You are on page 1of 23
The Law of Anomalous Numbers Author(s): Frank Benford Source: Proceedings of the American Philosophical Society,

The Law of Anomalous Numbers Author(s): Frank Benford

Source: Proceedings of the American Philosophical Society, Vol. 78, No. 4 (Mar. 31, 1938), pp.

551-572

Published by: American Philosophical Society

Stable URL: http://www.jstor.org/stable/984802 .

Accessed: 08/02/2014 11:22

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 support@jstor.org.

.

information about JSTOR, please contact support@jstor.org. . American Philosophical Society is collaborating with JSTOR

American Philosophical Society is collaborating with JSTOR to digitize, preserve and extend access to Proceedings of the American Philosophical Society.

http://www.jstor.org

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

THE LAW OF ANOMALOUS NUMBERS

FRANK BENFORD

Physicist,ResearchLaboratory,GeneralElectricCompany,

Schenectady,NewYork

(Introducedby IrvingLangmuir)

(ReadApril22, 1937)

ABSTRACT

It has beenobservedthatthefirstpagesofa tableof commonlogarithms showmorewearthando thelastpages,indicatingthatmoreusednumbersbegin

withthedigit1 thanwiththedigit9.

takenfromwidelydivergentsourcesshowsthatthereis a logarithmicdistribution

offirstdigitswhenthenumbersarecomposedoffourormoredigits. Ananalysis
ofthenumbersfromdifferentsourcesshowsthatthenumberstakenfromunre-

latedsubjects,suchas a groupofnewspaperitems,showa muchbetteragreement witha logarithmicdistributionthando numbersfrommathematicaltabulations

or otherformaldata.

viduallyare withoutrelationshipare,whenconsideredin largegroups,in good agreementwitha distributionlaw-hence thename" AnomalousNumbers." A furtheranalysisofthedatashowsa strongtendencyforbodiesofnumerical data to fallintogeometricseries. Iftheseriesis madeup ofnumberscontaining

threeormoredigitsthefirstdigitsforma logarithmicseries. Ifthenumberscon- tainonlysingledigitsthegeometricrelationstillholdsbutthesimplelogarithmic relationno longerapplies. An equationis givenshowingthefrequenciesoffirstdigitsin thedifferent ordersofnumbers1 to 10,10 to 100,etc.

A compilationofsome20,000firstdigits

Thereis herethe peculiarfactthatnumbersthatindi-

The equationalsogivesthefrequencyofdigitsinthesecond,third

-place

ofa multi-digitnumber,and it is shownthatthesamelaw appliesto reciprocals.

Therearemanyinstancesshowingthatthegeometricseries,orthelogarith-

miclaw,haslongbeenrecognizedas a commonphenomenonin factualliterature andintheordinaryaffairsoflife. Thewiregaugeanddrillgaugeofthemechanic, the magnitudescale of the astronomerand the sensoryresponsecurvesof the psychologistareall particularexamplesofa relationshipthatseemsto extendto all humanaffairs.The Law ofAnomalousNumbersis thusa generalprobability law ofwidespreadapplication.

PART I: STATISTICAL DERIVATION

OF THE LAW

IT has been observedthat the pages of a muchused table of commonlogarithmsshow evidences of a selectiveuse of the natural numbers. The pages containingthe logarithms of the low numbers1 and 2 are apt to be morestained and

frayedby use than those ofthe highernumbers8 and 9.

Of

PROCEEDINGS

OF THE

AMERICAN

PHILOSOPHICAL

VOL. 78, NO. 4, MARCH 1938

SOCIETY,

551

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

552

FRANK BENFORD

course,no one could be expectedto be greatlyinterestedin the conditionofa table oflogarithms,but the mattermay be consideredmoreworthyofstudywhenwe recallthatthetable

is used in the buildingup of our scientific,engineering,and

general factual literature. There may be, in the relative cleanlinessofthe pages of a logarithmtable, data on how we thinkand how we react whendealingwiththingsthat can be describedby means ofnumbers.

Methodsand Terms Beforepresentingthedata collectedwhileinvestigatingthe

possibleexistenceofa distributionlaw that applies to numer- ical data in general-,and to randomdata in particular,it may be wellto definea fewtermsand outlinethemethodofattack. First,a distinctionis made betweena digit,whichis one

of the nine natural numbers1, 2, 3,

whichis composedofone or moredigits,and whichmay con- tain a 0 as a digitin any positionafterthefirst. The method ofstudyconsistsofselectinganytabulationofdata thatis not too restrictedin numericalrange,or conditionedin some way too sharply,and makinga countofthe numberof times the

natural numbers 1, 2, 3,

decimalpointor zerooccursbeforethefirstnaturalnumberit

is ignored,forno attentionis to be paid to magnitudeother

than that indicatedby the firstdigit.

