Population Investigation Committee

The Measurement of Population Distribution

Author(s): Otis Dudley Duncan
Source: Population Studies, Vol. 11, No. 1 (Jul., 1957), pp. 27-45
Published by: Taylor & Francis, Ltd. on behalf of the Population Investigation Committee
Stable URL: http://www.jstor.org/stable/2172508 .
Accessed: 15/12/2014 03:46

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

Taylor & Francis, Ltd. and Population Investigation Committee are collaborating with JSTOR to digitize,
preserve and extend access to Population Studies.

http://www.jstor.org

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 The Measurementof PopulationDistribution
By OTIS DUDLEY DUNCAN

The analysisandexplanation ofpatternsofpopulation areproblems

distribution
for the demographer, geographer,humanecologist,and locationeconomist.
None of thesespecialists, however,has takenresponsibility forworkingout a
comprehensive method for dealing with the subject. Demographers,in
particular,have given it littlesystematic attention, althoughthe literatureis
richin elementary descriptivematerialson populationdistribution in various
regions. This paperattempts a summary of themajortechniques of describing
and measuringpopulationdistribution, indicatingsome unresolvedproblems
ofmethodthatmightwellbe thefocusoffurther research.
The following classification
is a tentative ofmeasures ofpopulationdistribution;
it does notpurportto be exhaustive and thereis evidentoverlapping of someof
thecategories.
A. Spatialmeasures
(I) Numberand densityof inhabitants by geographicsub-divisions
(z) Measuresof concentration
ofspacing
(3) Measures
measures
(4) Centrographic
potential
(S) Population
measures
B. Categorical
(i) Rural-urbanand metropolitan-non-metropolitan classification
(2) Community size distribution
(3) Concentration
by proximity to centresor to designated sites
Withinthe compassof thispaperit is possibleto give onlybriefillustrations
undertheseheadings.
NUMBERS AND DENSITY BY GEOGRAPHIC SUB-DIVISIONS
The basic information employedin most studiesof populationdistribution
is thecensusenumeration ofpopulation bygeographic ofa country
sub-divisions
or other territorialunit. It will appear below that summarymeasuresof
distribution may depend heavilyon the areal units by which population
enumerations are tabulated. Such units-geographicsub-divisions of the total
territoryover whichdistribution is described-areof threemain types: (i)
politicalunits,(2) unitsconsistingof combinations of politicalunits,but not
themselvespoliticalentities,and (3) units speciallydelineatedfor statistical
purposesnot necessarilyconforming to politicalboundaries. International
studiesof distributiontypicallyemploycountries and combinations of countries
as units,butthereis interestas wellin themoredetailedanalysisthatcombines
intra-country withinternationalcomparisons.'
1 Glenn T. Trewartha," A Case for PopulationGeography", Annalsof theAssociation of American
Geographers,xLIII (June, 195 3), pp. 71-97; Abbott Payson Usher, " The Historyof Population and
Settlementin Eurasia ", Geographical Review,xx (January,1930), pp. II-I 32, reprintedin Demographic
Analysis,edited by JosephJ. Spenglerand Otis Dudley Duncan (Glencoe, Ill.: The Free Press, I956)
pp. 3-25.
27

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
z8 OTIS DUDLEY DUNCAN

Geographershave developedseveralmethodsof portraying numberscar-

tographicallysuchas thedotmapor mapsemploying specialsymbols to designate
placesofspecified sizes. Each ofthese,whileappropriate forparticularpurposes,
encounters whentryingto depictwithequal fidelity
difficulties thedistribution
in regionsof dispersedsettlement and the large nucleationsof population.'
The basicdifficultywithanycartographic methodis,ofcourse,thatonlyrelatively
imprecise conclusions can be demonstrated witha map.
A step beyondthe mere listingor cartographic portrayalof numbersby
geographicsub-divisions is the computation of the ratioof populationto area,
i.e., populationdensity. In view of the wide use and frequentcalculationof
densityfigures,it is surprising thatthe demographic literaturecontainslittle
fundamental methodological discussionoftheconcept. A well-rounded treatment
of the subjectwould requirea separatepaper. Here it is perhapssufficient to
indicatethatthereareseriousunresolved problemsthatarisein theuse ofdensity,
or maps to studypopulationdistribution.First,the resultsobtained
statistics
will depend,to a significant degree,on the systemof areal unitsfor which
densitiesare calculated,inasmuchas thereis no wayto assigna uniquemeaning
to thenotionofdensity in thevicinity ofa givenpoint. Second,in constructing
dasymetric or isoplethdensitymapsthereis difficulty in decidingon a suitable
set of densityintervals. Each of the proposedsolutionsto this problem-
whetherit leans towarda " mathematical " or a " functional " criterion-is
unsatisfactory in one way or another. Third,any attemptto refinedensity
figuresby basing them on " net" ratherthan " gross" area encountersa
considerable indeterminacy in the notionof " net-ness ". Most presentations
of " net" densitiesareneitherclearnorconvincing.
It wouldappearthatthereis needforcooperationbetweendemographers and
statistical
geographers in developingtechniquesforstudying density.Whereas
thegeographers have regardedthe studyof densityas, primarily, a problemin
cartographic presentation,2 demographers should be in a positionto clarify
theanalytical purposesand limitations of density
measurements.
MEASURES OF CONCENTRATION
Ordinarily,themajorinterestin studiesof populationdensityhas to do with
thevariation overa territory
ofdensity ratherthanwithjusttheoveralloraverage
densityof theentireterritory.Formally,thisproblemis identicalwiththatof
1 John W. Alexander," An Isarithmic-DotPopulation Map " xix (October,
EconomicGeography,
1943), pp. 43I-432.
2 Alexander," An Isarithmic-DotPopulation Map ", op. cit.: John W. Alexanderand George A.
Zahorchak, " Population-DensityMaps of the United States: Techniques and Patterns", Geograpbical
Review,xxxiii (July,1943),pp. 457-466; JamesA. Barnesand ArthurH. Robinson," A New Method
for the Representationof Dispersed Rural Population", Geographical
Review,xxx (January,1940), pp.
134-1 37; PrestonE. James," The GeographicStudyof Population",Chapter4 in American Geography:
Inventory editedby PrestonE. Jamesand ClarenceF. Jones(Syracuse: SyracuseUniversity
andProspect,
Press, 1954); Eugene Mather, " A Linear-DistanceMap of Farm Populationin the United States",
Annals of the Associationof AmericanGeographers,xxxiv (September,1944), pp. 173-I80; John K.
Wright," A Method of Mapping Densitiesof Population,withCape Cod as an Example ", Geographical
Review,xxvi (January,1936), pp. 103-I0i. See also, Philip M. Hauser, Otis Dudley Duncan, and
BeverlyDavis Duncan, Methodsof UrbanAnalysis: A SummaryReport,Research Report AFPTRC-
TN-56-i (San Antonio: Air Force Personnel& TrainingResearchCentre,1956), pp. 67-69.

