ol reaonable<br />
data-ink mrximizing.
<strong>The</strong> same logic applies to many similar dcsigns,<br />
"pa<br />
such as rhis<br />
rllcl scle'natic plor. T\cong..,l required 8o seporrrc plac_<br />
ing, ofrhe v,aighredge. ;o lrorizonr:1. and 3o rerric-lr:<br />
F<br />
ir i.<br />
ilfl r!r<br />
|?uIu<br />
-L-<br />
i -,- T i<br />
_i<br />
'<br />
l_<br />
-I- -f<br />
An erased version requires only ro verticals to show the same<br />
in{6lmation:<br />
I<br />
<strong>The</strong> largc reduction in the amouat ofdrawing is relevanc for the<br />
u.e o[:uch de.ign. rn rnformal. cxntorarory dara analy,is. *here<br />
lhe re*art-h worker's rime 'fould be de\ oreo lo rrurrers orher<br />
than drawing lines.<br />
-!<br />
DATA-INK MAXITIIIZATION 1.2J
126 TIIIORY O! DATA CRA?HICS<br />
Redesign of the Bar Chart/Histogarn<br />
Here is the standard model bar chart, with the design endorsed by<br />
thc practices and the style sheets of many statistical and screfltfic<br />
publicarions:<br />
ks architecture difers little from Playfair s original design:<br />
&port! hd li+db oI S.oaI/ArlD b ana &no itfleent prs 6. oDe Y.q F(E cld(!e' r.o to chri{hlqr y3a<br />
rtk ,rg/+ Arv'?,t 2.. ?zn t'hazt/b-,t ?o'./: .a.l z'tu tla.t tua R 27tu d' /lll/'4/bu 1-?'*t<br />
dk-&ktut*z.r6r8e.d<br />
/'4^+ 'rz td r-e
<strong>The</strong> box can be erased:<br />
And the venical axis, except for th€ ticli:<br />
Even part of the data measures can be erased, maLing a ulire<br />
jrid, which shows the coordinate lines morc precisely than ticks<br />
alone:<br />
ts%<br />
1$h<br />
5%<br />
DATA-INK MAXll\,llZATION 127
128 TFIORY O! DAIA GRAPIIICS<br />
<strong>The</strong> white grid eliminates the tick marks, sircc the numerical labels<br />
on the vertical are tied direcdy to the white lines:<br />
t5%<br />
to%<br />
5%<br />
Although the intersection of the dncker bar with the thinner baseline<br />
creates an attractive visual efect (but also the oPtical illusion<br />
of gray dots at the intenections), the baseline can be erased since<br />
rhe bars dc6ne Lhe end-point at the bottom<br />
I<br />
ll.,<br />
llllr<br />
Still, a thin bascline looks good:<br />
I .<br />
lr.l<br />
illlt
Erasing and data-ink maximizing have induced changes in the<br />
plain old bar drart. <strong>The</strong> teclniques-no frame, no venical axis,<br />
no ticks, and thc white grid-apply to other designs:<br />
a .<br />
-<br />
a<br />
80<br />
.llj<br />
10<br />
Var "b F 1lrdlh N!i.hed Box plot<br />
I<br />
I<br />
T<br />
!'l I]<br />
il<br />
I<br />
I<br />
I<br />
f<br />
!<br />
tI<br />
DATA-INI( MAXIMIZATTON 129<br />
Rolert Mccirl, Jobn \v. Tder, dd<br />
Walhe A. Ilren, Vanations ofBox<br />
Plois,' Anoiuk Sknsticif, 32 (a978),
13O THTORY OF DATA CRA?HICS<br />
Redesign of the Scatterplot<br />
Considcr thc standard bivariate scatterplot:<br />
A useful fact, brought to norice by dre maxnnization and crasing<br />
principles, is dut the frrme of a graphic can become an effective<br />
data-communicating element simply by erasing part ofit. <strong>The</strong><br />
frame lines drould extend only to the measured limits ofthe data<br />
radrer than, as is customary, ro sorne arbitrary point likc thc ncxt<br />
round nurnber marking ofi'the grid and grid ticks ofthc plot.<br />
That prrt of the frame exceeding the limits ofthe observed dara<br />
is nimmed o{I:<br />
'lhe<br />
resrlt, a tange-Jtane, explicitly shows the maximum and minimr:m<br />
ofbothvariables plottcd (a1ong with the range), infbrmation<br />
available only by extr:polation and visual cstimation in the conventional<br />
design. <strong>The</strong> data-inh ratio has incrcased: sorne non-dataink<br />
has been erased, and dre remainder ofthc frarne, now carrying<br />
information, has gonc over to the side ofdata-ink.
Nothing but the tails of tha6"-e oeed cl,r.og"'<br />
D4rA-TNXMAXTMTZ_ATION 131<br />
Corrientional Scatteplot<br />
RdgFFlan€<br />
A range-Game does nor requLe any vievring or decoding instructions;<br />
it is not a graphical puzzle and most viewen en easily<br />
rell whar is going on. Since ir is more tuformarive abour tbe daa<br />
in a clear and precise nlanner, the range-ftame should rcplace the<br />
nonJata-bearing fiame in many graphical appticatioos.
132 TIIIORY OI DATA GRA?IIIC5<br />
A smrll drift in the remaining ink<br />
a quartile plot:<br />
turns each range-ftarne<br />
Irasnrg ald cditing has led ro the drsplay often cxrrx rrmbcrs<br />
(thc rninnnurn, rnaximum, two quartiles, and thc rncdian for both<br />
variablcs). TJre design is useful lor anrlytical and cxploratory data<br />
analysn, as *-eli as for publisired graphcs where sumnary characterizations<br />
ofthe Darginal distributions have interest. Thc dcsign<br />
is nearly always bener than the conventionally ftamed scaterplot.<br />
R.rnge-franes can also present ranges along a singlc dimemion.<br />
Here the historical high and iow are shown in the vcrticll fleme.<br />
This is an cxccilcnt practice and should be used widcly in all sorrs<br />
ofdisphys, both scientiic and unscicntfic:
Finally, the cntire frame can be tumed into data by framing the<br />
bivariate scattcr widr the marginal distribution ofeach variable.<br />
<strong>The</strong> dot-dash-plot restlts.l<br />
.<br />
-<br />
=<br />
("i,yit<br />
I tl<br />
xi<br />
rr rl I Iii | | | |<br />
<strong>The</strong> dot-dash-plot combines the two fundammtai graphical<br />
designs used in statistical analysis, the marginal frequency distribution<br />
and tLe bivariate disrribtfion. Do.-dash-plots make routine<br />
whar good data analysts do already-plotting narginal andjoint<br />
distributions together.<br />
An empirical cunulativc distribution ofresiduals on a normal<br />
gdd shows the ourcl 18 terms plus the 3oth term, with all 60<br />
points plotred in the marginal distribution:<br />
DArA-INX MAXTMTZAtTON 133<br />
I Tne terninology {ollows tradition, for<br />
ssrerploB w€re oo.c.aled dor diagmDs"<br />
lor mmple, in R. A. Fishcls<br />
Statiltitdl Methodt .fot Reyfuh wotkdt<br />
(ldinbqsh, 1e2J).<br />
Crthbert DaDiel,,4plrddd's o/sr,t'ti6<br />
to Indsstlial Exrynnent.tioa lNew \otL,<br />
r9?6),<br />
!.1Jj.
134 TIITORY OF DATA GRA?IIICS<br />
Similarly, this data-rich graphic ofsignals from lulsers shows both<br />
::<br />
lt r.S MHz<br />
FREOI.]ENCY<br />
CELL<br />
,<br />
Narro$bad sp..tr! ot irdn,idual subPrls$. Xactr poitrr ot rhe irreffiiy<br />
1"(l) plolted on the risht is the sun ot rhe disiribnrion of itrtensiies acrcFs the<br />
rcerr€r bud\ridth slioNn jn th. o!nte..,{t tte tor is ptoit€d th. spect.un i\cfused<br />
or.r tb€ puhe.In thc linit oi mrir rhousxDdr of Jjuls.s this Noutd shof rhe .crcjler<br />
Tilroih] H. HanIiN rd Bahey J.<br />
Rickett, "Pdsar SisDl Procesing," in<br />
B@i Aldei, et al,, edr., M.dodJ i,<br />
cdhlntdtiahdl Phrsi6, yolune 1a: Rddio<br />
lr,',,drl G\Iew Yoil,1e7j), p- 1o8.
<strong>The</strong> fringe ofdashes nr rhe dot-dash-plot can connect a serics of<br />
bivariate scaters in a /rg?lot (since it raembles a set offringcd<br />
rugs and covers the statistical ground):<br />
Reflecting the one-dimcnsional projections ftom each scamei, rhe<br />
dashcs encouragc rhe eye to notice how each plot filtcn and translates<br />
the data rhrough thc scatter from one adjacenr plor to the<br />
ne).t. Sometinet ir is usefui to thinl< ofeach bivariate scaner as the<br />
imperfect empirical representarion oferl underlying curve that<br />
transforms one variable into another. In rhe rugplot, the sequence<br />
ofvariables can wander offas appropriate. <strong>The</strong> quenritarive hisrory<br />
of a singlc observation can be traccd through a series ofone- and<br />
two-cllmenslonal contexb_<br />
DATA-INK MAXII4IZAT]ON IJJ
13O TIITORY OI DATA GRA?IIICS<br />
Conclusion<br />
<strong>The</strong> lirst parr ofa theory of data graplncs n in place. <strong>The</strong> idea, as<br />
describcd in rhc prcvious thrcc chaprcrs, is that most ofa graphic\<br />
ink s|ould vary in re"pon.e to d:rta variarion. l lre theory lr.r.<br />
sometling to say about a great variery of graplncs workadry<br />
scientifrc charts, rle unique drrwings ofRoger Hayward, the<br />
exemplars of graphical handbooks, newspaper displays, computer<br />
graphics, standard staristical graphics, and thc rccent invcntions of<br />
Chernof and Tukey.<br />
<strong>The</strong> observed increases in eliciency, in how much of the graphic's<br />
inL carrics information, are sometimes quite 1arge. ln several cases,<br />
the data-ink ratio increased fron,r or .z to nearly r.o. <strong>The</strong> transformed<br />
designs are 1es cluttered and can be shruak down rnore<br />
readily than the onghals.<br />
But, are the transforrned designs lener?<br />
(r) <strong>The</strong>y arc nccesarily bcncr wirhnr the principlcs ofthc thcory,<br />
for morc information per unit ofspace and per unit ofink is displaycd.<br />
And this is significant; indeed, the lnstory of devices for<br />
communicating infonn:tion is written in terms ofincreases in<br />
eltrciency of communication and troduction.<br />
(z) Graphics are alnost always going to improvc as thcy go<br />
through edldng, revisior, and tcsting agaimt diFercnt dcsign options.<br />
Thc principlcs of ma-rinnznrg data-ink and erasing generate<br />
graphical alternatives and aho suggest a direction in whch revisions<br />
should move.<br />
(3) <strong>The</strong>n there is the audience: will those looking at the new<br />
designs be confused? Some of the designs are self-explanatory, as<br />
in the case ofthe rangeframe. <strong>The</strong> dot-dash-plot is more dilncult,<br />
although it sti11 shows all dre srandard infonnation fbund in thc<br />
scartcrplot. Nodring is lost to those puzzled by the fiame ofdashes,<br />
and sornething is gained by thosc who do undqstand. Moreover,<br />
ir is a lreqtent mistakc in thinking abour statistical graphics to<br />
underestinute rhc audience. tnstead, why not asume drat ifyou<br />
urderstand it, most other readers wil1, too? crrphics should be as<br />
inrcJJrgent ar'd soo|i.ri, "red r\ rl-e rccomp )<br />
hg re\r.<br />
(+) Some ofthe new designs may appear odd, but dis is probably<br />
because we have not seen drem befbre. <strong>The</strong> conventional dcsigns<br />
for statistical graphics havc bccn viewcd thou$nds ofdmes by<br />
nearly cvcry rcadcr ofthis book; on thc other hand, the rangeframc,<br />
rhc dor-dasLplor, rhe white grid, the quarrile plot, the<br />
rugplot, and the half-facejust a few times. With use, dre new<br />
designs will come to look jusr as reasonable as the old.