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 THE MEASUREMENT OF POPULATION DISTRIBUTION 29

studyingthe " unevenness" or concentration of population. " Concentration",

it may be noted,is a termlike so manyin the scientificlanguage having a similar
Latin form,thathas two specificmeanings: (i) as referring to a stateor degree
of unevennessof population distributionat a given point in time,(z) as referring
to the process of increase over time in the degree of unevenness (the reverse
change, decrease of unevenness, is called " deconcentration"). Abstracting
fromthe spatialpatternof population distribution,any measureof concentration
(in the firstmeaning of the term) seeks simplyto make operationalthe notion
of the " degree of unevenness".
A device forgraphicpresentationand two index numbersassociated therewith
have been most widely used for measuringconcentration(although many other
measuresare conceivable,some of which have been seriouslyproposed). These
are (I) the Lorenz curve and Gini's " concentrationratio ", both originally
suggested for measuring inequality of income or wealth' and adapted to the
measurementof population concentration; and (z) what will be called here
simplythe " index of concentration", or more generallythe " index of dissimi-
larity",A. Figure IA illustratesthe principleon which the Lorenz curve is
constructed. Area units are arrayed in order of decreasing density. Then,

,oo Density Interval

f f ~~~(Pop.
/ Ml 2)

90.
so
8 45- 89.9

ox III E ^ /:18- 44.9

Under 18
? 60 - -1

QL

a0-

---- ..............
E
`7
20 4

0 20 40 60 80 t00 CUm *b POP

Cumulative Per Cent of Population

FIGURE I. (A) Lorenz Curve for Measuring Population Concentrationin the

United States, in Relation to DensityIntervals (County Basis): 1950. (B) Index
of Concentration,A, in Relation to Lorenz Curve.

Dwight B. Yntema," Measuresof theInequalityin thePersonalDistributionof Wealthor Income",

oftheAmerican
Journal Statistical xxviii (December,1933),pp. 423-433; MaryJeanBowman,
Association,
" A Graphical Analysis of Personal Income Distributionin the United States", AmericanEconomic
xxxv(September,
Review, 1945), pp. 607-628.

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
30 OTIS DUDLEY DUNCAN

treatingtheseunitsseparatelyor groupingthemintointervals, one computesthe

cumulative percentage of populationand of areawiththeadditionof each unit
(or intervalof units),and plotsthe severalcumulatedpercentages of area (Y,)
againstthe corresponding cumulatedpercentages of population(X,). Such a
curvewouldfollowthediagonalthroughout, inthecasewhereall unitshad equal
densities(even distribution), and would coincide with the X-axis if all
populationin the territory were concentrated at one mathematical point.
Variation betweenthese hypothetical extremesof complete evenness (no
concentration)and completeconcentration is indicatedby the degreeto which
thecurvedepartsfromthediagonal.
The concentrationratio,CR, is givenbytheformula:
k k
IO1000 CR= 2 Xi,1Yt- E X,Y l,
i=I i-I

whereX and Y are therespective cumulative percentages, and k is thenumber

of arealunits(or intervalsof units,iftheyhave beengrouped). Geometrically,
thisformulaexpressesthe area on the graphcontainedbetweenthe cutveand
thediagonalas a proportion of theentireareabelowthediagonal. The Lorenz
curveand concentration ratiohave been studied,as measuresof population
distribution,mostintensively byWright, who,however,considersonlythecase
in whichall unitsareofequal areasize,an unnecessary and one which
restriction
virtually
precludes ofthesemeasures
application to actualdata.1 Wrightdiscusses
also some indexnumbers,otherthan CR, that dependon the shape of the
Lorenzcurveas wellas thedegreeto whichit deviatesfromthediagonal.
The indexof concentration, A, algebraically is simplythe maximumof the
setof k valuesof (Xi - Y,). Geometrically, it is themaximum verticaldistance
fromthediagonalto thecurve,as illustrated in FigureiB. An alternative method
of computation is givenby Hoover,2perhapsthe firststudentto employA as
anindexofpopulation concentrationextensively.Ifxi andy,aretheuncumulated
percentages of populationand arearespectively, then
Ik
A=-E~Xl-y,l
2 Z

or simplythe sum of thepositive betweenthe two percentagedis-

differences
tributions.
theeffect
Incidentally, ofgrouping beforecomputing
intervals
unitsintodensity
CR or (usually) A is to give a slightlylower value than that obtainedin
computation fromungroupeddata. It is evidentintuitively, thatthe Lorenz
curve encompassesa slightlylargerarea when drawnwith manyshortline
segments thanwhendrawnwithonlya fewlong ones. When A is computed
fromgroupeddata,the same value is obtainedas forungroupeddata onlyif
is themeanoroveralldensity
thelimitofoneoftheclassintervals oftheterritory.
1 JohnK. Wright," Some Measuresof Distributions,AnnalsoftheAssociation
ofAmericanGeograpbers
xxvii (December,I937), I77-2II.
2 Edgar M. Hoover, Jr.," InterstateRedistribution
of Population,i850-I940 ", JournalOfEconomic
I (November,
History, 1941), I99-205.

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 THE MEASUREMENT OF POPULATION DISTRIBUTION 31

For mostpurposes,however,a half-dozen intervalsor so are usuallysufficient

forcomputingCR.
Fromthegeometric relationships
ofCR and A to theLorenzcurveofpopula-
tionconcentration,
it can be deducedthat
A < CR ? 2 A- 2.
Thus if A =o 5 (so%) thenCR is at leasto 5 and can be no greaterthan0 75.
OftenCR approximates the averageof theseextremevalues. For example,
inFig. i, withA =o0*587,zA - A 2=o 83o,andtheaverageis o *7o8,as compared
withthe actual value, CR=o 73o. For comparisonsof situationsof widely
varyingconcentration, eitherindexservesaboutas wellas theother,thoughthey
wouldnot,of course,rankcloselysimilarsituations exactlythesamein all cases.
Inasmuchas A canbe computedwithoutarraying arealunitsin orderofdensity,
it is usuallythesimplermeasureto compute.
It can be shownthatbothCR and A can be expressed in termsof indexesof
the dispersionof the densitiesof the areal units(the meandifference and the
mean deviation,respectively).Hence a measureof concentration can be
regarded as a measureofthedispersion ofunitdensitiesabouttheoveralldensity.
This beingthe case, the earlierremarksabout the mathematical indeterminacy
ofthedensity conceptapplyas wellas to measuresofconcentration.
Perhapsthe best way to appreciatesome of the problemsof interpreting
measures ofconcentration is to examine
indexnumbers computed underalternative
conditions.Table i showsindexesofconcentration (A's) fortheUnitedStates,
I900 to 1950, forvarioussystems ofdividingthecountry intosub-areas.These
data serve to bring out some importantcharacteristics of the index of
concentration:(i) The value of theindexis, in general,directly relatedto the
numberofarealunitsintowhichtheterritory is sub-divided,or inverselyrelated
to the averagesize of the units. A morerigorousstatement is thefollowing:
An indexcomputedfora givensetofarealunitscan be no largerthantheindex
computedfora set whichcomprisessub-divisions of the firstset. Thus, the

Table i. Indexesof Population

Concentrafion, on theBasis of Counties,
computed
Economic and Geographic
Sub-regions, for theUnited
Divisions, States: 9pootoq9o.

County Economic Sub- Geographicdivi-

Year basis regionbasis sion basis

I950 ... ... ... ... 58 9 50 I 39.2

I 940 ... ... ... ... 59*I 49 2 40 5
I930 ... ... ... ... S 59 49*I 40? 8
I920 ... ... ., 538 48 3 4I.2
I9I0 ... ... 52 8 48 7 4I16
I900 ... ... ... .. * 5X6 43 6

* Not computed.
Source: United StatesCensus publications.

indexbased on countiesmustbe at leastas greatas the indexbased on sub-

ofcounties),and willbe greater
regions(combinations ifthereis anyunevenness
by countieswithinsub-regions.(z) Consequently,
of distribution the index

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 32 OTIS DUDLEY DUNCAN

provides no unique answer to the question of what degree of population

concentrationcharacterizesa territory. Any index value must be considered
relativeto the systemof sub-areas on which it is based. (3) Moreover, the
index does not give a unique answerto the question of whetherthe unevenness
is increasingor decreasing(thesecond meaningof "concentration",
of distribution
noted above), as is evident fromthe fact that the series in Table i change in
contrarydirections.
Although comparisonsof concentrationindexes are subject to these sources
of ambiguity,it does not follow that they are empiricallymeaningless. For
example, subject to qualification upon more intensive study, the changes
exhibitedin Table i mightbe given the followinginterpretation.The decreasing
concentrationindexesbased on geographicdivisionsreflectthe spread of popula-
tion over the continentalarea of the United States,a continuationof the historical
patternof settlement fromeast to west and the openingup of regionsto intensive
occupancy. The increasing indexes based on counties indicate urban and
metropolitanconcentrationof population, and the arrestingof the increase
afterI940 may reflectlocal deconcentrationof population within metropolitan
areas. The comparativelyslightchanges shown by the indexes for sub-regions
may representa balancingof the two tendenciesof regionaldeconcentrationand
metropolitanconcentration,since many sub-regionscontain both metropolitan
centresand extensivenon-metropolitan areas. If,in fact,thechangesin population
distributionover the last half-century have followed the complex patternjust
suggested,it would, of course,be impossibleforanysingleseriesof indexnumbers
to describe those changes adequately. The contraryresults obtained with
alternativeindexes may then reflecta basic ambiguityinherentin any concept
of concentrationthatdoes not specifythe systemof areal unitsto which it refers,
ratherthana defectin the operationaldefinitionof the measureof concentration.

MEASURES OF SPACING
An approach to the analysisof distributionthat is closely related to density
and concentrationmeasures is the measurementof the spacing of population
units. Theoretical contributionsto this problem are due primarilyto plant
ecologistsworkingon the spacing of membersof speciesin the plantcommunity'
and to the geographersand economistswho have developed hypotheticalsystems
of economicareas on theassumptionof even populationdistribution.2However,
the most directapplicationof the idea of spacingto the representation
of popula-
tion distributionwas developed by Barnes and Robinson3 in theirtechnique of
the " linear distance map ". This type of map is recommendedby the authors
for displaying variations in density for relativelydispersed populations. It
involves a transformation of densityby the formula: Average distance= *i i /
VDensity. In their illustrationof the technique,densityis taken as number
of farmhousesper square mile,and the average distanceof a farmhouseto its six
1 Philip J. Clark and Francis C. Evans, " Distance to Nearest Neighbor as a Measure of Spatial
xxxv (October, 1954), pp. 445-453.
Relationshipsin Populations", Ecology,
2 August Losch, TheEconomics ofLocation(New Haven: Yale UniversityPress, I954).
3 " A New MethodfortheRepresentation of Dispersed RuralPopulation", op. cit.

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 THE MEASUREMENT OF POPULATION DISTRIBUTION 33

nearest neighbours is deduced on the assumption that farmhousesare evenly

distributed. A correction of the formulais given by Mather,' replacing the
constant I II by I .07.2 In both these papers it is indicated that values of
average distancecomputed by formulaagree ratherwell with observed average
distances separating farmhouses,even in areas where the distributionof farm
houses appears somewhatirregular.
Evidently,the lineardistancemap commendsitselfprimarilyon impressionistic
grounds. Since the quantitydepictedis a simplefunctionof densityit can contri-
bute nothingto the clarification of ambiguitiesinherentin the conceptof density.
But it is perhaps true that a linear measure is more " understandable" than a
ratioof populationto area, which is somewhatdifficult to visualize.
Whereasthe techniqueof lineardistancemaps employstherelationshipbetween
spacing and densitymainlyas a device for improvingcartographicpresentation,
otherstudieshave been concernedwith measuresand patternsof spacingin their
own right. Clarkand Evans state: " The patternof distributionof a population
of plants or of animals is a fundamentalcharacteristicof that population, but
it is a featurethat is extremelydifficultto describe in precise and meaningful
terms".3 Plantecologistshave devoted considerableeffort to devisingappropriate
tests of the randomness of a distributionin space. Recent studies continue
this interest,but endeavour as well to specifydegree and patternof departures
fromrandomness. Some of the resultsof this work will be indicatedbriefly.
For a given universe of territorycontainingn units (individuals or defined
groups of individuals)of population, let ri be the lineardistance of the ith unit
to its nearestneighbouringunit,irrespectiveof direction. The r, are measured
for the entirepopulation of units, or a random sample thereof. Let p be the
densityof population units,area being measuredon the same scale as is used in
determiningthe r5. If r is the mean of the r5,it can be shown that-rE=0-5 P
is the expected mean in an infinitelylarge random distributionof density p.
The observed mean, 1A) varies below this,to a theoreticallower limit of zero,
as distributionsbecome more " clumped " or " aggregated". In the limiting
case each unit is contiguous to at least one otherunit. The observed mean may

vary above rE to a limit of I .0746/Vp, which occurs for a perfectlyuniform

distributionin which each unit is equidistantfrom 6 other units. The ratio,
R, of actual to expected distance,or rAIrE, may thus vary fromzero to 2 *149I,
with a value of unityoccurringfor a random distribution. A significancetest
forthe departureof R fromunityis available.4
In studies of human populationsindividualswould, of course, be found in a
patternof aggregationto at least the household level. With the methodreferred
to, it would be possible to make an exact test of the assumptionof Barnes and
Robinson and Mather that farm households in the United States tend to be
1 " A Linear-DistanceMap of FarmPopulationin the UnitedStates", op. cit.
2 A more precise value of this constantis I0746.

3 " Distance to Nearest Neighbor as a Measure of Spatial Relationshipsin Populations", op. it.
P. 445.
4 Ibid.