Maximizing data ink (within reason) is bur a single dimension of<br />
a complex and multivariate design task. <strong>The</strong> principle helps conduct<br />
expcriments in graphical design. Some ofthose experiments<br />
will succeed. <strong>The</strong>re remain, however, many other considerations<br />
in the design ofstatistical graphics-not only ofefrcienry<br />
but also ofcomplexity, structure, density, and even beauty.<br />
DATI-rNK MAXIMTZATTON 137
7 Mubifunctioning Graphical Elements<br />
<strong>The</strong> same ink should often serve more than one graphical purpose.<br />
A graphical elernent may carry data information and also perform<br />
a design finction usually left to non-data-ink. Or it might show<br />
several different pieces of d*a. Stuch nufuifunctioning graphical<br />
alernrnts, if designed with care and subtlety, can ef,ectively display<br />
compiex, rnultivariate data.r<br />
Consider, for e
14O TIIIORY O! DATA GRAPIIICS<br />
Building &ta measures out of the data increases the quantitative<br />
detail and dimensionality of a graphic. <strong>The</strong> stem-andJeaf plot<br />
constructs the ihsftibution of a variable with numben themselves:<br />
0l9 - 900 rccr<br />
h€ights of 218 volcan@s, lnn 100 fel.<br />
19 3 = 19,300 aer<br />
9476562<br />
9??19630<br />
69947 1 6654442t1 t009 I 50<br />
47665541209955t4?5<br />
99884433t92943336110?<br />
97 6666655 442221 gJ97<br />
0<br />
2<br />
1<br />
5<br />
7 3 |<br />
6 898665441077?61065<br />
1 988554311006521080?3<br />
8 6533t12t17<br />
9 37765542t9J0493<br />
l0 09u4131652t2<br />
ll 496Xmt63r<br />
t2 4542IL64<br />
l3 47810<br />
00<br />
t5 676<br />
l6 52<br />
t1 92<br />
l8 5<br />
l9 39?30<br />
<strong>The</strong> idea ofmaking every graphical clernort effective was behind<br />
the desi$ of the stem-andJeaf plot. In presenting his invention,<br />
John Tukey wrote: "If we are going to make a marL, it rnay as well<br />
be a meaningfrrl one. <strong>The</strong> simplest-and most usdul-meaningful<br />
mark is a digit."'<br />
Here. too. the &ta form the data measure. Note the bimodal<br />
distribution in the histogram of college studorts arranged by height.<br />
,"Some cnthic dd Senig.aphic Di5-<br />
!la}," in T. A. Bddoft,.d., Stdriri.al<br />
Papas in Hotul oJ AeotSe w. sftdecol<br />
(An6, Iow,, 1972),<br />
!. 296.<br />
Arim l-.Joiner, Living Histogrim,'<br />
Inttutioel Stdtitti.dl Pcww.4j G9? j),<br />
339-34o. But, for 6:rtha ilevelopments,<br />
s MuL F. Sclillitrg. Am E. WatLi$,<br />
ud Wilid WnkiB, h Htu]6 Height<br />
Bitu'l^l?" Thz A6di@ Statistkidi, s6<br />
(Aug6t 2@2), 223-2t9-
A distinguished graphic thar builds data rneasures out ofdata *,as<br />
designed by Colonel Leonard P. Ayres for his staristical hnrory of<br />
World WWar<br />
I, a book with several notable graphics al1 done by<br />
typewriter and rule. Consrructing rhe dara measues out ofeach<br />
American division's name (a numerical designation) tums wh:rt<br />
might have been a routine timc-scries into an elegant dnplay. (Note<br />
that rhe cumuladve design depends on the facr that none ofthe<br />
divisions retumcd before October 1918.) <strong>The</strong> triple-functioning<br />
data measure shows: (r) fie number ofdivnions in France for<br />
each rnonth, June r9r7 to October r9r8; (z) what particular divisions<br />
were in Fnnce in each morth; and (:) the duration ofeach<br />
division s prcsence in France.<br />
MUlTrtuNcTtoNINGGtApIrlCAtXrrMrNTS 141<br />
lit !a<br />
a<br />
& let<br />
a7 61<br />
.lo rlo 40<br />
t9 a9 5t<br />
a3 aa 3a<br />
al al 31<br />
? 7 7<br />
a5 35 0!<br />
15 !6 t6 t6<br />
79 79 79 f9<br />
fa 76 7a la<br />
'9,'2t29',<br />
a t q e t s q<br />
9o 90 90 90 90<br />
92 92 92 92 9a<br />
0, a9 a9 69 39<br />
6t 0t €a 6t 05<br />
?amtatat0<br />
a0 a0 !o a0 3o 00<br />
!o 50 !0 !o lo lo<br />
t! 5t 5:t t5 !! !t<br />
6 6 6 5 6 6<br />
tt zt zl zt dt 9l<br />
! t a a 4 a a<br />
2A2A?A,l2a?€<br />
t5 55 t6 t5 66 !5<br />
a2 a2 62 A2 A2 AZ<br />
7t 7t 71 77 tt 7t t,<br />
! t 5 5 ! ! ! 5<br />
5 5 5 5 6 6 5 5<br />
az a2 a2 62 a2 A2 a2 A2 52<br />
al at ,l]' at 4t ro {!. ,11 .u ,11 ,tt<br />
&424.€.c42a2 {:42 1E 42 4,42<br />
26 26 A 26 26 21 26 e6 26 2A 26 6<br />
2 2 2 8 l a 2 2 A 2 2 2 2 2 2 2<br />
t l 1 l L 1 1 1 : . r 1 1 1 r l 1<br />
191? 191a<br />
Leomrd P. Al,res, ?r. t/,,,,id cer<br />
rJ,,r (wrslington,D.c.,<br />
lere),!.io2.
142 TIIEORY OF DATA GRAPHICS<br />
Encoding of data measures can be frr more elaborate. <strong>The</strong><br />
ploted points here are Chemofffaces, which reduce well, maintaining<br />
legibility even with in&vidual areas of.o5 square inches<br />
as shown.3 <strong>The</strong> analyst would observe the standard X-Y scatterplot<br />
and then turn to the within-scatter detail, seeking clusters of<br />
similar observations over the X-Y plane. Oudying faces and those<br />
inconsistent with others in the neighborhood-drey are, of course,<br />
strargers should be identified by observation number or name.<br />
o<br />
o o<br />
o<br />
,"n /al<br />
\-,/ \ -/<br />
/_n<br />
1 '4, ,i6A<br />
rn L-_-]<br />
fPxA<br />
/:]\<br />
, r t<br />
o o<br />
With cartoon faces and evm numben becoming data-measures,<br />
we would appear to have reached the limit of graphical economy<br />
ofpresentation, imagination, and, let ir be admined, eccentriciry.<br />
to Repreent Points in k-Dimtri(Dl<br />
sp e cnphic l\y," Jounal oJ the Anet<br />
im Stdtistirdl Alecidtion 68 (l@e 1e73),<br />
36r-168. For d apl,lication of Aces le<br />
cated over two dinmsions, sec Howard<br />
'Wainer<br />
dd David Thisfn,<br />
"craphic<br />
D^n tu lysi,s," Awadl Reri* oJ Psy-<br />
.nobsy, i2 (r98t, 19r-24r.
But let us consider this shaped poem, "Easter Wingi' by George<br />
Herbert (1593-1633), which uses space-the lorgth of each Lineto<br />
depict quantity, all done r5o years before Playfair.a <strong>The</strong> lines<br />
double-function: the longer Lines describe wealth, plenty, largese,<br />
and rising to flight; shorter Lines tell ofpoverry and becoming<br />
"most<br />
thinne"; and lines ofintcrmediate lenerh indicate rransirion<br />
:nd chrnge (decrying. ri(ing. combining. be;oming):<br />
-EASter-utn{s.<br />
fOrd, rrho createdst man in wealth and store,<br />
l_7 Though foolishly he lost thc same,<br />
Decaying more and more,<br />
Till he became<br />
Most Doore :<br />
wrth thee<br />
O let me rise<br />
As larks, harmoniousl6<br />
And sing this day thy victories:<br />
<strong>The</strong>n shall th; fall turihe.ihe ffight in mc.<br />
My tend€r age ih sorrow did beginne:<br />
And still with sicknesses ahd shame<br />
Thou didst so punish sinne,<br />
I nat I became<br />
Most thinne.<br />
With thee<br />
Let me combine<br />
And feel this daythyvicto c:<br />
For, if I imp mv'wiie on thine.<br />
Affiction shall a'doaice thi flight in-me.<br />
MULTIIUNCTIONING cRA?ErCAI, ELXMENTS 143<br />
a For a remrkable orsoc liketo!. of<br />
the maDy tFographicai varlant shapes<br />
of 'Esier Wingi in its long publcation<br />
hnrory, see lhe esay tr^r tLUx," by<br />
"Rrndom cloud" in Randall Mcleod,<br />
eC,.,C'8it tt Editing: Texx oJ the E'glish<br />
Rrfuisard (New Yo.k, 1994).61 172.
144 THTORY OF DATA GRA?HICS<br />
Fnrally, this graphical pun: the visual d:ta as the data measure,<br />
as in thc living histogram. <strong>The</strong> chart shows how states once dil<br />
{ircd in their engineering standards for painting lane stripes on<br />
road pavemenr. Some states marked the road lanes with short<br />
dadres and long gaps; others used only solid lines. Portrayed in<br />
the graphic is the acrual physical pattcrn painted on the road, with<br />
48 U.S. srates ordered by thc lcngth ofthe painted mark:<br />
Ohio<br />
1@ 1Jo 20o 25o 3oo 3Jo<br />
R€dnm Gom A. R- Laucr, Prycholosical<br />
Factos h Efec.ive Trafiic conttot<br />
Ddi.ccs, Ttdfrc Q,an$]L 5
Data-Based Grids<br />
Very occasionally the grid can repot dircccly on the data. This<br />
grid is formed by the locarion of measuemcnr instruments; the<br />
plain dots register a zero reading, in contrast with the white background<br />
where no readings were taken. Erasing the grid would<br />
erase measured data (rather unevcnt6rl, to be sure). Such is not<br />
the case for most grid dots, ticLs, and lines.<br />
<strong>The</strong> arrangement of data in this table-graphic yields an intemal<br />
grid, a rare exanple ofdara as grid:<br />
ADU!I CHILDREN<br />
r!e; lleiShE I ad !.viarioft sm 68!in ,lks,<br />
MUITIFUNCTTONING CRAPHICAL TTEMTNTS 145<br />
K. v. Robeirs md D. E. Porier, "MagnetohydrodyDnic<br />
Calolations," in<br />
Bemi Ald€r, et al-, edi., Merl,o6 ir<br />
Conpltational Phyti.t: Vol"ne 9, Pldma<br />
P/4/rid (New York, 1970),<br />
!. 4o2.<br />
Ktl Peasan, mo Life,lLfi,s mti Ldbout\<br />
oI Ftun.i! Galtan lctmbideq
146 THSORY OF D,\TA GRA?IIICS<br />
_elndgh!TQvdM!Ds<br />
sJBiq,h nto! dd'4 rdsq. a<br />
b d! r,ir&r eiq vh{ Joi! Mrd<br />
vidddlhedigid)o9iidIi!qish|<br />
TL€ Unied Sab in Nodh Are.icl<br />
(Mirhell Mip)<br />
i.e +@is .h. &dsEd.d d & .sd Fd@ d Nod<br />
ic F4hc/ d Nod Aire 6d d. s<br />
liFole.ddei.lFid<br />
Here the grid is the elenent ofinterest, rather than the rrap. IrsEr J. Ca!!on, Barbara Batz Petche<br />
nik, md John Hamilton I@9, ,4116 o/<br />
Eatly AneticM Hi$ory efir.etm, 196),<br />
P. J8.
<strong>The</strong> gnd that follows presents the data on rhe surface of the ro&;<br />
on the sides, the grid n conventional. <strong>The</strong> rwo displays comprre<br />
the efi'ect ofreligion, taLing into accormt parry afiliation, on a<br />
person's votc for preJdent in 19j6 and in 1960 (when a Catholic<br />
ran lbr president). Note there is no reliable slope associatcd widr<br />
religion in 1956, once party is controlled; in 1960, a sysrematic<br />
effect is found. Readnrg the slopes in the other direction shows the<br />
persistent efi'ect ofparty in both elecions:<br />
z<br />
g<br />
Weak Strong<br />
Independenl Republi€n<br />
E<br />
1@%<br />
75%<br />
50%<br />
25%<br />
w<br />
MULTIFUNCTIONING GRArrrrc4[ rrxMrNTs 147<br />
Plilip E. Convere, _Religion and Po1itics:<br />
<strong>The</strong> 196o Election. in Angus<br />
Canrlbell, Pltiip E. Convdsq Warrcn<br />
E, MiUe., andDoDld E. Stokcr, Eie.nonr<br />
ari ke Pohicdl On10 (tlew \ork,<br />
W*k Strong<br />
lndelend€nt Republican
148 THEORY OF DATA GRA?IIICS<br />
Playfair tled the grid ro the data in his skyrocketing debt graphic.<br />
Although thc implicit piotting coordinates are based on regular<br />
intervals, the verticalgrid lines nr the published version are irregularly<br />
spaccd, ke,r'ed to significant events. <strong>The</strong> data-based grid is<br />
a shrcwd graphicrl device, sewing rather than fighting with the<br />
data. It is a technique underused in contemporary graphical work.