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
34 OTIS DUDLEY DUNCAN

spaceduniformrly. No doubttherewouldbe considerable regionalvariationin

thevalue of R. The writerhas experimented a littlewithmeasuresof spacing
of townsand citiesin two regionswheretherewas reasonto believethata
tendency to uniform spacingwouldevidenceitself.Valuesof R approximating
different
I.4, and significantly fromunity,werefoundforsettlements in Iowa
and in the northern two-thirdsof Indiana,takingas unitsthe urbanizedareas
and urbanplacesoutsideurbanizedareas. No doubtin otherregionsa tendency
towardaggregation of communities, ratherthanuniformity of spacing,would
be discoveredby thistechnique. Whilethisexerciseproducedonlyverygross
results,owingto theuse ofroughmeasurements on a smallscalemap,it suggests
thatthetendency towarduniformity is notnecessarily a functionofthecloseness
of spacing,forrA was aroundI 3 milesin Indianaas comparedwithi8 milesin
Iowa.
One can imaginepatterns of distributionin whichthetechniquedescribedi$
oflittlevalue. Foronething,itfailsto distinguish" clumped" patterns inwhich
each unithas but one close neighbourfrompatterns in whichthe numberof
unitsperclumpis largerandvariable. ClarkandEvansindicatesomeelaborations
ofthetechnique thatwouldbe helpfulforsome,butprobablynotall,suchcases.
It seemslikelythatthemeasurement of spacing,as appliedto humanpopulation
units,willprovemoreusefulfordetecting tendencies towarduniformity (which,
as Losch indicates,are of considerabletheoretical importance)than for dis-
criminating amongpatterns of aggregation.

CENTROGRAPHIC MEASURES
Owingto itspopularization bytheU.S. BureauoftheCensus,one ofthebest
known" centres ofpopulation" is themeanpoint,centroid, or centreofgravity,
i.e., " thepointupon whichthe UnitedStateswould balance,if it werea rigid
plane withoutweightand the populationwere distributed thereonwitheach
individualbeingassumedto have equal weightand to exertan influence on a
centralpointproportional to his distancefromthatpoint"-1 A measureof
dispersionaroundthe centreof gravityhas been proposed2and criticized ;3
butitseemsto havebeenusedverylittle,ifat all,in empirical studies.
Two othermeasuresof centraltendency in arealdistributionshave received
considerable attention.The " medianpoint" is definedas the intersection of
two orthogonallines,each of whichsplitsan area into two partswithequal
numbers ofinhabitants.The locationofthispointis to somedegreeindeterminate
inasmuchas it varieswiththerotationof theaxesusedin itscalculation.Lines
parallelto theseaxes maybe calculatedto dividethe distribution into fourths
in each direction. Their pointsof intersection are knownas " quartilides".
The principle can be extended to " decilides", " centilides", etc., yielding a
setofpointsthatdescribes ofdispersion
thedegreeandpattern ofthepopulation
1 U.S. Bureau of the Census, I950 Censusof Population, Vol. ii, Part I (Washington: Government
PrintingOffice,I953), p.9.
2 D. WeltyLefever," MeasuringGeographic Concentration by Means of the StandardDeviational
Ellipse ", American Journal xxxii (July,I926), 88-94.
ofSociology,
3 P. H. Furfey," Note on Lefever's ' StandardDeviational Ellipse' ", American JournalofSociology,
xxxIII (July,I927), pp. 94-98.

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 THE MEASUREMENT OF POPULATION DISTRIBUTION 35

overtheterritory.1 A somewhat conceptis thatofthe" mediancentre"

different
or moredescriptively, pointof minimumaggregatetravel. This is the point
fromwhichthe sumof the linearradialdeviationsis a minimum.It has been
discussedby Quinn as an optimumlocationin his " hypothesisof median
thepointofminimum
location.f"2 Methodsareavailableforcalculating aggregate
travelwheretravelto thecentreis alongrestrictedroutesratherthanby shortest
airlinedistance.3
It shouldbe notedthatthe locationof the centreof gravityis affectedby a
changein the positionof any unit of the population,whereastherecan be
considerablemovementwithoutaffecting the medianpoint. Hart suggests
that theirmathematical propertiesrenderthe centreof gravitymost useful
forstudyingthe areal shiftsof a distributionover time,the medianpointfor
comparingdifferent at the same time,and the mediancentrefor
distributions
investigating locationaloptimaforcentralized
services.

POPULATION POTENTIAL
As was indicated,the centreof gravityis computedon the assumption that
eachindividualexertsan influenceproportional to his distance
fromthe central
point. The term" influence " here,of course,is construedin the sense of
mathematical weight. But investigators concernedwith social influenceas
affected
by space have pointed out thatinfluence is moreprobablyan inverse
thana function
direct of distance. Professor Stewart's commenton thecentre-
this
concept4expresses pointforcefully:
of-gravity
The factis thatthispopulationcentreof gravityis principally
bureaucratichocus-pocus.
It is nonsenseto computea centreon a basissuchthat" theleverageofa dozenpersons
in California,at the long end of the teeter-totter,could conceivablycounter-balance
IOO persons in New York, the short end of the teeter-totter". . . . There is no evidence
fictitious
whatsoeverthattheimpossibleteeter-totter's has anyspecificsociological
fulcrum
meaning. Thereis on thecontrary a wealthof evidencethatpeople exertmoreinfluence
close at handthanfaraway,and thata thousandpeople a hundredmilesaway exertabout
thesameinfluence as fivethousandpeople fivehundredmilesaway.
If the" influence" of each individualat a pointis considered to be inversely
proportionalto his distancefromit, thenthe totalpotentialof populationat
a point,L0,is thesumof thereciprocals of thedistances of all individuals
in the
populationfromthepoint. In practice,of course,thecomputation is madeby
assumingthatall theindividuals withina suitablysmallareaareequidistant from
Lo, whence,
n p
Potentialat Lo== D,'
iD1'

1JohnFraserHart, " CentralTendencyin Areal Distributions", Economic Geography,xxx (January,

pp. 48-59; E. E. Sviatlovskyand Walter Crosby Eells, " The CentrographicalMethod and
I954),
Review,
", Geographical
RegionalAnalysis xxvii (April,I937), pp. 240-254.
2 JamesA. Quinn, " The Hypothesisof Median Location ", American Review,viii (April,
Sociological
1943), pp. I48-I56.
3 D. E. Scates and L. M. van Nortwich," The Influenceof RestrictiveRoutes upon the Centerof
Minimum Travel", Metron,
Aggregate xIII (I937), pp. 78-81.
4 Taken froma letterto the New YorkTimes,October 7, Ig9.

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 36 OTIS DUDLEY DUNCAN

wherethePi are thepopulationsof then areasintowhicha territory is divided

and theDi are therespective distancesof theseareasfromLo (usuallymeasured
fromthe geographiccentreor fromthe approximate centreof gravityof the
population,in each area). Aftercomputingpotentialfora numberof points
such as Lo, it is possibleto obtainvalues for otherpointsby interpolation;
or one may construct isometricmaps showinglineswhichare loci of points
havingequal potentials.'
It maybe observedthatthenotionofpotential ata pointis inprinciple
perfectly
precise,whereasthe conceptof densityat a pointis meaningless and thatof
density in thevicinity
ofa pointis ambiguous. Whilethedensity ofanyportion
of a territory dependsonly on the numberinhabiting thatportion,potential
at anypointdependson thedistribution of populationovertheentireterritory.
The two conceptsarerelated, however,in thattheconfiguration ofequipotential
linesis determined by thepatternof variationof densityovera territory;one
could say the same,of course,withrespectto theinterrelations of thevarious
othermeasuresof distribution thathavebeendescribed.

3
<~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

~~~~~~~~~~~~~~~3
_

FIGURE 2. Isolines of Population Potential,United States: I950.

Figure a is a map showingisolinesof populationpotentialfor the United

States. The principaldeterminant of the configuration
of the equipotential
lines is the massiveconcentration
of metropolitanpopulationin the Middle
AtlanticStates,butthecontours
areelongatedto theWestbythehighlyurbanized
1 For details see John Q. Stewart," Empirical MathematicalRules Concerningthe Distribution
and Equilibriumof Population", GeographicalReview,xxxvii (July,I947), pp. 46I-485; reprintedin
DemographicAnalysis,editedby Spenglerand Duncan, op. cit.,pp. 344-37I,

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 THE MEASUREMENT OF POPULATION DISTRIBUTION 37

zone stretchingalong the GreatLakesto Chicago. New York City,wherethe

peak potentialoccurs,maybe regardedas the " centre" of the countryfrom
the standpointof demographicinfluence, grantedthe assumptionson which
thismapis constructed.
Figurez maybe comparedwithFigure3, whichshowsisolinesof aggregate
traveldistancefor the same underlying populationdistribution.This index

250DSATS > \ \- t s5

FIGURE 3. Isolines of Aggregate Travel Distance, United States: I950.

reachesa mimimum, i.e., the "mtediancentre" occurs,at a pointlyingwithin

the lowestcontourshown. At thatpointaggregatetraveldistanceis around
92 billionmiles; thisis the totaltravelrequiredto move everyindividualin
thecountry tothatpointbyshortest airlinedistance.The travelperpersonwould
be somewhat over600 miles.
To thewriter'sknowledgeit has notbeen observedpreviously thatpotential
and aggregatetraveldistanceare closelyrelatedconcepts. The value of either
at a point,Lo, can be expressedby the formula, Z P,D7, wherem= -I forthe
potentialcomputation and m=I forcalculating aggregatelineartraveldistance.
The two shouldbe regardedas alternative butnotmutually exclusiveconcepts;
anditmaywellbe thatvaluesofmotherthan?: I will be foundsuitableforcertain
problems.1
1 Theodore R. Anderson, "Potential Models and the Spatial Distributionof Population", Papers
theRegionalScienceAssociation,
and Proceedings, II (I956), pp. I75-I8z. It is a curiositythatpopulation
potentialis emphasizedby an investigator(Stewart)workingat PrincetonUniversitynear the point of
maximumpotential,whilethe" hypothesisof medianlocation" was setforthbyan investigator (Quinn)
at the Universityof Cincinnati,in close proximityto the point of minimumaggregatetravel. It is left
to speculateon thepossible"demographicdetermination"
forstudentsof Wissenssoaiologie ofdemographic
concepts.

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
38 OTIS DUDLEY DUNCAN

RESIDENTIALCLASSIFICATION
The measuresof distributionconsideredthus far may be considered" purely
spatial" measures,i.e., theycan be computedwithoutreferenceto any qualitative
or classificatoryinformation on theareal sub-divisionsof the universeof territory.
In the other main class of devices for describingdistributionare found those
that depend on a prior classificationof areas as to size, location, rural-urban
character,or the like. The measure of spatial distribution,in effect,becomes
a summarystatisticdescribingthe frequencydistributionof the population with
respectto such a classificationscheme.
The most familiarclassificationused in describingdistributionis the division
of the population into rural and urban residents. It would require a sizeable
paper even to summarizesketchilythe variationsin definitionof thesecategories
among various censuses.' Probably the most frequentlyused criteriaof urban
statusare size and legal or administrative of communitiesor political
classification
sub-divisions. Thus the major componentof the urban population in the U.S.
Census has always consistedof incorporated(legal criterion)places exceedinga
specifiednumberof inhabitants(size criterion). To this core have been added
certain other populations specially designated as urban; for example, in the
I950 Census, inhabitantsof unincorporatedplaces of z,500 or more (boundaries
of which were delimitedaccording to geographicalcriteria)and inhabitantsof
portionsof urbanizedareas not classifiableas urban exceptin termsof proximity
to largerpopulationcentres. The importanceof modifyinga primarily administra-
tiveconceptof urbanpopulationis indicatedbythesize ofthepopulationincluded
underthe supplementary criteria: in I950, when 64 -o?/% of the U.S. population
was urban,53 44% residedin incorporated
placesof 2, 500 inhabitants
or more,
while 6- 60% were residentsof separatelyidentifiedunincorporatedplaces of this
size or partsof urbanizedareas otherthan incorporatedplaces of z, 500 or more.
However defined,the proportionof the population residingin urban territory
is undoubtedlythe most widelyused measureof population distribution,owing
to itsreadyavailabilityin manycensuses. The ubiquityof themeasurein regional,
national,and temporalcomparisonsperhaps accounts for the tendencyof many
demographersto treatit as an aspect of population " composition" ratherthan
distribution".
The categories," urban" and " rural", are so broad that sub-categoriesof
each need to be introducedinto any refinedanalysis of distribution. In the
United Statesthe breakdownof ruralpopulationinto " farm" and " non-farm"
residentshas been widely used for the last three decades; but it has become
increasinglyclear that the rapid growthin the " rural-non-farm" sector of the
populationis not interpretableas " rural
" expansion,but moreprobablyrepresents
"suburbanization " of the urban population. Changes in the meaning of
qualitativelydefinedcategoriesof populationdistributionare thusto be expected
when the spatialpatternof distributionundergoeschange.
1 A detailedaccount of the complicatedhistoryof rural-urbanclassifications
in the United States is
given by Leon E. Truesdell," The Developmentof the Urban-RuralClassification in the United States,
I874 tO I949 ", Current
PopulationReports,SeriesP-z3, No. I (Washington: U.S. Bureau of the Census,
1949).

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 THE MEASUREMENT OF POPULATION DISTRIBUTION 39

Cutting across, though not independent of, the rural-urbanclassification,

is the classificationof areas by metropolitanstatus. Although,again, operational
definitionsof the classificationvary, they generallyinclude rules for assigning
to a metropolitancentrecertainterritorycontiguous to it and thoughtto share
in its metropolitancharacter. The designationof metropolitancentresis inevita-
bly a somewhat arbitrary matter,the criteriaof metropolitanstatususually being
stated in terms of size and spatial separation from centres of larger size.
Conceivably, functionalcriteriafor metropolitanstatus could be developed,'
but their routine use in statisticalcompilations would present considerable
difficulty.The following figures(estimatedpopulation in millions for I956)2
illustrate the relation between urban-ruraland metropolitanclassifications:
Total ...plt ... ... ... I64 3
Metropolitan ... ... 96*z
Urban ... ... ... ... 79*4
Rural ... ... ... ... 17 43
Non-Metropolitan ... ... 68*I
Urban..- ... ... ... 24. S