Double.-Functioning Labels<br />
MUITIIUNClIONING<br />
Data-based coordinate lircs lead. to data-basei la6els, as, for example,<br />
at the boftom of?la),{air's debt graphic. Again, the issue is rhe<br />
same: why not use the ink to show data? Beginning with conventionally<br />
labeled frame<br />
0<br />
10<br />
and erasing to the range-frame<br />
_------.1<br />
2 0 3 0 { o<br />
leaves those lonely ticks and numbcrs out on the raiis, workine to<br />
help rhe eye ger a berrer readrng on nhere the line ofrhe ransF<br />
frame ends. Bur tharjob can be done better by inuestbg rhe irme<br />
ink in data: radrer than showing the minimum round number<br />
and the maximum round number at the ends of rire ftame, show<br />
the actual mininum and maximum realized in the data:<br />
10<br />
_ - - - - . ' - - - -<br />
20 30 37<br />
Witl itr greater precision ard rwo tick-marks less ofnon-datainL,<br />
the range-frame wirh range-labels is superior to rh€ range-<br />
Game with round number labels. Both improve on the shnalard,<br />
pasive frame.<br />
Numbers also double-function when used both to name things<br />
(like an identification number) and to refect an ordering. In this<br />
graphic (in which the circled numbers fail to double-function),<br />
each number idcntiies a parricular study ofthe rhermal conduc,<br />
tivity of tungsten, ordered alphabetically by the last name of the<br />
first author. Ifthat list were order-ed by &te ofpublication instead,<br />
then the code would also indicate the time order in which<br />
40<br />
GRA?EICAIEtttrlENTS r49
4.4<br />
3<br />
150 TIIIORY OF DATA CRAPI{ICS<br />
the various conduoivity determinations were made. Thus "1"<br />
would indicate the earliest study, and so on or, altematively,<br />
"6rc"<br />
would be the third study published in 1961. Such informadon<br />
has interest, since we could see which of the *rly studies got<br />
the right answer. tn addition, the movement ofthe studies toward<br />
the "conect" recomrnmded values could be tracked. This extra<br />
information requires no additional ink.<br />
THERMAL CONDUCTIVITY OF<br />
TUNGSTEN<br />
In most graphics, the coordinate labels are far from the dara<br />
measures. Consequcntly the eye ofthe viewer must move back<br />
and forth between the path formed by the data and rhe coordimte<br />
positions arayed along the margins ofthc graphic. Sometimes this<br />
eye-work c:n be eLiminated cntirely by turning the coordinate<br />
labels inro dara meesurcJ. Jnorhcr doublFftmcrioning maneuver.<br />
Take the exarnpie from the style sheer ofthe-/ormal o.f the Anetia<br />
Statisttual A:sociation:<br />
C-Y.Ho,R. W- Powell. dd P. E- Liley,<br />
Thdnal Cobiludivq a.f the Eknenrs: A<br />
Conpfthen\ite Rerie'!, srpplenle^t ^o.<br />
r,lor al o.f Phy'nd| dnd Che\nat ReJierc<br />
Datd, j \r9J4),1492.
<strong>The</strong> grid inoement! ofthe X-axis are relocated upward to mark<br />
the path of the data:<br />
.15<br />
.10<br />
.05<br />
, t n ,o,r rr.ruros<br />
A:rd since the issue in this display is the probabiliry at each integer<br />
value, the round-number Y-scale is replaced by exacr values:<br />
.t77<br />
.rL4<br />
.u5<br />
.052<br />
.034<br />
.025<br />
.004<br />
.w2<br />
, , n TorTrr13ro$<br />
<strong>The</strong> Y-scale now resembles rhe dashes ofthe dot-desh-plor. wirh<br />
rhe r ertical column of&u-positioned numben servine as tbe<br />
dasher to rndicete Lhe mareinal disnibudon.<br />
MUI,TIFUNCTIONING GRAPHICAL !lEMTNTS 1J1
1J2 THEORY OF DATA GRA?IIICS<br />
<strong>The</strong> method ofdata-based markers for the marginal distributions<br />
suggests a further enhancement ofthe dot-dash-plot:<br />
m.3<br />
L4.6<br />
11_3<br />
70.7 .<br />
8.4<br />
Now the numben in the margin eliminate the standerd frame<br />
and everr a rangeframe, replace the coordinate ticks, show thc<br />
marginal distribution ofboth variables, and rccord the exact values<br />
of rhe two measurements made on each udt of observation. Tlis<br />
graphical anangernent performs better for smaller data sets (say 3o<br />
observations or les) and when a fine level of detail is required.<br />
Fnrally, a srriking design with data-based labels:<br />
Desiened by Carol Moor€, Corpoate<br />
Amul Reports, Inc., in Wwaltd H-dcg,<br />
Gqhn/Diastdns (zwi&, r9j6), p. 23-
Puzzles and Hierarchy in Graphics<br />
<strong>The</strong> complexity ofmultifunctioning elements can sometrmes tum<br />
data graphics into visual puzzles, crypto-graphical mysreries for<br />
the viewer io decod€. A sure sign ofa puzzle is that rhe graphic<br />
must be interpreted rbrough a verbal rather than a visual proces.<br />
- For example. despite i* clever and multifuncioning dara measure.<br />
formed by crosing two four-color grids, this is a puzzle graphic.<br />
Deployed here, in a feat oftechnological virtuosity, are 16 shades<br />
ofcolor spread on 3,oJ6 counties, a monurnenr ro a soDhisticated<br />
computer graphics system.s Bur. it is surely : graphic eiperienced<br />
verbally, nor visu:Jly. Over and ovrr. rhe viewer' musr nrn<br />
linle phrases through their minds, trying to maintain the righr<br />
paitem of words to make sense out of the visual montese: i Now<br />
let's see, purple represents courties where rhere are both hieh leveh<br />
of nule cardiovascular disese mortality aad r r.6 to J6.o p;ent<br />
ofthe households have more trhan 1.or persons per room.<br />
. . . What does thac mean anl,way? . . . And the yellow-green<br />
counties. . . ." By conrrasr, in a non-puzzle graphic, the translation<br />
ofvisual to verbai is quickiy l€amed, automatic, and implicit<br />
- so th::r rhe rrsu.al image flows ight through the verbal decoder<br />
inirially necessary ro understand rhe grephic. As paul ''Seeing<br />
Valdry wrore,<br />
is lorgetring tfie name ofrhe tlunq one sees."<br />
MUI,TIFUN ClIO NIN G G&rlPrrrcar rLlMtNTs 1J3<br />
5<strong>The</strong> .r.chnique n desdibed in Vincent<br />
P. D&bb: ed Alva L. Iintner, <strong>The</strong><br />
Utilizrion of Prim*y Printing ColoB<br />
in <strong>Display</strong>ing More rb,n One vdiatte,"<br />
iD Buceu ofthe CeNui Technical paper<br />
No. 41, Ctuphitul Pr.htdtkh of Statistiral<br />
I"Iotadtior (wasbi\st n, D.i., 1s78),<br />
14-2r. <strong>The</strong> m1p! are as6ed in Howrd<br />
Wain€r aDd C. M. Francolini, ..An<br />
Empnid Inguny Concehing Hund<br />
UndeBbnding of TwGveriable Color<br />
M,ps, Anenan Sldtutittdn, 34 (1sqo),<br />
8143-<br />
!*:':<br />
! t"'' *"
1J4 THEORY OF DATA GRA?IIICS<br />
Color often gclerates graphical puzzles. Despite our experiences<br />
with the specrnrm in science textbooks and raiubows, the rnind's<br />
eye does nor readily give a visual ordering to colors, except possibly<br />
for red to refect higher levels than orher colors, as in dre<br />
hot spots ofthc cancer rnap. Attempts to give colors an order<br />
result in drose verbal decoders and thc mumbling oflittle mental<br />
phrase.s again-indeed, even rnnemonic phrases drotl the phras€s<br />
required for graphical decodiqg:<br />
A method ofcoloring ingenious in idea but aot very satisfactory<br />
in practice was used by L. L. Vauthier. It was called thc<br />
mountain-to-the-sea method. White was uscd for thc representation<br />
ofthe greatest intensity ofdre fact because it indicated<br />
the sunrmit of a mountain with its etemal snow, irext came<br />
greer reprcsenring the forcsts farther down tlc slopcs, thcn<br />
yelJow for the grain ofthe plains, and fnally for the ninimum<br />
the blue of the waters at sea level.6<br />
Becausc drey do havc a natural visual hiemrchy, varying shades of<br />
gray show varying quantities bcttcr than color. Ten gray shades<br />
worked ef'ectively in the galaxies map:<br />
<strong>The</strong> success ofgray compared to rhe visually more spectacular<br />
color gives us a lead on how multifunctioning graphical elemena<br />
can communicate complex information without tuming into puzzles.<br />
<strong>The</strong> shades ofgray provide an ea.sily comprehendcd order to<br />
the data measures. This is the key. Central ro mainraining clarity<br />
in the face of the complex are graphical methods that organize ond<br />
order the Jlow oJ graphical infornation presanted to the eye.<br />
How can graphical architecture promote the ordered, sequenced,<br />
hicrarchical flow ofinfotrnation from the graphic to the mindi<br />
eye? How can dre data-information be arranged so that the viewer<br />
is able ro peel back layer after layer ofdata from a graphic?<br />
Multiple layers ofinformation are cre*ed by nulupb t iewing<br />
deptlu and nubQle uiewing aryles.<br />
6H, Gay F@lhotrsr, Histoiiel Dt<br />
velopncnr of thc cnphn,l<br />
tion of Sbti*io1tub," OJ'tn, 3 (<br />
126, who cites E. Cheyson, L6 D&<br />
thodes de statisnquc ghphiq@ il<br />
snion eiv6c11c dc 1878,"Jolb.ld.L<br />
Sadltl de *atnri|,e de Pdtn,I (9fi\
, Graphics can be desigftd ro have at le3st rhree viewing depths:<br />
(r) what is seen from a &stance, an overall sru.*.. *,rllly "ggr"_<br />
gated fiom an lnderlyng microsftucture; (z) wh"t is "een op1t"e<br />
and in detail, the frne structure ofthe data; aad (3) what is seen<br />
1rulicidr,.gd1lylng<br />
ttre graphic- that which i, U.ti"a th. gophi".<br />
Look at all the difi'erenr levels ofdetail created by thts popj"ion<br />
densiry map ofLhe Unired Stares, a glory ot -"a.r" ."rog.rphf<br />
prepared by rhe Bureau ofthe CersLrs rr.l, dot. .*..pr tn"url"o<br />
centers, represe[ts jcrc peoplel Note the corridors connecting the<br />
major urban complexes; the e6ec* ofJandforms on th" popJl"tion<br />
di"smbudon lrhe cennal valJey ofCrlifornia. the raJlrys'and ridges<br />
oI Appahchra. and the clu,rer. along ;ven1r ana the imalt rown'.<br />
erong rhe highways, linled like; srring ofpearls. <strong>The</strong> map arrays.<br />
Inrde-cr. lome 4oo,ooo points on its implicir grid.<br />
Differenr visualangles for differenr aspecrs ofrhe dau aLo or_<br />
ganize graphical inlorm:rion. f,ach separare line ofsrght sbould<br />
remain undranging (preGrably horizontal or verricel) as the eve<br />
watches for dara variation of the fat ofthe Iine ofsisht. For mulrivariare<br />
wor[. sereral clear lines can be crerred. nccJJ Ayres.<br />
display ofAmerican divisionr in France. Even with its comolex.<br />
inrerwo\en dr(a.th€ grrphic is nor a puzzJe. Tt ... .eg"rar. ;.*l<br />
angles make the flow ofinformation coherent: the profle of the<br />
horizon for rhe upward-moving time-series, the vertical for the<br />
composition ofthe bar, and the horizontei for each division's stay.<br />
Thus while every drop ofink serves three dillerent data dispiay '<br />
functions, each of the tluee comes to the eye v/ith irr ovrn inde_<br />
pendence and integrity.<br />
..,iiii+iE{ttitti<br />
--.**ffi-*<br />
MUITIFUNCTIONING GR4PIIICAL 3!EMSNTS 1
158 TTTEORY O! DArrl GRAllIIcs<br />
Current R€ceipts ofcovemment as a<br />
Percenage of Gros Domestic<br />
Prcd\\cr, rg7o ^nd. lgjg<br />
Belgiu<br />
1970 1979<br />
37.5<br />
34,9<br />
Ir.ly 3o.4<br />
43.2 Sdgim<br />
3J.? I$ly<br />
27.1 Sp.in
, SimiJarly. this rable-gnphic organize dau for viewing in sevenl<br />
drrecions. Tbe chart, wber read vertically. r.arr[, ,5 ."un;es by<br />
govemmmt tax collections in r97o and again in ry79, with the<br />
names spaced in proportion ro dre percenrages. Acros *e mlurrurs,<br />
rhe parred comparisons show how rhe numbers chaneed over rhe<br />
yean. <strong>The</strong> dopes are also compared by readirg d"*n ,t " - .._<br />
tjon oflines. and lines ofunuruai slope srand our 6om rle overalt<br />
upward panern. <strong>The</strong> informarion pftown is boLh integnted and<br />
separated: integrated chrough im donnecrJ conrrnr, separared iD<br />
that tLe eye follows several difiLlent and uncluttered patlu in look_<br />
ing over the &ta: :<br />
Such an analysis ofthe uieuhg archttr.tnrc of a gr:Lphic will help<br />
tn aDd<br />
-creating<br />
evrluadng designs Lhar organize complex<br />
inlormadon hierarchically.<br />
MULTITUNCTIONING GIAPIIIC LEIEMSNTS 1J9
I uant to rcach that state of tondansatioh<br />
oJ scl$ations<br />
uhith .onstitutes a pi.tute.