Rural ... ... ... ... 43 .6
Such a tabulationbrings out a facet of population distributionpatternsthat is
obscured by a simple rural-urbanor even a community-sizeclassification:
the clusteringof tributarypopulation around major centresthat is characteristic
of highlyurbanizedcountries.
COMMUNITY SIZE
Anotherfairlycommon practiceis to sub-divideurban population (and rural,
forthat matter)by size of community. Demographic differences by community
size withinthe broad urban categorymay oftenbe as importantas gross rural-
urban differences.3As in the definitionof urban places, the community-size
distributionwill be, in considerablepart,a functionof proceduresfor compiling
the data. A paramount problem is whether to treat each politically distinct
urban place as a separateunit in the distribution,or to regarda clusterof such
places as a single unit, as in the United States Census definitionof " urbanized
area " The distributionsin Table z show what a wide difference in commumty-
size distributionis produced by changingfromone to the other of these bases.
A community-size distributionmay be summarizedwith a varietyof statistics.
One such statisticthat has been proposed for income distributionsmightwell
be adapted to this purpose. The " equatorial communitysize " is calculated
by interpolatingfor the size of communitythat divides the population in half.
Thus, the data in Table z show that one-halfof the U.S. population in I950
lived in places of approximately8,8oo inhabitantsor more, on an urban place
1 Rupert B. Vance and Sara Smith, " MetropolitanDominance and Integration", Chapter 6 in
The UrbanSouth,editedby RupertB. Vance and Nicholas J.Demerath(Chapel Hill: University of North
CarolinaPress, '954).
2 " Civilian Population of the United States,by Type of Residence,March 1956 and April 1950 ",
Current PopulationReports,SeriesP-20, No. 7I (Washington: U.S. Bureau of theCensus,1956).
3 Otis Dudley Duncan and AlbertJ. Reiss, Jr.,Social Characteristics
of Urbanand Rural Communities,
19f o (New York: John Wiley & Sons, 1956).

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
40 OTIS DUDLEY DUNCAN

Table z. Population by Site of Community,

Distribution on
Compiled
Two Bases,for the UnitedStates: 197o.
Urban place basis Urbanizedarea basis
Size of l
community Number % of total Number % of total
of places population of places population

United States,total .. ... i8,548 IOO-O I7,2I7 IOO 0

Urban. ... 4,74I 614-0 3,410 640 0

x,ooo,oooor more ... ... 5 II.5 12 251I
500,ooo to I,000,000 .3 ... I3 6i I3 5.8
250,000 tO 500,000 ... 23 5 -5 24 5.8
I00,000 to5o,ooo . ... 65 6*3 70 7* 2
50,ooo toIOO,000 ... ... I26 5 9 59 2X9
25,000 to 50,000 ... ... 252 5.8 172 3 9
IO,000 to 25,000 ... ... 778 7 9 547 5 5
5,ooo to IO,OOO ... ... I,I76 54 908 4.I
2,500 to s,0oo ... ... ... I,846 4.3 i,605 3 7
Under 2,500t ... ... ... 457 0*4 -

Other urbant ... 49 -

Rural ...3 ... ... ... ,807 36- 0 13,807 36 0

I,OOO tO 2,500 ... ... ... 4,158 4 3 4,158 4 3
Under i,ooo (incorp.) ... ... 9,649 2.7 9,649 2.7
Otherrural? 290... ... 29 o29 o

* Includes 38 urbanizedareasand 2i urbanplaces.

t Incorporatedplaces insideurbanizedareas.
t Unincorporatedpartsof urbanizedareas.
? Includesunincorporatedplaces of underi,ooo, as well as otherruralpopulationnot in nucleated
settlements.
Source: 1950 Censusof Population,Vol. i, NumberofInhabitants, Tables sa and 5b (Washington:
GovernmentPrintingOffice,I952).

basis,or in citiesof approximately 28,300 or moreon theurbanized area basis.'

The difference betweenthese two resultsemphasizesagain the relativity of
measuresof distribution to the systemby whichthe statisticsare compiled.
The equatorialcommunity size is not to be confusedwiththe mediansize of
community.The lattercouldalso be computedfora distribution of townsand
citiesbysize,butnotfora distribution likethatin Table a inwhicha considerable
proportion of thepopulationis not actuallyclassified by size of community and
couldnot be describedas livingin nucleatedcentres. The equatorialsize does
not dependon the lowerlimitof the size classification, except,of course,that
thislimitmustbe low enoughforhalfthepopulationto fallabove it. On the
otherhand,the mediansize will varydirectly withthe lowerlimitof the size
distribution of citiesand towns,usuallyfallingclose to it, inasmuchas small
communities greatly outnumber largeones.
Severalinvestigators have been intrigued withthe observation thatthe size
distribution of citiesand townsexhibitsa kindof regularity fromplaceto place
and overtime. For anylargeterritory in whichurbanization is at all advanced
it is almostuniformly foundthatsmallplaces outnumber largeones, and the
size distribution is in generalhighlyskewed. Auerbach2is creditedwiththe
Interpolatingon the logarithmof communitysize.
Felix Auerbach, " Das Gesetz der Bevolkerungskonzentration
2 Mitteilungen,
", Petermanns LIV

(19I3), pp- 74-76,

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 THE MEASUREMENT OF POPULATION DISTRIBUTION 41

firstpresentation of a formulato summarize the distributionof citiesby size.

He observedthatwhenthecitiesof Germanywererankedin orderof number
of inhabitants theproductof a city'srankand itssize tendedto be moreor less
constant.Moreover, hesuggestedthattheaveragevalueofthisproduct, expressed
as a proportion of the nationalpopulation,affords an indexof populationcon-
centrationsuitablefor comparingdegreesof urbanization among countries.
If the " rank-sizerule" were to hold precisely, the productof rankand size
wouldbe equal to thesize of thelargestcity. An averageof theproductsmay,
therefore, be regardedas an " adjusted" or " estimated " sizeof thelargestcity.
Hence,Auerbach'ssuggestedcoefficient comesdownto takingthe" adjusted"
sizeofthelargestcityas a proportion ofthenationalpopulation.
In a briefdiscussionof Auerbach'swork,Lotka1indicatedthatforthe IOO
largestcitiesin the UnitedStatesthe " law of urbanconcentration " was more
precisely (rank) 0.93 x size=constant. Lotka stated,however: " It may be
leftan open questionhow muchsignificance is to be attachedto thisempirical
formula".
Other earlytreatments of city-sizedistributionformulm apparentlywere
producedwithoutknowledgeof Auerbach'sand Lotka's work. Goodrich2
notedthatthenumberofcitiesabovea givensizelimittendsto be equal to twice
thenumberofcitiesabovea limitoftwicethatsize,a formula whichis equivalent
to the " rank-sizerule". Saibantesexperimented withfitting a Paretocurve
to the city-sizedistribution
and presented an analysisof concentration indexes
analogousto Gini's indexesof incomeinequalityor concentration.Gibrat4
suggestedfittingthe city-sizedistribution with a log-normalequation,and
proposedan indexofconcentration relatedthereto.
Both the " rank-size " and the Pareto-curve approachhave been further in-
vestigatedin thelasttwodecades. Singer6drewattention to theparallelbetween
thecity-sizedistributionand Pareto'slaw of incomedistribution and presented
evidence of goodness-of-fit of the Pareto curveto a numberof city-size
distributions.Allen6recently reviewedand considerably extendedSinger'swork.
The " rank-sizerule" wasmuchpublicized byZipf7whoevidently beganhiswork
withoutreference to Auerbach'sand Lotka'spriorstudies. Mathematical as well