8 Data Density and Small Multiples<br />
Our eyes can make a remarkable number ofdistinctions within a<br />
small area. With the usc ofvery light grid lines, it is easy to<br />
locate 625 points in one square inch or, equivalently, too points<br />
in one scluare cenrimeter,<br />
Or consider how an 8o by 8o grid over a squarc inch-about 3o<br />
by 3o over a square centimeter-divides the space:1<br />
With the help ofconsiderable redundancy and conrexr, our eyes<br />
nake fine disrinctions ofthis sorr all the time. Measuremenr insrruments<br />
used in engineering, architecrural, and machine work are<br />
engraved with scales ofuo increments to thc cenrimeter and 50<br />
to the inch. Or consider the reading ofline prinr. <strong>The</strong> rype in rhe<br />
U.S. Statistual Abstrca is set at rz lines per vertical inch, with each<br />
line running at abour z3 characers per inch for a manmum density<br />
ofz76 characters per square inch. <strong>The</strong> actual density, given<br />
the white space, is in this case r85 characters per square inch or<br />
28 per square centimeter.<br />
No, 1,t50, SrE L PRoDUCB-NE1 SsrlMEMs, By MAnnr CussEs: 1960 ro 19?8<br />
.. " na4 lot eb.vhse dNjnedl<br />
co.dsbdi]eebgoB.$'|pl'iP'..<br />
T{<br />
l1g<br />
1 A squre grid formed on c,c! side by<br />
! pdald black,nd n 1 pmlel vlite<br />
lncs contains D2 iniersections of two<br />
black lines (coners ofsqlrr€r, (n 1),<br />
intenections oat*o white liDes (whire<br />
sqEret, dd,n(n 1) ineseciions ofl<br />
black ard whnc [n. (sides ofsquder,<br />
for. total of (2n-r),llne inteBediols<br />
U,S. Bueau of the Ce$us, Srd,s/kdl<br />
Ab4tdtt o.f rhe unit.d stdtes: 1979 (tv.shingron,<br />
D.C., 1979), p.822.
1O2 THIORY OI DATA GRA?IIICS<br />
Maps routinely prescnt even finer detail. A cartographer writes<br />
that "the resolving power of the eye enabler it to diflerentiate to<br />
o.r mrn where provokcd to do so. Clearly, $erefore, conciseness<br />
i.s of the essence and high resolution graphics are a common<br />
denominator of cartography."2 Distinctions at o.1 inm mean<br />
254 per inch.<br />
How many statistical graphics takc advantagc ofthe ability ofthc<br />
eye to detect large amoutrts ofinformation in small spaccs? And<br />
how much informatioa should graphics show? Let us begin by<br />
considering an empirical measurc ofgraphical pcrformancc, thc<br />
data density.<br />
Data Det$ity in craphical Practice<br />
<strong>The</strong> numbers drat go into a graphic can be organized into a data<br />
maftix ofobseffrtions by variables. TaLing into account the size<br />
of the graphic in relrtion to the arnormt of data displayed yields<br />
rhe data densitll<br />
data density of a graplnc :<br />
number ofentnes rn dau m:mx<br />
area ofdata graphic<br />
Data matrices and data demities vary enormor:sly in practice.<br />
At one extreme, this overwrought display (originally printed in<br />
five colors) presents a data natrix offour entries, the names and<br />
dre numbers for the two bars on the righr. <strong>The</strong> left bar is merely<br />
the total ofthe other two. <strong>The</strong> graph coven 20.5 square inches<br />
(r7r squarc ccntimeters),<br />
resulting in a data dcnsity of.r5 nurnbers<br />
per square inch (.oz numbers per square centimeter), which<br />
is thin indeed.<br />
,D. P. Bi.knore,<br />
"<strong>The</strong><br />
Relmc of<br />
cartogmphy," inJohn C. Davis md<br />
MichaelJ. Mcculash, eds., Dirpial dfld<br />
,4nalysn oI Spdtidl Datd (L@dot, rgj s),<br />
P 331
TOTAL<br />
PARTICIPATION<br />
DATA D!NSITY AND SMAIL MULTI?TIS 103<br />
lnadult<br />
-35<br />
30<br />
25<br />
z0<br />
Exc@dve Onice ofthe Prcsid.nt, Olice<br />
of MMgcftent and Bldget, Sd.td1<br />
Ldi.at r, rr7_r Nqrshingron, D.C.,
164 THEoRY oF DATA GRAPIIICS<br />
<strong>The</strong> exernplar ftom theJASA style sheet comes in at a lightweight<br />
3.8 numben per square inch (o.6 numbers per square centimeter)<br />
and a small data natrix of 32 enrries:<br />
ln contrast, the New York wearhet history, in this reduced<br />
version, does very weii at r8r numbers per square inch (28 per<br />
square centimeter):
An annual sunshine<br />
rccord reporrs about r,ooo numben per<br />
square nrch (16o per squarc cenrimcter):<br />
Thc visual nctaphor corresponds appropriarely ro the data ifthe<br />
image is reverscd, so that the light areas are the tines whcn the<br />
DATl Df\5I-! A\D SI1A.t vtttI?l!S rh<<br />
F.J. Mo*house and H. R. WitkinsoD,<br />
M4," dla DidCldr4 (lond.n, rhnd<br />
ednion, 1971),<br />
!!. 2+L2,13.
166 THxoRY or DATA GRA?Hrcs<br />
This map (:7 square inches, r75 square centimeters) shows che<br />
location and boundaries of 3o,ooo communes of France. It would<br />
rcqlrire at least 24o,ooo numbcrs to recreate the data ofthe map<br />
fuo,ooo latitudes, 3o,ooo longitudes, and perhaps six numbers<br />
describing the shape ofeach commune). Thus thar data densiry<br />
is nearly 9,ooo numbers per square inch, or r,4oo numbers per<br />
square cenqmcrcr,<br />
<strong>The</strong> new map ofthe galaxics locates 2,27J,328 cncoded recrangles<br />
on a two-dimensional surface of61 square nrches (39o square<br />
centimerers). Each rectaDgle represents thrce numbers (two by ia<br />
location, one by its shading), yielding a data density ofrro,ooo<br />
numben per .qu.rre incf or 1-.ooo numbers pe( rqurre cenrimerer,<br />
That is the current record.<br />
J,.$es Bertiq Se,tol,rie Crdprt ,e<br />
(?aris, sc.ond €didoq, 1e7j), !. 1j2.
Data Density atld the Size ofthe Data Matrix:<br />
Publication Pnctices<br />
DATA DlNSTTY AND<br />
<strong>The</strong> able shows the data density and the size ofthe data matrix<br />
for graphics sampled from scietrrfic and news publications. At least<br />
zo graphics from each publication were examined.<br />
Thc table records an enonnous divenity ofgraphical performances<br />
both within and betwecn publications. A fcw data-rich designs<br />
appear in nearly every publication. Thc opportunity is there but<br />
it is rarely exploited: the average published graphic is rather thin,<br />
Data Dasity and Size df,Data Mitrix,<br />
Statistiel cnphics in Sel€td Publicatiols, Cn@ 1q79 1980<br />
oruben !d squre in.h)<br />
medio rnnimum maxnrq<br />
n<br />
2L<br />
19<br />
18<br />
18<br />
t7<br />
13<br />
t2<br />
9<br />
8<br />
8<br />
7<br />
7<br />
6<br />
5<br />
5<br />
2<br />
o.2<br />
3<br />
5<br />
f<br />
5<br />
2<br />
2<br />
3<br />
1<br />
I<br />
1<br />
1<br />
1<br />
2<br />
1<br />
7<br />
1<br />
2<br />
362<br />
115<br />
754<br />
1l<br />
122<br />
lo<br />
113<br />
923<br />
5L<br />
17<br />
74<br />
39<br />
1X<br />
72<br />
t3<br />
25<br />
69<br />
23<br />
10<br />
Sir ofData MaEix<br />
m€did hinimum ruihm<br />
r77<br />
2ctr<br />
lo9<br />
0.1 1<br />
5<br />
2!J<br />
%<br />
50<br />
150<br />
29<br />
84<br />
36<br />
35<br />
32<br />
23<br />
44<br />
3a<br />
16<br />
15 3780<br />
10 1,160<br />
376<br />
2A 1aA<br />
42 156<br />
14 44A<br />
46 I&fi<br />
L5 472<br />
I 3600<br />
3 192<br />
1L 3t2<br />
a 420<br />
14 735<br />
6<br />
t4<br />
2<br />
t2<br />
580<br />
96<br />
96<br />
540<br />
14 652<br />
I 164<br />
sMAl! MULTTPL!S 107
168 TEEoRY o! DATA CRA?IIICS<br />
based on about 5o numbers shown at the rate oflo per square<br />
inch. Among the world s news?apen, the Wall Steet Jounal, <strong>The</strong><br />
T;nes (rondon), and zrali publish data-rich graphics, with data<br />
densities equal to those o{ the Jowul of the Aneican Statistical<br />
,4rsoriarron. Most ofthe Amencan papers and magazines, along<br />
with Praala, publish less data per graphic than the major papers<br />
of other industrialized countries.<br />
Very few statistical graphics achieve the information display<br />
rates found in maps. HiglJy detailed naps portray loo,ooo to<br />
lSo,ooo birs per squarr inch. For example. che average U.S.<br />
Geological Survey topographic qua&angle (meauring r7 by z3<br />
inches) is estimated to contain over 1oo million bits ofinformation,<br />
or about z5o,ooo pet square inch (4o,ooo p.r squ"r"<br />
centimeter).3 Perhaps some day staristical graphics will perform<br />
as succrssfLill) as mJps in carrling infonnation.<br />
Iligh-Information Graphics<br />
Data graphics should often be based on large rather than small<br />
data matrices and have a high rather than low data density. More<br />
information is better than less information, especially when the<br />
marginal costs ofhandling rnd interpreting ad&tional info nation<br />
are low, as they are for most graphics. <strong>The</strong> simple things belong<br />
in tables or in the text; graphics can give a sense oflarge and<br />
complex data sets that calrnot be managed in any other rnay. If<br />
the graphic becomes overcrowded (although several thousand<br />
numbers lepresented may bejust fine), a variety of&ta-reduction<br />
techniquet averaging, clustering, smoothing-can thin the numbers<br />
out before plotting.a Summary graphics can emerge from<br />
high-information displays, but there is nowhere to go if we begin<br />
with a low-information daign.<br />
Data-rich designs give a context and credibility to statistical<br />
evidence. Low-informstion designs are suspect: what is left out,<br />
what is hidden, why are we shown so little? High-density graphics<br />
help us to compare parts ofthe data by displaying much information<br />
within the view of the eye: we look at one pege at a time and the<br />
more on rhe p€e, de more efective and comparative our eye<br />
can be.s <strong>The</strong> principle, then, is:<br />
Maximize &ta density and the size ofthe data<br />
matrix, within reason,<br />
High-information graphics must be designed witb special care. As<br />
the volume ofdata increases, data measules must slrink (smaller<br />
dots for scatters, thinner lines for busy time-series). <strong>The</strong> clutter of<br />
I Mo[is M. Thompson, Ma?Jrr ,u ertd<br />
(washinston. D.C.,1e7e). p. 187.<br />
'Paul<br />
A. TGey sdjohn W- Tukey,<br />
''s@mriatior:<br />
Smoothing: Sulplemenied<br />
Views," in Vic Baben, ed.,<br />
Iiteryrcting Mdnuatidt, Ddta (Chid\6t{,<br />
lDgland, 1e32), ch. 12r dd william s_<br />
Cleveland- "Robst Loely wcighrcd<br />
Resre$ionand SmoothinsScatterplots,"<br />
Joufldt oJ &e Anqrrdn Sratkti.dl AMtiation,74<br />
(1979J, E?D 336.<br />
slt is sucgstcd in the am\ris ofx{ay<br />
frlms to "smrch a reduced imse so that<br />
the whole dirllay cm be pereived on<br />
.t lest one o.6ion without large eye<br />
novement." xdwud LlewcllF<br />
Thoms, "Advic<br />
to the Ssiclld or<br />
'What<br />
Do We TeU Tbem?" in Richrd<br />
A. Monty andjohn W. S@dds, eds.,<br />
Eye Moteftentt tud Pstclblogkal Ptur.ss!<br />
(HiUsd.re, N.J., 1926), l. 349.
charqjunk, non-data-ink, and redurdant data_ink is cven more<br />
costly than usual in data-rich daigns.<br />
<strong>The</strong> way to increase data derxity other than by arlarging the data<br />
matrix is,to lcduce the arca ofa graphic. <strong>The</strong> Shrink pn;ciple has<br />
Graphics can bc shruaL way down.<br />
Marry data gmphics can be reduced in erea to halftheir currently<br />
publrshcd nze uith rirru:ll1 no lo- rrr legrbilirl and rnforrnadon.<br />
for e),ampje. Ber rin s crisp ,nd eleganr line rllow. the jsplal 6f<br />
r7 "nrll-scJe graphics on a single pagc aJong wiLlr exren.irerexr.<br />
Repcared.applic,rion of rhe Shrinl princ pte lead-r ro a pou erlii<br />
and effecrir e gr:phicrl desrgn. the small muJnple.<br />
^tr--:<br />
ll_]<br />
ihfilifltl'ml]<br />
lMl'<br />
i 4 - -<br />
&YW+\l<br />
'E=<br />
i'<br />
--<br />
ilruulllil<br />
Fffi<br />
.l'-l<br />
)<br />
lL*./ I<br />
t3T TI rl<br />
t\h4'!h1<br />
Lrl 1l<br />
lHl'<br />
DATA DINSITY AND SMALL MUTTI?TES 169<br />
POSES PAR LES CHRONIOUES<br />
;lt:',i:ip.tT,"t:;l:ii":.* ^',''" p,. d.*<br />
co!rb
1'O TH!ORY O! D-{TA GRAPHICS<br />
Small Multiples<br />
Snall nruldplcs rcscniblc the lrames of a nrvic: a scries ofgraphics,<br />
shosng the samc conibrnaron ofvariablcs, n,dcxcd b.v chrnges<br />
in aro cr vanable. Trvenq.-tlucc hours ofLos Angclcs air pollution<br />
arc organized into drn drsp1a,v, bascd on a cornputcr gcneratcd<br />
vidco tape. Shown is the hourlv :rvcragc distribution of<br />
reactiic h,vdrocarbon emisions. <strong>The</strong> design rema,Ds corlstant<br />
tlrrough all tLc f.arnes, so drat ancution x devotcd cnrirclv to<br />
slrrfts in the data:<br />
I'.on vid.. trte b-! Grego.\ J Nl.R:.<br />
Crlilon!r Iin.nne oaTdhn.l.g!.<br />
Th. n,od.l n deribed in c.J. I4cRr.<br />
Ww. R. a;oodnr, andJ, H. S.nrf.]d.<br />
''DevelopDrent<br />
of tr Sc.ond-Cicncrro.r:<br />
Nhdr.Dr.i..li!4.dci Lr UrbrL An<br />
!ollun... L Mod.l F.mulxtior,'<br />
Ah"oijl)nr Lrrnan"Ert, 16 (198)),
<strong>The</strong>se grim small multiples show the distribution ofoccurenc€<br />
ofthe cancer melanoma. <strong>The</strong> sites ofz69 primary melanomes are<br />
recorded, along with the distribution between men and riromen.<br />
Note the data graphical arithmetic, similar to that of rhe<br />
multiwindow plot.<br />
Abb. 1. Yerteilug von t69 pnmdm l[eleodm rnj Kopf<br />
md Ea,ls<br />
Abb. 2<br />
Abl. 3<br />
Abb,2 u,3. Diff6@ierung dor MeleDon@t€itunq<br />
@h G@NtuLt€n<br />
DATA DINSITI AND SMAfL MULTIPIES 171<br />
Arihu Wistelm,<br />
"Zu<br />
Mddom€ntrtehug<br />
dtrch chronishe Li.hreinwiF<br />
k@e," Dd Eautaftt,2s (rs71), t1.
172 THXORY OI DATA GRA?IIICS<br />
<strong>The</strong> efi'ecs ofsampling errors are shown in these rz disuibutions,<br />
each based on a sample of 5o random normal deviates:<br />
j"1<br />
/.I./rJL-.-r<br />
*:]<br />
'l ,J-\1,r. Iaa.<br />
Thcsc six distributions show the age composioon ofherring catches<br />
each year from r9o8 to r9r3. A ttemcndous numbcr ofhcrring<br />
were spawned in r9o4, and that clas began to dominate the 1908<br />
catch as four-year-o1ds, then the rsos catch as {ive-year-olds, and<br />
at5676q$11 12 I11 1516<br />
1908,<br />
1909.<br />
h 1910.<br />
r+L 191'1.<br />
) l\\ 1e12<br />
/ \ re13,<br />
5t56iE9m1n6h15t6<br />
This next design compares r comglex set ofdata: shown are the<br />
chromosomes of (from left to right) man, chimpanzee, gorilla,<br />
and orangutan. <strong>The</strong> similarities between humans and the great<br />
apes are to be noted.<br />
xdmo'dA. Muphy, "One caqe?<br />
Many Catres? Thc Argumflt from the<br />
Binodal Ditiburion," Jo,',al oJ<br />
chtonit Dives,17 (re6d, 3ae.<br />
Johan Hjori. "Fluctlario$<br />
in rle Grc,t<br />
Fishoies ofNonhem E!rope," Rdr'pdt<br />
et Ptoteev,th4s,, 20 (1e14, tn Stsn<br />
S.blee, <strong>The</strong> Edge oJ e UnfMiliat Watld<br />
(New Yorl, 19?t, !.226.<br />
Jorse J. Yuis dd Om Phkash, "<strong>The</strong><br />
O.jgjn of Mdr ,{ Cnromoionrr Pntorial<br />
Legacy. ' srrfte, 21j (March 19,
:in<br />
nEn<br />
|':t-'<br />
trt<br />
lt!<br />
!tFftl<br />
rlt<br />
tii<br />
trt<br />
!!"<br />
IFi<br />
l.ts l<br />
'l!l t - -<br />
ltt<br />
t,,1 r I<br />
l l l<br />
-a:1<br />
83tr<br />
iE-<br />
III<br />
=F!<br />
iti<br />
=FF<br />
,TEB<br />
TII<br />
!." I<br />
. t - l . l<br />
l t t<br />
-rit|:<br />
i F r<br />
I F E<br />
:!lli5<br />
! i t<br />
E::<br />
! t l<br />
F ; :<br />
t t t<br />
i i t<br />
.1<br />
!!: ti<br />
l l t<br />
."it l l<br />
III<br />
r-i ul<br />
1a<br />
II<br />
T<br />
l.'l<br />
!<br />
:<br />
i:t<br />
r i<br />
t!<br />
tJr<br />
n':<br />
FI<br />
tt<br />
-fl<br />
!l<br />
0.'<br />
ili<br />
ilr<br />
FEF<br />
! i !<br />
,''r<br />
! !<br />
il<br />
rll<br />
!!!<br />
19<br />
Li!t<br />
I t t<br />
tlt-<br />
i:E<br />
8Fr I<br />
tt!<br />
F: :t t:<br />
llt<br />
-r ! t_l<br />
!::<br />
itt<br />
n:a<br />
niir<br />
!q r r.l<br />
l t l<br />
:1 r"r<br />
-Fi<br />
!!!<br />
trrl<br />
!r'fl<br />
20<br />
...1<br />
!!I<br />
il!<br />
l r'!<br />
!ti<br />
I t t<br />
! t i<br />
l t -<br />
=tt<br />
fl -, ll<br />
ttt<br />
ltt<br />
ilI<br />
tft<br />
l r t<br />
iti'l<br />
lti f<br />
itt<br />
!r!<br />
t-t l-t rl<br />
!!:<br />
TTI<br />
Itt<br />
!tl<br />
lrl<br />
'l<br />
l' :l<br />
E Et<br />
|lf- l-i -.t -.1<br />
:i- r,ii:l E=:=<br />
ii: ;iii! .i::l<br />
iE! ',: ! ii +. i ::<br />
ilr I!-:= l==:=<br />
10<br />
t.lLi!:<br />
Frl<br />
HT<br />
,ir'! lir<br />
i iii<br />
--ii<br />
16<br />
!t!<br />
2l<br />
,E:<br />
11<br />
i<br />
f ! i<br />
itE<br />
t = i<br />
!3t<br />
!-!l<br />
'17<br />
12<br />
tiE<br />
trf<br />
FE:<br />
rll<br />
l l t<br />
llr<br />
ttr<br />
F-F<br />
lrt<br />
iLlF - II<br />
':-'rri<br />
rl ! !:
174 TIIEORY OF DITA G{APEICS<br />
And, faally, a visually similar small multiple, the Con*ner<br />
Reports fteqnenctafaepair records for autonobiles built from<br />
1926 to 1981. This is a panicularly ingenious mix oftable and<br />
graphic, portralng a complex set ofcomparisons berween manufacturers,<br />
tJpes ofcan, yeer, and trouble spo6.<br />
o-ed*tu-4' o=|.futuddn o-r@ o-$!{tuao( a=tu*'ftil'ft,<br />
ooooo<br />
ooooo<br />
ooooo<br />
ooooo<br />
ooo oo<br />
ooooo<br />
ooooo<br />
ooooo<br />
ooooo<br />
ooooo<br />
ooooo<br />
oooo<br />
oooo<br />
oooo<br />
oooo<br />
aoaao<br />
ooooo<br />
ooooo<br />
ooooo<br />
oooo<br />
oaooo<br />
ooooo<br />
ooooo<br />
oooo<br />
ooooo<br />
ooooo<br />
oaooo<br />
ooooo<br />
ooooo<br />
ooooo<br />
ooooo<br />
ooooo<br />
ooooo<br />
aoooo<br />
ooooo<br />
oooo<br />
oaooo<br />
ooooo<br />
ooooo<br />
oooo<br />
oo ooo<br />
ooooo<br />
ooooo<br />
oooooo<br />
oooooo<br />
oooooo<br />
oooooo<br />
oooo<br />
ooooo<br />
aoooo<br />
oooo<br />
oooo<br />
oooo<br />
oooo<br />
oo<br />
oooo<br />
oooo<br />
oooo<br />
oooo<br />
oooo<br />
ooao<br />
ooao<br />
oooo<br />
oo<br />
o oooo<br />
oooooo<br />
oooooo<br />
oooooo<br />
oooooo<br />
oo oooo<br />
oooooo<br />
oooooo<br />
oooooo<br />
oooooo<br />
oooooo<br />
ooooo<br />
ooooo<br />
oooooo<br />
oooooo<br />
oooo<br />
oooo<br />
CoMni Retfit, a7 ( ptn1982),<br />
oooooo<br />
ooooo<br />
aaoooo<br />
oooooo<br />
oooooo<br />
oooao<br />
ooo oo<br />
ooooo<br />
ooo ooo<br />
ooaoo<br />
oaooo<br />
ooaooo<br />
oooooc<br />
oooooo<br />
oo ooo o<br />
ooooo<br />
oooao<br />
ooooo<br />
ooooo<br />
ooooo<br />
ooooo<br />
oooooo<br />
oooooo<br />
ooooo<br />
ooooo<br />
ooooo<br />
ooooo<br />
aaooo<br />
o oooo<br />
o oooo<br />
ooo oo<br />
oooo<br />
oooo
Conclusion<br />
Well-designed snull multiples ate<br />
. inevitably comparative<br />
. deftly multivariate<br />
. shruaken, highJarsity gaghics<br />
. usually based on a large d.L -"trii<br />
. drawu almost entirely with daa-in_k<br />
' often narrative in content, showing shifts in the relationship<br />
between variables as the index variable dranges (thereby<br />
revealing interaction or multiplicative effectsJ.<br />
Small multiples reflect much ofthe theory ofdata graphics:<br />
For non-data-ink, le*s is more.<br />
For data-iok, less<br />
is a borc.6<br />
DA:lA DENSITY AND SMALI, MUI.TIPLES 175<br />
6<strong>The</strong> iwo alho.isms o the nani-c of<br />
'%s" are, readivcly, cdited to iu&<br />
wis Mi6 Ie dd Rohc dd io Rohst<br />
V&tnti, Convbitf 6rt Corbedkne i^<br />
-4rdrtt 6.r (Nry YdL, s@d.diti@,<br />
tw),p.a?.
s""'"i. r r " r"'<br />
eark ,frauraliua a,, o.t"r^*'-L.-.<br />
"<br />
*.t:^- + a",al,J ^a-* a 6w.<br />
6\,L<br />
^ Ar. ^ h"s,*-'14<br />
r :x-t Li6d,4q&+<br />
Ct"J. | 'r- att )<br />
'q*-^ded@<br />
-La'1""t*<br />
c4trai<br />
ug<br />
?Rlr s,.ltttr!<br />
&rtz &a"nlu * *"*-.8--- rr'o^,t Ja"a. tuL d@ ' .X@n ^stz -t8ts.<br />
"<br />
tu)+.^ 4t a) La--q-t ,.t ",4;a :.'^* q-- Z u,a--n ,*"<br />
-t:t*;EE:*fr . t *':;:-H.::*,: ; t t x;3<br />
;q1t .atr. rLqa, Lqv. L.fae Lc.-trls -4r +4 J.rt./;* "j'* + L 4 aLL;<br />
1..-;Fan:uat.'*rn;'l;>*rlLaL*;z=::x:8.*"*^^<br />
dd&i AR;e!&d:!a<br />
4Lr.-,b?;1e-;";.t"F.-<br />
'.t ',tL*.!t-b'*\<br />
;-'<br />
- "i29,=ffi=t*t<br />
.'
9<br />
t . 1 . l a 1 '<br />
Aesthetics and Technique in Data Graphical Design<br />
Along with the anazing graphic ofthe French losses in the Russian<br />
invasion, Minard includes a second<br />
"Carte Figurative." Ir portrays<br />
Hannibal s fading eiephant campaign in Spain, Gaul, and Northern<br />
Italy. Minad uses a light transparent color for flowJines, allowing<br />
the underlying type to show through. Thh relined use of color to<br />
depict more information conrrasts with the garish tones too often<br />
seen in modern graphics.<br />
Whar makes for such graphical elegance? What accounts for<br />
the quality of Minard's graphics. ofthose of Playfair and Marey,<br />
and of some recent work, such as the new view ofthe gala-xies?<br />
Good design has cwo key elements:<br />
Graphical elegance is often found in simplicity<br />
ofdesign and complexiqr ofdata.<br />
<strong>Visual</strong>ly attractive graphics also gather power from contcnt and<br />
interpretations<br />
beyond the immediate display of some numbers.<br />
<strong>The</strong> best graphics are about the usetul and imporrant, about life<br />
and death, about the universe. Beautiful graphics do not tra{fic<br />
with the trivial.<br />
On rare occasions graphical archirecture combines rrr'irh th€ dara<br />
content to yield a uniquely spectacular graphic. Such performances<br />
can be described and rdmired but there are no easy compositional<br />
principles on how to create that one wonderful graphic in mi11ions.<br />
As Bamett Newman once said. "Aestherics<br />
is for the afist Like<br />
ornithology is for the birds."<br />
What can be suggested, rhough, are some guides for enhancing<br />
the visual quality ofroutine, workaday designs. Attractive displays<br />
of statistical infomarion<br />
. have a properly chosen format and desig.<br />
. tse words, numbers. and drawing together<br />
. teflect a balance,<br />
a proporrion, a sense<br />
ofrelevant scale<br />
. display an accesible compleity of detail<br />
. often have a narrative quality, a story to rell abour rhe data<br />
. are drawn in a profesional nanner, with rhe rechnical details<br />
ofproduction done with care<br />
. avoid content-free decomtion, including chafijunk.<br />
Cnades Joseth Minard, Tdrler,, c'e<br />
thiqu* et Catta Fiqnarius de M. Minahl,<br />
'r4J ir69, a pofifolio ofhis work held<br />
by tle Bibliothiquc de l'Ecole Nationale<br />
de Ponts et Chau$des, Paris.