as empiricalcontributions ofthe " rank-size
to theliterature rule" weremadeby
Stewart8and Hammer.9These investigators dealtwith the generalization of
1 AlfredJ.Lotka, Elements ofPhysical
Biology(Baltimore: Williams& WilkinsCo., I925), pp. 306-307.
2 Ernest P. Goodrich. " The StatisticalRelationshipbetweenPopulationand the City Plan ", in The
UrbanCommuniy, editedby ErnestW. Burgess(Chicago: Universityof Chicago Press,I926).
3Mario Saibante " La Concentrazionedella Popolazione ", Metron, vii (I928), pp. 53-99-
4 R. Gibrat,Les Inegalitis
Economiques (Paris: Librairiedu Recueil Sirey,I93I).
5 H. W. Singer," The ' Courbe des Populations : A Parallel to Pareto's Law ", Economic Journal,
XLVI (June,1936), pp. 254-263.
6 G. R. Allen, " The ' Courbe des Populations,': A FurtherAnalysis", Bulktinof the Oxford

Universit.Institute
ofStatistics, xvi (May-June,I95)' pp. I79-I89.
George KingsleyZipf, NationalUnityand Disunity(Bloomington,Ind.: The PrincipiaPress,Inc.,
I941); HumanBehaviorand thePrinaipleof Least Effort(Cambridge,Mass.: Addison-WesleyPress,
Inc.,1949).
8 " Empirical MathematicalRules Concerningthe Distributionand Equilibrium of Population
op. cit.
9 Carl Hammer, " Rank Correlationof Cities and Refinement
", mimeographedreport,Bureau of
Applied Social Research,Columbia University,195 I.

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
42 OTIS DUDLEY DUNCAN

therulein whichtherankof a cityis allowedto have an exponentotherthan

unity,andpresented formulaforestimating thenumber andaggregate population
of citieswithinspecifiedsize limitsfromthe parameters of the fittedcurve.
Anotherrecentcontribution is Madden'sdiscussionof thetemporalpersistence
" relationship
of the " rank-size as evidenceof stabilityin the growthpatterns
of cities.' Aside fromthe empiricalinvestigations incompletely listedhere,
a numberofwriters havecommented on thesignificanceofthe" rank-size rule"
fromthe standpoint of thegeneraltheoryof populationdistribution and space-
economy.2
It is curiousthatone findsintheliterature littlediscussionoftherelativemerits
of the " rank-size " and Pareto-curve approachesto summarizing the city-size
distribution.The mathematical relationship betweenthetwo formula:may be
indicatedas follows:
Let x= size of cityor town(numberof inhabitants) and y=y(x) = numberof
placesofsizex or larger(rank).3 Thentheformula fortheParetocurveis logy-
log A - a log x, ory= Ax-, oryxa-=A, whereA and a are constants deter-
minedfromthedata. The generalized " rank-size " formulais log x= log B
- i log y,or x =By-, or xyP- B, whereB and ,8are constants determined from
the data. These equations implythat ,B=i/ctor c=ilfl, and A=BI1P or
B= Al/at. However,ifA and othavebeendetermined byminimizing thesum
of thesquaresof theresidualsin logy whileB and ,Bhave beendetermined in-
dependently by minimizing the sum of the squaresof the residualsin log x,
theforegoing wouldnot,in general,holdprecisely.
relationships
The bulkoftheworkon fitting " rank-size " or Paretocurveshasbeencarried
out somewhatinformally.In the majorityof cases, constantshave been
determined graphicallywithoutany statedcriterionof fit,or the data have
simplybeen plottedon double-logpaperand theirapproximation to a straight
line has been noted by visual inspection. One findsin the literature little
discussionof mathematical for
criteriaand techniques fitting curves to city-
size distributions,and, correlatively,littleformalattention to goodnessof fit.
Singerand Allen presentcalculationsof " errors" and " averageerrors" in
computingthe numberof citiesin varioussize groupsfromthe equationof
thefitted curve. However,comparisons basedon suchcalculations areinvalidated
to a degree,by variationfrom case to case in the number and size of intervals.
In no case,to thewriter'sknowledge,has a formal test of goodnessof fit,such
as x2, beenpresented. One suspects that in the majority caseswherethefit
of
a
has been judged acceptablygood, stringent mathematical test would detect
statistically departures
significant from the mathematical model. One consequence
of the " rank-size"approach has been to focus attention undulyon the fitof
theformula at theveryupperendofthedistribution.Thus,in numerous verbal
discussionsthe writer has encountered the statement that the "rank-size ".
1 Carl H. Madden, " On Some Indicationsof Stability in theGrowthof Citiesin the UnitedStates",
Economic andCulturalChange,iv (April, 1956), pp. 236-252.
Development
2 For example,Losch, The EconomicsofLocation,op. cit.; RutledgeVining," A Descriptionof Certain
SpatialAspectsof an Economic System", Economic Development and CulturalChange,in (January,1955),
pp. 147-195; WalterIsard, LocationandSpace-Economy (New York: JohnWiley & Sons, Inc., i956).
3 Singer and Allen let y (x) = number of places larger than x.

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 THE MEASUREMENT OF POPULATION DISTRIBUTION 43
ruledoes not hold forsuchand sucha countryon the groundsthatits largest
cityis muchtoo largeor too small. The netimpression is thatscepticshave
oftenrejectedthe possibilityof a simpleformulaadequatelyfittingcity-size
distributions on thebasisofan examination of onlya smallpartof theevidence
whereasproponents of one or anotherformulamayhave been guiltyof failing
to giveas muchattention to negativeas to positiveevidence.
One mightwellbe inclinedto adopta fairly liberalcriterion ofwhatconstitutes
an acceptablefitofa curveto thedataifon sucha criterion numerous independent
instancesof acceptablefitwere to be discoveredand if instancesof poor fit
could be explainedsystematically ratherthan on ad hocgroundsdiffering for
each instance. The generality of anyproposedformulainvolvesat leastthree
considerations:the numberor proportionof countriesor regionswherea
good fitis obtained,theoccurrence ofa good fitin varioustimeperiodsand the
oftheformofthesizedistribution
stability overtime,andtherangeofcommunity
sizesforwhicha good fitis obtained. On thelastpoint,virtually all investigators
recognizethatthelinearformula(in the logs) can be to
expected hold onlyfor
sizes above a certainminimum. This minimum, however, mayvaryspatially
and temporally;moreover, may it sometimes fallwithin the rangeof available
data and thusbe approximately or
identifiable, mayit be lower thanthe size
forwhichfrequencies are tabulated. Both incautious statements and unjustified
speculationon this matter are to be found in the literature; however, Allen'
presents careful empirical estimates of the minimum size for a number of
countries.On the generality of " rank-size" and relatedformulas over space
and time,it mustbe recognizedthatthe fundof availabledata constitutes a
quite biased sample of countries and periods. Moreover, the comparability
oftheavailabledatais greatly impaired byvariation incensusmethods ofrecording
and tabulating community size. The largestcollectionof instances is presented
by Allenwho,however,workedmainlywithdatareadilyavailablein secondary
sources. It is evidentthatcasesareto be foundwheretheParetoor " rank-size "
formulafitsquite poorly,many,thoughnot all, of whichcan be " explained
away" by reference to deficiencies in thedata. Intuitively, one feelsthatthere
are certainkindsof " countries " and, perhaps,certain" abnormal" periods
in whichsuch a formulawould not be expectedto fitwell. However,until
an adequatetheoretical rationalization forthe formulais available,it is difficult
to organizethesehunches.
A carefulappraisalof thetheoretical significance of theParetoor the " rank-
size" rulewould probablyassignit a positionmidwaybetweentwo extremes:
on the one hand,a merelyempiricalcuriosity, and, on the other,a " law "
rigorously deducedfroman acceptedtheoretical schemeand verified underfully
specifiedconditions. In otherwords,such a rule has " plausibility " as well
as a modicumof empiricalsupport. A " plausible" connectionbetweenthis
typeof regularity and his doctrineof thehierarchical organization of economic
regionswas set forthby Ldsch2and restatedby Hoover.3 Moreover,Vining
1 "' The ' Courbe des Populations', A FurtherAnalysis", op. cit.
2 TheEconomics ofLocation,op. cit.,pp. 431-438.
2 Edgar M. Hoover, " The Conceptof a Systemof Cities: A Commenton RutledgeVining'sPaper ",

Economic andCulturalChange,iII (January,I955), pp. I96-I98.

Development

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
44 OTIS DUDLEY DUNCAN

has suggestedthe possibilityof developinga relevantmodel based on the

assumption" that some processof developmentleads in the limitto stable
distributionsof the sortthatwe have described ".1 As yet,however,none of
theseideas has been pursued farenough to yield a theoryof community-size
that
distribution predicts what conditions are both to
necessaryand sufficient
give riseto the Paretoor " "
rank-size form. Under the there
circumstances, is
indicated
definmitely a need forconcurrent effortsto discovertheempiricallimits
of validityof these rulesand attemptsto formulate in theoretical
termsthe
relationshipofthecity-size to otheraspectsofpopulationdistribution
distribution
and ecological organization.Pendingthese achievements, the Pareto curve
may be accepted as a convenient means of summarizing community-size
distributionsundercertaincircumstances.

OTHER CATEGORICAL MEASURES

largenumberof measuresof distribution

An indefinitely can be generated
arealunitson somequalitative
byclassifying basisand tabulating
or quantitative
the numberand proportion fallingin each class. For example,
of inhabitants
Hawleyand Bogue, followingMcKenzie,tabulatedthe populationof counties
lyingwithinapproximately fiftymilesof the seaboardand GreatLakes in the
2 A trendtowardincreasing
UnitedStates. concentrationwithinthiszone was
observed from I900 to 1930, with a decrease occurringbetween I930 and I940.
Hagood studiedthe distribution of rural-farm
populationby classifying it
accordingto size of largestcityin each county.3 A numberof investigators
have been concernedwithpopulationdistribution accordingto distancefrom
majorurbancentres.
Ordinarily to examinenotonlytheproportions
itis desirable ofthepopulation
fallingin the severalcategoriesof such a classificationscheme,but also the
aggregateland area occupiedby the populationof each category. It is then
possibleto computedensitiesand to studytheirpattern.Severalinvestigators,
forexample,have studieddensitygradients accordingto distancefromcentres.4
It is alsopossibleto generalizetheLorenzcurveandconcentrationratiotechnique
foruse in thisconnection.Unitareasmaybe ordered,ratherthanby,density
(as previously described),byanyvariablethatis ofinterest.The resulting
curve
is no longerconstrained to be concaveupwardtowardthe diagonal,but may
followan irregular path,even going above thediagonalin whole or in part.
The concentration ratio, computedby the formulaalreadystated,is then
interpreted as a measureof concentration withrespectto thevariableby which
areasare ordered. It will take on negativevaluesif halfor moreof the area
1 " A Descriptionof CertainSpatialAspectsof an Economic System", op. cit.,p. I 85.
2 Amos H. Hawley and Donald J. Bogue, " Recent Shiftsin Population: The Drift toward the
Metropolitan District,
I930-40 ofEconomic
", Review Statistics,
XXiv(AugUst, 1942), 143-I48.
3 Margaret JarmanHagood, " Rural Population Characteristics ", Chapter Iz in Carl C. Taylor,
etal., RuralLifein theUnited States(New York: AlfredA. Knopf, I949).
4 Colin Clark, " Urban Population Densities", Journal of theRoyalStatisticalSociety,cxIv (No. 4,
I95I), pp. 490-496; Amos H. Hawley, The Changing Shape of MetropolitanAmerica: Deconcentration
sinceI920 (Glencoe, Ill.: The Free Press, I956); Donald J. Bogue, The Structure of the Metropolitan
Community (Ann Arbor: Universityof Michigan,I949).

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions
 THE MEASUREMENT OF POPULATION DISTRIBUTION 45

enclosed by the curve and the diagonal is above the diagonal. If, for example,
areas are orderedby distancefromcentres,the concentrationratio may be inter-
pretedas an " index of centralization".1
More elaborateproceduresforstudyingdistributionsresultfromcombinations
of two or more of the techniquesdescribed. Bogue's unusuallycomprehensive
investigationinvolved the computation of densities (or " habitationratios")
for urban and rural population separatelyaccording to distance to nearest
metropolitancentre,size of centre,geographicregion,and proximityto highways
and to non-metropolitanurban centres.2 In this study the focus of attention
was on spatial gradientsin density,as affectedby the other attributesunder
consideration; but it mightwell have included measuresof concentrationalso.
The limitsto the possibilitiesof this kind are practical,ratherthan conceptual,
in nature. The ingenuityof futurestudents of population distributionwill
doubtlessproduce a varietyof new techniquesof analyzingdistributions.
In conclusion, it should be recognized that the study of population
distributionhas wide ramificationsand connections with other branches of
demography. If distributionsof different classes of the populationare compared,
it is possible to develop useful measures of segregationand areal association.3
Not onlypopulationcomposition,but also populationdynamicshas an important
distributiveaspect. One importantphase of the studyof migrationis, of course,
the investigationof its impact on patternsof population distribution. The same
can be said of differential fertilityand mortality,while, on the other hand,
importantclues for the study of differentials are turned up in researchon the
variationin levels of the vital processes fromplace to place. But the purpose
of thispaper is not so muchto emphasizethesewell known distributionalaspects
of general demography,as to suggest the advisabilityof making population
distributionitselfthe subjectof concertedresearchefforts.

I Hauser,Duncan, and Duncan, Methods of UrbanAnalysis,op. cit.

2 TheStructure of he Metropolitan
Community, op. cit.
3 Otis Dudley Duncan and BeverlyDuncan, " A MethodologicalAnalysisof SegregationIndexes ",

American Sociological xx (April,I95 5), pp. 2IO-2I7;

Review, " ResidentialDistributionand Occupational
Stratification
", AmericanJournalofSociology,LX (March,I95 5), pp. 493-503 .

This content downloaded from 128.235.251.160 on Mon, 15 Dec 2014 03:46:46 AM

All use subject to JSTOR Terms and Conditions