178 TIIEORY OF DATA CRAPSICS<br />
<strong>The</strong> Choice of Design: Sentences, Text-Tables, Tables,<br />
Semi-Graphics, and Graphics<br />
<strong>The</strong> substentive content, extmsiveness oflabels, and volume and<br />
ordering ofdata all help determine the choice ofmethod for the<br />
display of quantitative materials. <strong>The</strong> basic structr.ues for showing<br />
data are the sentence, the table, and the graphic. Often two or<br />
three of these devices should be combined.<br />
<strong>The</strong> conventional sentorce is a poor way to show more than<br />
two numbers because it prevmts comparisons within the &ta.<br />
<strong>The</strong> linearly organized flow of words, folded over at arbinary<br />
points (decided not by content but by the happenstance ofcolumir<br />
width), otrers less than one efi'ective dimension for organizing the<br />
data. Instead of:<br />
Nearly 53 percent of the type A<br />
group did something or other<br />
compared to 46 percent ofB and<br />
dightlymore than 5 7 percent ofC.<br />
Arrange the typ€ to facilitate comparisons, as in tlns text-rablei<br />
<strong>The</strong> three groups dift'ered in how<br />
they did something or other:<br />
GrouP A 53%<br />
GroDpB 46qo<br />
CnonP C S7o/o<br />
<strong>The</strong>re are nearly always beaer sequences *lrr alphabetical-for<br />
example, ordering by content or by data values:<br />
Group B +6Vo<br />
Group A szEo<br />
Group c stVo<br />
Tables are clearly the best way to show exact numerical values,<br />
although the entries can also be arranged in semi-graphical form.<br />
Tables are preferable to graphics for many small data sets.1 A<br />
table is nearly always better than a dumb pie chart; the only<br />
wone design than a pie chart is several ofthem, for then the<br />
viewer is asked to comparc quantities located in spatial disartay<br />
both within and between pies, as in this heavily encoded example<br />
ftom an adas. Given their low daa-density and failure to order<br />
numben along a visual dimension, pie charrs should never be used.,<br />
.*-4<br />
Depanhdt of Suseys, Ministly of<br />
hlour,.4d6 d/Ivdel 0roten,<br />
19J6 ), vol. 8, p. 8.<br />
1 OD the deden of abl€s, @ A.S.C.<br />
"Rudimots<br />
Ehr6bdg.<br />
of Nmmcy,"<br />
Joutul of the Roydl Stunakal Socieq,<br />
A, 14o (!97i,277-297.<br />
,t<br />
'This loiDt is n.de decisively inJacqGs<br />
Ber.n, Aryhia dnn Qa ic Infohati@<br />
r@ari'J (3dlin, 1981). Benin dgcribes<br />
multiple pie charts s 'tompletely
Tables also work well when the data<br />
Presentation requires many localized<br />
compadsonl In this 41o-number table that I<br />
designed for the Neu Yol& Tifl$ to show how<br />
different people voted in presidential elections<br />
in the United Sates, comperisons berweefl rhe<br />
elections of 1980 and 1976 are read across each<br />
line; within-election analysis is'conducted by<br />
reading downward in the clusters ofthqee to<br />
sevm lines. <strong>The</strong> horizontal rules divide the data<br />
into topical paragraphs; the ows are ordered so<br />
as to tell an ordered story about the elections.<br />
This type of elaborate table, a supertable, is1*ely<br />
to attract end intrigue readers through its<br />
organized, sequential detail and reference-liLe<br />
quality. One supertable is far better than a<br />
hun&ed iitde bar charts.<br />
Ne yor& Tirrr November 9, 1980,<br />
!,-^ 28.<br />
ArsTrrBTrcs aND tEcHNtQuE 179<br />
How Diflerent Groups Voted for Prestdent
18o TFEoRY oF DATA GRA?Hrcs<br />
For sets ofhighly labclcd numbers, a wordy data graphic<br />
coming close to straight text-works well. This table ofnumbers<br />
is niceJy organized into a graphic:<br />
Matring Complexity Accessible: Combining rVords,<br />
Nunbers, and Pictures<br />
Explanations that give access to the richness of the data make<br />
graphics more attractive to thc viewer. Words and pictures are<br />
sometimes jurisdictional enemies, as artists feud wirh wirers for<br />
scarcc space. Arr unfortunate legacy ofthese craft-union difi'erences<br />
is the artficial separation ofwords and pictures; a few sryle sheers<br />
everr forbid printing on graphics. What has gonc wrong is drat the<br />
techniques of production instead of the information conveyed<br />
have been given precedence.<br />
Words and picturcs belong together. Viewers need the help that<br />
words can provide. Words on graphics are data-ink, naLing<br />
eft'ective use of thc space &eed up by erasing redundant and nondata-ink.<br />
It is nearly always helpful to write litde messages on the<br />
ploning iield to explain the &ta, to label oudiers and itrteresting<br />
dara points, to write equations and sometimes tables on ghe graphic<br />
itself, and to integrate the caption and legend itrto the design so<br />
that the eyc is not required to dart bacl and forth beoveen rextual<br />
material and the graphic. (<strong>The</strong> size ofqpe on and around graphics<br />
Netu Yo* Tin6,J^nwy 2,1979,p. D 1.
can be quite small, since the phnses and senterces are uiually nor<br />
too long and therefore the small tne will nor fatigue viewers<br />
the way it does in lengthy texts.)<br />
<strong>The</strong> pturcipb of data ltext integatiol is<br />
Data graphics are paragraphs about data and<br />
should be treated as such.<br />
Words, graphics, and tables ire dif'erent mechanisrns with bur a<br />
singlc purpose the presentation ofinformation. Why should the<br />
flow ofinformation be broLen up into dift'erenr places on the page<br />
because the information is pacl:ged one way or another? Somerimes<br />
ir may be useful to have multiple story-lines or multiple<br />
levels ofpresentation, but rhar should be a deliberate designjudgment,<br />
not something decided by conventional production requirements.<br />
Imagine if graphics were replaced by paragraphs of words<br />
and those paragrapls scattered ovcr rLe pages out of sequefte with<br />
the rest of the text-drat is how graphical and tabular information<br />
is now treatcd in the layout of many published pages, parrict arly<br />
in srientjfi c journrlurd profesronal book..<br />
Tables and graphics should be run inro rhe texi whenever possible,<br />
rvoiding the clunxy and diverting segregarion of "See Fig.<br />
2," (figures all too often located on the back ofthe adjacent page).3<br />
Ifa display is discused in various parr of the text, it might well<br />
be printed afrcsh near each reference to it, perhaps in reduced<br />
size in larer showings. <strong>The</strong> principle of te*/gnphic/table integration<br />
also suggess th:t the same typeface be used for text and<br />
graphic and, further, that ruled lines sepanting diferent types of<br />
information be avoided. Albcrt Bidermaa notes that illustratiorx<br />
were once well-integrated wirh rext in scientifc manuscripts, such<br />
as those ofNewron and Leonardo da Vinci, but that statistical<br />
graphics became segregated from te* and table as prinring technology<br />
developed:<br />
<strong>The</strong> evolution ofgraphic methods as an elemenr ofthe scientific<br />
enterprise has been handicapped bv therr adiuncive. seeregated.<br />
:nd margrnal po.ition. the'exigencrJs of rypog.lphy<br />
rhar_moved graphics ro a segregared position in rhe prirred<br />
wo'k ha\ e in rhe past conrribuled ro rheir inrellecruJ segregrnon<br />
and marginaliry as wcll. <strong>The</strong>re wx a correspondinq<br />
orgrnizarional segregatJon, uirh decisions on graphic' often<br />
pasrng out ofrhe hands ofthe orisinal allsi anid communicaror<br />
Into tho\e of qraphic .pecjalisr-the comrnercial arri*,<br />
and de'igncn of eraphic departrrenrs and audjo-vjsual aids<br />
shops. fiir example, whosc predilecrions and st
182 TExoRY oF DATA GRAlItIcs<br />
Page after page ofl-eonardot manuscripe have a gende but<br />
thorough integration oftext and figure, a quality rarely seor in<br />
c6uqi t car udtt* atkx tatn n blta t6r 6ii)+'<br />
lrno u ua /an'qan npu alauift qfu<br />
flt rt4u*tft Cdttd t4/Lttb /otrr. o, r-k l"to i;rt<br />
Tiin { laru ya& alla oa auhQitra.,<br />
" 'hlknfi,ary<br />
[ra
Accessible Complexity: <strong>The</strong> Friendly Data Gtaphic<br />
Allr occasional data graphic displays such care in desiqn that it is<br />
particularly accesrible and open ro rhe eye. as rfthe Jesigner had<br />
the.viewer in mjrdar every tum<br />
'fhts<br />
wLile consrructing the graphic.<br />
is the friendly data gd?hi..<br />
<strong>The</strong>re are ma:ry specifc diferences between Giendlv and<br />
unfriendly graphics:<br />
wo.ds de spelled our, tuysrGious md<br />
elabohte @codins.voided<br />
wods m &om ieft to nshq rhe<br />
Bual direction to! r@dirs occidmtal<br />
iitde nAsases beb qdain dat<br />
elalohtely osded shadings, mse<br />
batching, dd coloa are avoided:<br />
iE@4 labeli .re pl,.ed on lhe sDphic<br />
nElf, no legod ; requiEd<br />
glaphic attact! view6, provotes<br />
coloa. ifused, de.hosd s rb,fthe<br />
.olor-deficidr od olq-blind (5 to<br />
10 !d@r of vi€w6) .d make sdF<br />
ofthe gnlhi. (blue @ be disrjnguisbed<br />
from oth€r .olox by hosr<br />
olor-defi.i@t psple)<br />
tlpe is d@, precise, modesq teHdoe<br />
nay be done by hdd<br />
t]?e is uppq+nd-lowd €r. wirh<br />
abbreviatioE abourd, &quinng the<br />
n€wer ro son tuough terl ro<br />
words m veEiquy, p.nj(ul,iy aloDg<br />
lhe Y{ns: qords m in reltnl<br />
graphic is crypric, requires repared<br />
refersces to sanded rext<br />
obffe .odiugs reguite soing lack<br />
dd forth betwem leged dd galhic<br />
sraphic is iepdlenr, frlbd virh<br />
dedgn in$sitive to coloriefi€imr<br />
viems; red md green uscd for<br />
rype n doned. ovdbedng<br />
t}?e i {ll dlirals, ss sif<br />
'With regard ro typogaphy, Josef Albers wntes:<br />
AESTEETICS AND TsCIINIQUE 183<br />
<strong>The</strong> concepr rhat "rhe rimpler tbr form ofa lener $e simpler<br />
1t5 reaorng w_as aD obsesron of beginning coDvmcdvism. Ir<br />
became somethirg like a dogma. and i, .;ll foflowed by<br />
moderrusrc rypographers. . . . Ophrh:lmoloey ha, disclosed<br />
thar rhe more rhe lerrers are difrerenLiared Goml_ech other, the<br />
ersrer is rfie reading.<br />
!irhour going irro comparisons and<br />
oerrrs. tr $lolrld be retlrTed $rr word\ consistine ofonlv<br />
caprul leners presenr rhe mo* di_6cuJr readire -b.."us.'of<br />
Lheir.equal heighr, equal volLrme, and. wirh ri-osr. rheir equal<br />
wrdrh. when comprring seriIlerters with sans_seril the ljtter<br />
provrde an uneasy reading.-I'he fashionable preGrence for<br />
sans-\eril in rexr sbows Dejther hivorical noiprrcrical 6loer lJbds, Ltela.tion oJ atu (Ns<br />
HN@, 1963, Evised e&rjo 197J), p.4.
184 TFxoRY oT DATA GRAPHICS<br />
P"oportion and Scale : Lirc Weight and L€ttedng<br />
Graphical elements look better together when their relative pro.<br />
portroh 2re in balarce. An inagraced qualiry. an appropriare<br />
visual linkage between the various elements, results. This musical<br />
score ofKarlheinz Stockhausen exhibits such a visual balance:<br />
In contrast, this next design n hear,ry handed, with nearly every<br />
element out ofbalance: the ciotted ink, the poor style ofletering,<br />
the puffed-up display ofa smail data set, the coarse texture of rhe<br />
entire graphic, and the mismatch between drawing and surrounding<br />
t€xt:<br />
?o%<br />
60%<br />
a<br />
a<br />
a<br />
oa<br />
a<br />
DEMOCRATIC<br />
l a<br />
j<br />
SHARE OF<br />
\Actuolrrlult:<br />
.I<br />
SEATS<br />
D.mocrol6 r.c.lY.d<br />
a I 50.9 % rot6r,<br />
i 55.4 % 6.ott<br />
5o% ---.--f\------<br />
(5O% vota.. 50% !.ol.l<br />
aa<br />
4010;<br />
50i 60*<br />
DEMOCRATIC SHAREOF<br />
VOTE<br />
.L<br />
I
Lirres ia &ta graphics sLould be thin. One reason eighreeothand<br />
nineteenth-century graphics look so good is that tley were<br />
engraved on copper plates, with a charecteristic hair-thin lina<br />
<strong>The</strong> drafting pens of twentieth-century mechanical drawing<br />
thicLened lineworl, making it clurrsy and unattractive.<br />
An cf,ective aestLetic device is the orthogonal intencction of<br />
lines of ditrerent weiehts:<br />
Nearly every intersection ofthe lines in tlis desiga ftased on a<br />
painting by Burgoyne Ddler) iovolves lines ofdif,ering weights,<br />
and it makes a dilGrence, for the paintirg's cbaracter is diluted<br />
wittr lines ofconsant width:<br />
aBSTEEIICS IND TECSNIQUB 18J<br />
Po.tq fot th. .nibiri@<br />
"MoD&ie<br />
ed<br />
N+Plalticid in Amdie," Y.lc Univdity<br />
An Gallcy, Ocrobcr 18 to<br />
Dc
i86 TEToRY oF DATA cRA?Hlcs<br />
Likewise, data graphics carr be enhanced by thc perpendicular<br />
intenections of lines of dilfernrg weights. <strong>The</strong> heavier linc should<br />
be a data measure. In a time-serics, for example:<br />
<strong>The</strong> contrast in line wcight represcntr contrait in meaning. <strong>The</strong><br />
greater mcaning is giv€n to the greater Line weight; thus the data<br />
line should receive greater weighr than the connecting verticals.<br />
<strong>The</strong> logic here is a restatemert, in differcnt lan$rege, ofthc<br />
principle of data-ink miximization.<br />
Proportion and Scale: <strong>The</strong> Shape of craphics<br />
Graplncs should tend toward the horizontal, greater in length<br />
than height:<br />
Several lines ofreasoning favor horizontal over venical displays.<br />
First, analogy to the horizon. Our eye is naturally practiced in<br />
detecting deviations Gom the horizon, and graphic design should<br />
taLe advantage ofthis fact. Horizonhlly stetched time-series<br />
are more accessible to the eye:<br />
,,'.-'\/w
<strong>The</strong> analogy to rhe horizon also suggests that a shaded, high contlast<br />
display might occasionally be better than the floating snaLe-<br />
<strong>The</strong> shadbg should be crlm, wi$our moirC rffece.<br />
Second, ease of labeling.<br />
that read from left to right<br />
6eld:<br />
instead of<br />
lzbels<br />
Third, emphasis on causal influence. Many graphics plor, in essence,<br />
etret<br />
ft al<br />
It is easier rb write and to read worils<br />
on a horizontally stretched ploning-<br />
.- LbeL<br />
and a longer horizontal helps to elaborate the worLinqs ofthe<br />
causal variable in more detail.<br />
AlSTEXTICS AND TECFNIQUs 187
188 TExoRY ot DATI GlaPgtcs<br />
Fourd:" Tukey! couasel.<br />
Most diagnostic plos involve eithet a more or less de6nite<br />
dcpcodence that bobblcs around a lot, or a poilt spatter. Such<br />
olots are retha more often bener made !,i/; tLan iall. widerihan-tall<br />
shapes usuelly make it easicr for the eye to follow<br />
Eom left to rieht.<br />
Perbaoa tbe-most smeral euid{rce we cutr ofer is that<br />
s-ootdy-cfianging ;rves Ju stend being aller thau wide,<br />
b"t " *ilgly ciruiha& to be wid€r th; tal. . . J<br />
,bd, 6ndly, Play6iis example, Of thc 89 graphics in six differan<br />
books by-William Play6ir, most (92 percmt) ere wi
If graphia should tend toward the horizontal rather than the vertical,<br />
then how much so? A venerable (ffth-century n.c.) but<br />
dubious mle ofaesthetic proportion is tLe Golden Section, a "divine<br />
division" ofa line.3 A length is divided such that tle smaller<br />
is to the greater part as the greater is to the whole:<br />
a b<br />
-..<br />
!J<br />
b<br />
b\<br />
a*b<br />
Solving the quadratic whed e - 1 yields [, :<br />
In tum the colden Rectangle is<br />
r's +j<br />
2<br />
1616. . .<br />
= 1.618....<br />
<strong>The</strong> nice geometry oftLe Golden Recungle is not unique;<br />
Birkhoffpoins out rhat at least 6ve oth* reci:ngles (induding<br />
the square) have one simple mathematical property or another foi<br />
which aesthetic claims mieht be madc:e<br />
lmrlNul<br />
r=t4r4 t=r,618 r I t.132<br />
Plafair favored proponions beoveen r.4 aad r-8 in aboft rwothirds<br />
ofhis published graphics, with most ofthe exceptions<br />
moving oore roward the horizonral thar rhe golden prescriprion:<br />
ABSTEDIICS AND TECIINIQUE 189<br />
3<strong>The</strong> combitutio of go@t y dd<br />
mFticilm s@ding the Golilen Retangle<br />
cm !e sm in Miloutiqe Borig<br />
ellieaircl\, <strong>The</strong> Coldn Nanbd &n nE<br />
Sciclttitt Aesthltis oJ At&ite.tur G.lew<br />
Yo*, r9J8) ed Tons Btu4 7t Sdek<br />
oJ Aftieflt C.otuttf lcop&has6.<br />
ecstge D. Bi*hofi ,46rr.ri, Meerc<br />
(C@b.idge, r9l3), pp. 2F3o-
19O TrrlORY OF D.{T.{ GRA?rrrCS<br />
<strong>Visual</strong> prefercnces for rectangular proportions have been studied<br />
by psychologists since 1860, but, even given the implausible assumption<br />
that such studies are relevant to graphic design, the 6ndings<br />
are hardly decisive. A mild preGrence for proportions near<br />
to the Golden Rectangle is found anong those taling parr in the<br />
experiments, b11t the prcGrred height/lcngdr ratios also vary a<br />
great deal, ranging between<br />
And, as is neaiy always dre case in experiments in graphical<br />
perception, viewer responses were found to be highly contextdepcndent.l<br />
o<br />
<strong>The</strong> conclusions:<br />
. If the nature of rhe data suggests the shape of the graphic,<br />
follow that suggestion.<br />
. Otherwise, move toward horizontal graphics about 5o percent<br />
wider than tall:<br />
lol have relied on Leomrd zuse, zedi<br />
Pd..lrior oJFoh 0{er Yo.}, 1970),<br />
c!. ro, for a sqtuary o{ the inmenF
Epilogwe: Designs for<br />
the <strong>Display</strong> of Information<br />
Design is choice. <strong>The</strong> theory gfthe visual display ofquantiative<br />
information consists ofpdnci/les thar\generare design options and<br />
that guide choices among options. <strong>The</strong> principles should not be<br />
applied rigidly or in a peevish spirit; they are not logically or rnattrematically<br />
certain; and it is bener to violate any principle than to<br />
place graceless or inelegant marks on papbr. Most princifes of<br />
design should be_greered with some skepticism, for word authority<br />
can dominate our vision, and we may come to see only through<br />
the lerxes ofword authority rather than with out own eyes.<br />
What is to be sought in designs for the display of infomntion<br />
is the clear portrayal of complexity. Not the complication of the<br />
simple; rather the task of dre designer is to give visual acces to<br />
the subtle and the dilficult-that is,<br />
the revelation of the complex.
194 rNDXX<br />
resthetics, gBphicrl 177 r91<br />
rir lolhtion 42, 1zo<br />
,4edd,4 ( Red fl!g") 83<br />
Alder. Berni 134. 14t<br />
Andndn Edutdtiab ia<br />
Aneti.ab Palitirdl Stietue Re,iew 167<br />
Atu"dne Statistiqre de la FnnE 167<br />
Anscombe's qurnc! r3 14<br />
area d.l q@tiry 69F73<br />
dnonoDicrl ghplics 26-29,154, 166<br />
AFes, Leoerd P. 141, rJj<br />
Altes, Ri.hard E. 9J<br />
B.mlqgd, Clam Frmds 82<br />
Bmooft, T, A. j3, 14o<br />
bzr cntn 96 97, 12612a, b9<br />
Brabba, Vincmt P. iJ3<br />
Dmeit, Vic 1r4,168<br />
bcfors.ftatimeseris 39<br />
Bcniger, James R. 20, 36<br />
Bdtin, Jacqles r12, 166, 169, qa<br />
Bi.tmore, D. P. 162<br />
Bidemd, Albett D. 181<br />
lil.tehl symetry 97<br />
Biachinica er Biothytid A.td ia<br />
Bi*ho4 George D. r39<br />
Slot, WiliamJ. 16<br />
Bonner, JolD Tyld 9a<br />
boring statistics 7F8o, 82<br />
Borisadievitch, Miloutine 189<br />
Boutique Data Gnphic 113<br />
Bowfl, William c. 9j<br />
box plor 97, 123, i29<br />
Brinton, \vilad C. 112<br />
BroM. DeniF Scott 116-117<br />
B"\ikB Week 63,83,167<br />
caLdtrs, gaphical 46<br />
Cdryonid Wata Alds rls<br />
Crmplcll, Angs 147<br />
Camplell, Donald T. 7j<br />
@ret 1e2o. 47, 171<br />
dcd m,ps 16 2oj 121<br />
ca!!on, r€$.rJ. 146<br />
.ars, frequmcy ofrepair 1?4<br />
Caftesim coordinats 9<br />
.ase andetrded 37, 47,82, 1a7<br />
chemical dem@ts 1o>1oj<br />
Chemoftrces 97,142<br />
Chemofl Hdmd 136,142<br />
chi.keD, 4,j4c!oud ?3<br />
.blomosorns 18173<br />
Clev€lmd, william S. j8,168<br />
color-defici@t viewds 183<br />
ColtoD, Farnond R. 112<br />
comluter graphics 26-27, 29. 42, 1\.,<br />
116. 72e12r, 136, 153, 17a<br />
Connecti@t speding 74-?j<br />
onstant dolars 65 63<br />
Qoaner Repots 8a, \74<br />
CoDverse, Philip E. 142<br />
copler, coDducrivily ol 49<br />
Comford, Eain M. 1o9<br />
couty mals 1G2o, lJl<br />
Crotty, Wiliahj. eJ<br />
Daniel, Cdhbert rl3<br />
data-bAed labels 14e lj2<br />
data-built data-mss66 1j9 144<br />
data-ink mNimiatim 96 1oJ,<br />
12! 137,175, ra6, r:31<br />
dat. meas@6 119 1.14<br />
data/tex integhtion 18e182<br />
deodd, visul 1J3 1J4<br />
deflating money 6t 63<br />
Det Spiegel 83<br />
dsign mdation 6G1r3<br />
Dewey, ndmrd R. 1t<br />
Die Zet 81<br />
Diller, Burgoyne 13r<br />
distonion, gsphial 5J-J9
Doll. R. 47, 82<br />
dollars, consta.t 6J-68<br />
do!.lash-plot 1ll, rj6,1j9, 1J2<br />
double-fudioning labels 149F1t2<br />
Ducbam?, Mdcel j6<br />
@onomr. dat: 15, 32-:14, 33, J4-5t,<br />
61 64,7a.91j2, ro8, 126, 4a, 1J2,<br />
<strong>The</strong> Eronanbt 63,83,167<br />
edu.ad@L tesL', graphic 86<br />
Xbiobers, A. S. c. 178<br />
election graphie r47, r79<br />
electrodaphaloghm 9l<br />
elegance, graphicrl r77<br />
Uiot, T. S. ro<br />
erasnrg principles 96-100, 136<br />
*cend.e, gqphical 1l-jr<br />
Fi@terg, Stethen !. 14<br />
Finlnd, Alvr L. rt3<br />
Flanigm, William H. 8J<br />
Frd.olini, C. M. 1J3<br />
Ftunk_f|tt{Allseheine 8J<br />
Fmumeni, Joselh F. i6<br />
Frmcb commees 166<br />
Frmch winc map 2J-26<br />
fris.ly d.tt gh?hid r83<br />
fuel.cononygMphic 57-jq<br />
Imkhose., H. Gr.y ro, ?3,1j4<br />
Gab.glio, ADrotrio ?2<br />
gal.xies mp 26-22 154, 166<br />
Ccological S@ey Mps 168<br />
Goldenbers, Edie N. 1?o<br />
goYemmdt sP@ding 64 58, 1J8<br />
gsde scnool ghlhid 86<br />
Ghph ofMasicrl Parallelepipeds 68<br />
grid!, data-based r45-r48<br />
grids, white 127-129, 1j6<br />
Groth,ndwddJ- 26<br />
Gunett, Donald A. 29<br />
Hanlins, Tioothy H. 134<br />
Hamibal's @hpdgn 176 177<br />
Hayward, Roger 1o2, 116<br />
Hcrbert, George 14j<br />
Herdeg, Walter 80, 1J2<br />
hierarchy in graphiG 1jl-1t9<br />
high-informtion graplics 168-169<br />
histogr.m, lividg r4o<br />
horizon, an,logy to 18c18?<br />
Ho6e of Relrsat1tiv6 l?<br />
idmtificltion numbes 49 1ro<br />
Jatd,<br />
gmlhi6 in hiyeBiry entrece<br />
JalmeF beedc 43, 121<br />
JapareF ghphi6 8'84<br />
JASA siyle sleet ilo, 1JGiJ1, 164<br />
Joau.l oJ the Aneti.al Chmkdl<br />
Joa<br />
dl oJ Anen.6 Mdi.ll Astiation<br />
Joutnal of the Anen.q Statiai.al<br />
Asa.i4tion io. 1 So-t't, 164"<br />
Jotnal oJ Bi'logkal Chdhtry ro<br />
Jounat ofchehkal Phrn6 io<br />
JounaloJthe Rof,l Stdtini.al So.i.tf 167<br />
K.t!], Wilim L. 1r9<br />
Kellet, stai ey 9J<br />
Kouchoukos.Nicholas ro9<br />
Kuaicki, James T- 1oo, 1o9<br />
rNDtx 19J<br />
hbels on ghphics 18e182, 183, 18?<br />
Lambert, Johoo Hetuich 29, j2.<br />
L.s, Wayne 4. 129<br />
kw School Adftission Tesr,<br />
lt@ing.fim lit V.S6 rt.1r1<br />
I@do d: Vinci 131 182<br />
lits, dcf@ of ?6-?7<br />
Liley, P. E. 49, 1jo<br />
livira hi5t Dm 14o<br />
Long,Jobn rlniltoD 146<br />
Lor Angel6 smog €, 1?o<br />
bt Atg.Lt Tina p, @<br />
lua (aco 16-20, 4?<br />
Me.domld-Ros,Michr€l J6<br />
Me.Gr€€or, A.J. 112<br />
tugneric monoPole 39<br />
mail, HoE of Rc!@obtivd 3?<br />
M.lcoln, An&.w H. 84<br />
mP, F@.h wine 2J-26<br />
tuP, galdi.s 26-27, 154, 166<br />
ma!, thcmatic 16 2?<br />
ma!, Yii Chi Thu 2c2r<br />
Mey,E.J. 3r, 34-36,40, lls 116, 12r<br />
Mdon, Thoms J. i6<br />
M,ste.ton, William 3J<br />
maxihitngdata-dmiry 168<br />
maximizing data-irl 96 1o5, r23 1!?,<br />
Mccl@ghan, Willian 8,
196 rNDrX<br />
Mcculash, MichietJ. 162<br />
MccutcheoD, N. Broe rd, ro9<br />
McKay, FErlk W. 16<br />
M.Pae, G.egoryj. 42. 1?o<br />
Mcihoe{e., H.J. 56<br />
Mid em dcr Rohe, Ludwig i?5<br />
Miller Arthu H.<br />
'waEd<br />
Miller !-<br />
ro<br />
4? I<br />
Miurd, ChrlesJqclh 24 2j,19,<br />
4e4, 57, 121, 176 r77<br />
Mitchell, H, L, jo<br />
mois: vibEti@ 1o7 112<br />
Molkhous, F.J. 112, r6t<br />
Monty, Richard A. 168, 132<br />
Moscow, Nalolsnt €npaign 40 41<br />
mo@t.in-teih+e 1J4<br />
multi{imctioning elements r39FlJ9<br />
multiple viewiilg ogles 1J4 1rJ<br />
multiple viewiDg delth rJ4 rjt<br />
mdtiwindow plob 114, 1?1<br />
Mulhy, ndmon84<br />
Ptu eedibgt o.f tha Natiogl Acadenf oJ<br />
Stktres, U.5.4. tro<br />
Psl*oloeiulBllLtin 167<br />
psychological expuimdts 5J-J6<br />
qdile llot 124, 136, 139<br />
FDg+liame 13o, 116, 119,149,1t2<br />
redudat data-in! 96-100<br />
relaii@lgraphica 43-Jo<br />
Rh6ne bridge 39<br />
Rickett, &meyJ. 134<br />
Robinion, &thw H. 20, 40<br />
Robyn, Dorothy L. 20<br />
Roge$, Alm C. 112<br />
Ro$, H. Lam.e 7J<br />
bydty, goelogy t4 It<br />
$mpson, RoyJ. 8J<br />
Serf, F. L. 29<br />
Schnid, Calvin I- 112<br />
Schnid, Stanton E. i12<br />
Srieu. 39. 3t, 1o9, 1ro,167, r73<br />
S.iefte hdir.tan, 1974 6a<br />
Scientfu Ahentb So, 73, 167<br />
Sei.feld, J. H. tF<br />
semi-gDphi6 178-r8o<br />
s@d6. Jol'n W. 168, 182<br />
shape, gEphial 186 r9o<br />
Shinohm, Miyohei 32, 8t<br />
Shiskin, Juliu 38<br />
siryocl, Holy S. r13<br />
siebds,3. H. 26<br />
siegel, Jacob S. 113<br />
shall multiple 42, 48, 17eqs<br />
So.ial Iidi.dto4, t9n 163<br />
sopbitiation,shphisl 8t86<br />
Slear, Mary llanor 81, ii2, 123<br />
St,tlstir4l Abdtad oJ nE Unned stat6<br />
$@dd-.laf plot 14o<br />
Stoclhausm, Kulhdnz r3a<br />
StoLs, Don.ld E. 142<br />
sMdar Tik t lImdon)<br />
Sphe., Wylie 139<br />
63<br />
taUe-gnphic 1aj, 1t8 1J9<br />
rel4ilion gn?hi6 ?6, 31<br />
TerldiD& IrMJ. 38<br />
tq.tbooks. statisti.al grlPhics 112
Thon,s, Edwrd LLwetlyn 168<br />
Thonas, Lavche st<br />
ThoDpson, Mo.rn M, 168<br />
Tlrower, Norhan J. w. 2:r<br />
rjct rnarks iz7-r29, 149<br />
Tilling, Laura 32,45 46<br />
Thz Tin6 (Lo[dan) 8!,16j<br />
linHsies, beforqfter 39<br />
thnl scheddc al, 98, 11j 116, r21<br />
Tdte, Edward R, 29, 53, 7t 95, rg4<br />
Tukey, John W. t, r14.123,12t,<br />
129, 136, 14o, 163, 188<br />
Tulet, Pd A. 114, 16s<br />
tfrgsren, conductiriry of lJo<br />
Vcnturi, Robert i17, 119, 17J<br />
uewing ,rchit€ctue 1J9<br />
Wrire., Howard 79, r42, rJ3<br />
Waldrop, M, Mirchell j9<br />
wal sndJosndl s3, 167-168<br />
W4:hinpofl Past 62, ?o,81<br />
White,X.B.8r<br />
white grid 127-129, 136<br />
Tne foliowi.s illtsiBrio ae relrintcd 6y pcrmnsio.:<br />
crrprER 1. An$oDbCj qn2dt c] 1973 Ameri.r Sotisbcr Asc<br />
oidoD Mrp orrlc C x,6. p.J/n6 F peeb,$, r.in.eron UnF<br />
r^in. Pado em:s.o..f Lp,,rr, Dmrtd {. cum. r, .e<br />
_<br />
Americr Asoci*ion for tne Advecdot of scten.e. N* y ork<br />
Ciiy senher hi$ory 6 1931 Tho New ].ork Times Con!u,y.<br />
Tat on fftnled mail o 197J <strong>The</strong> New ].ort Tjhes Conr!{y.<br />
Mrgnetic non.pole o 1982 Anerier Asocjation for itE ^dv ce_<br />
ndt of Sciece. Bndge on Rh6ft, Bibliothdque de I tco_<br />
N"LiorJcde. PoDF at tru.treq to. AjCetc. !o,os, or Boo {!e1<br />
0 !9 4 rh. ror Ajde e. Time.. b:ed oh c:b toT \ ,eeo-yj.<br />
M.c-a,c-ljJoa,d t1 ttuFo.tc hnotag..J.p!n*o€;-. .ooJ<br />
i rus 5oos. I oado!. pf,U p! oruc, O, Srldon<br />
.or I .oDoTi.<br />
( oopeHrion "!d DAeloDn c! . p, .. Do8. o r"7o ..en.b.<br />
^rfr tu.In!. i H^p.LF 2. PryhcoF ro DdrraadB o .97s<br />
T're \.r wdl I Dei Conodb.. Dns,ng by, rv_" rqr: r ue<br />
.\ei rdr* \4.s,ne, h.. Ilel e.onoDr'.and.rc. , rq , n.<br />
p., Jit D. !-. o roTE I he Nrq<br />
rorr IIn6( ompan). Ol prie .979,1im-,rh..O,<br />
brie o<br />
ra ! <strong>The</strong> V "shiDg on Poj. p.J ph.e o'<br />
'o<br />
o L Auj.n*. WeeL, @<br />
o MCirw-rJl, tn!. aeJprr-ofoi o raToTtFSu.Ly<br />
Tim6. I oodob. ccj.e<br />
F ofo.t o .o-q <strong>The</strong> | .ohodi., { ronl<br />
d govrmror c, 1977 ].ale Uriv*sit]' ?Es. New yort budget<br />
o j nTb" \s v.rl,l.nes.onpan).st-nlljJcdo,or,by<br />
liob AJen rn P{. BtuF.oir, o .4-o ttr Lo( {ngels l.n.fl.<br />
irldrB dour r r0/3 Tb- Wrhingbn p$r pi @ of ojt dd_<br />
i,-{3 o '03. Tb- \er vorl Time, LoDos.. i H^orsp I. Viewr<br />
oo c.i Dom)_" ,o3o <strong>The</strong> Ne\ york I Da Compuy. Dre of.ry<br />
.lc o .u-o.T\e Nes Yo.t T.hs (ohpry. -;{;r",4.Et(_<br />
F@.mhJog?n 'ai 'a7r HJper a Ror. cflehdoo in c @ .uo,<br />
"-n,.bo<br />
Utu\srv PE*. pegi,orboD<br />
rie o rodT {.nehoD<br />
roln.r \ocnr A.tciri.on. Regrthrion nki o ,97o Add\ot<br />
WiLinson, H. R. r12, r6j<br />
wnson. Janes Q. 8J<br />
Wiskenmmn, Afthlr 17r<br />
Wond War I smlhic 14r<br />
Wumu, tuclard Saut 90<br />
Yn Chi Thu nap 2e2i<br />
Zitgrle, Nmcy H. 8J<br />
Zhe, leonrd 9?, 19o<br />
rNDXX 197<br />
Wtrle\ PDL.hrg I oDsnv. aegrrrior nk. o.o7o Hok,<br />
tubd fd wiB,on. o,u uErE. Kj-k Brld, priniru,<br />
uto,eNtv dd rh. ahcn, r Nrereorolog:.:l Sdi-,) di,.orr,<br />
t. J(lhdi, pro! ro_ & Ldmb,e,.um!r).<br />
"oA.i Hrvsrd dnrDg ofperod, pr.Fre ot cteDer6, -.nlr<br />
J. vcm.J bb. g,u.oe tufo, t"io V.<br />
codqd. o.o3r {h.r,d {$a:ron ro_hc Ao\seDe1, or<br />
\oft, -. Muluq.ldow ptor. paLt A. tuL.). B.B Du. t, p&. d _.<br />
I droar doo.. \ke oacJifom.r. O{!. oiptqms do Feqr..l<br />
]\ewsprler cont€nt @ 1979 ADelicm potitic, Sciftcc Asbcranon.<br />
' H{p. tr 6. Prr:llcl ! Tcr piat o .o-- Add.,on_w6.ey prb-<br />
!flrc \ ompa!) Bo\ ptor."rir,oh o r973 Arrh.rn \bri.-.ct<br />
A.o. rrioo. PLle' nsnrls o rrr AcrdeD.i. 1J6. Enprrialonu_<br />
r:bw d^hb' h.D o ro-dJoll wilcy. -H\prr,<br />
? \-em marerf<br />
old o r97: T\. loq: \Lr L b \e^,w pr- t \ oC 1Dbs.a !<br />
brs toiDe'. w:Cn.robydrodrDn, . o roTo Ao&mi. pE.s. Irsr<br />
m:p @ 0-6 PiDc.roD Ur^6i,y pts- rglo rd reoo ee, rio, s o<br />
r066John V/.1.) .HrF -p 3.rm.hiohmre\<br />
md cdthn V'lld A.ff rmal m,JDpte o ro-i Mouron "1d<br />
n Dutsonr @ i!6{ Ptrcaooo prN,. I b Chror,o-oD$.Jor8cJ<br />
Yuni,. o ,oB) Ansidb Aso-:boo ror rhe AorrD-m;rr ol<br />
\o.oe Auronob i- rrcqwDcyof Jeprir o .q8z CoJM.e^ UnioD<br />
q oe utoFd 5ubri h!. , q^pr,R o. How o.fmr s-oupr<br />
voGd irr p'eidot o o3o Tte \s vo-k TimF com;dy. roE.<br />
4n.c o .o-o <strong>The</strong> Neq yo,( T,nc( corphy. tonr;o a, \.n,<br />
oBw'na. rr.rm'le 6 ro