The Law ofLarge Numbers An effortwas made to collectdata fromas manyfieldsas possible and to include a varietyof widely differenttypes. The types rangefrompurelyrandomnumbersthat have no relationotherthan appearingwithinthe covers of the same magazine,to formalmathematicaltabulationsthat admit of no variationfromfixedlaws. Between these limitsone will recognizevarious degreesof randomness,and in generalthe titleofeach line of data in Table I willsuggestthe natureof the source. In everygroupthe count was continuousfrom thebeginningto the end, orin the case oflongtabulations,to

a sufficientnumberof observationsto insurea fairaverage.

9, and a number,

9 occur as firstdigits. If a

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

THE LAW OF ANOMALOUS NUMBERS

553

The numberscountedin each groupis givenin thelast column ofTable I.

PERCENTAGE

DIGITS

T itle

TABLE I

OF TIMES

THE

NATURAL

NUMBERS

1

TO

9 ARE

USED

AS FIRST

IN NUMBERS,

AS DETERMINED

 

BY 20,229OBSERVATIONS

 

First Digit

 
 

_

_

_

_

_

_

_

-

_

_

_ -

_

_

_ -

-~ount

1

2

3

4

5

6

7

8

9

C

A

Rivers,Area 31.0 16.4 10.7 11.3

7.2

8.6

5.5

4.2

5.1

335

B

Population

33.9

20.4

14.2

8.1

7.2

6.2

4.1

3.7

2.2 3259

C

Constants

41.3

14.4

4.8

8.6

10.6

5.8

1.0

2.9

10.6

104

D

Newspapers 30.0

18.0

12.0

10.0

8.0

6.0

6.0

5.0

5.0

100

E

Spec.Heat

24.0

18.4 16.2

14.6 10.6

4.1

3.2

4.8

4.1

1389

F

Pressure

29.6

18.3

12.8

9.8

8.3

6.4

5.7

4.4

4.7

703

G

H.P. Lost

30.0

18.4

11.9

10.8

8.1

7.0

5.1

5.1

3.6

690

H

Mol. Wgt.

26.7 25.2

15.4 10.8

6.7

5.1

4.1

2.8

3.2 1800

I

Drainage

27.1

23.9

13.8

12.6

8.2

5.0

5.0

2.5

1.9

159

J

AtomicWgt. 47.2

18.7

5.5

4.4

6.6

4.4

3.3

4.4

5.5

91

K

n-, i/n,*.**25.7

20.3

9.7

6.8

6.6

6.8

7.2

8.0

8.9 5000

L

Design

 

26.8

14.8

14.3

7.5

8.3

8.4

7.0

7.3

5.6

560

M

Digest

33.4

18.5 12.4

7.5

7.1

6.5

5.5

4.9

4.2

308

N

Cost Data

32.4

18.8

10.1

10.1

9.8

5.5

4.7

5.5

3.1

741

0

X-RayVolts 27.9

17.5

14.4

9.0

8.1

7.4

5.1

5.8

4.8

707

P

Am. League

32.7

17.6

12.6

9.8

7.4

6.4

4.9

5.6

3.0

1458

Q

Black Body

31.0

17.3

14.1

8.7

6.6

7.0

5.2

4.7

54

1165

R

Addresses

28.9

19.2

12.6

8.8

8.5

6.4

5.6

5.0

5.0

342

S

n',n2

n!

25.3

16.0

12.0 10.0

8.5

8.8

6.8

7.1

5.5

900

T

Death Rate

27.0

18.6

15.7

9.4

6.7

6.5

7.2

4.8

4.1

418

Average.

.

.

.30.6

18.5

12.4

9.4

8.0

6.4

5.1

4.9

4.7

1011

Probable Error 4-0.8

+0.4

-0.4 +0.3

+-0.2 +-0.2 +-0.2 +-0.2 +0.3

At the foot of each column of Table

I the average per-

given for each firstdigit, and also the probable

centage is

errorof the average. These averages can be betterstudied if the decimal point is moved two places to the left,making the sum of all the averagesunity. The frequencyoffirstl's is thenseen to be 0.306, whichis about equal to the common logarithmof 2. The frequencyof first2's is 0.185, whichis slightlygreaterthan the logarithmof 3/2. The difference here,log 3 - log 2, is called the logarithmicintegral. These resemblancespersistthroughout,and finallythereis 0.047 to be comparedwithlog 10/9,or 0.046.

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

554

FRANK BENFORD

The frequencyof firstdigits thus follows closely the logarithmicrelation

Fa

= log (a

+

),

(1)

whereFa is the frequencyof the digit a in the firstplace of used numbers.

TABLE II

OBSERVED AND COMPUTED FREQUENCIES

Natural

Number

Observed

Logarithm

Interval

Observed

Prob. Error

Number

Interval

Frequency

- Computed

ofMean

1

1 to 2

0.306

0.301

+0.005

?t0.008

2

2

to 3

0.185

0.176

+0.009

?40.004

3

3

to 4

0.124

0.125

-0.001

-0.004

4

4

to 5

0.094

0.097

-0.003

?40.003

5

5

to 6

0.080

0.079

+0.001

?40.002

6

6

to 7

0.064

0.067

-0.003

?t0.002

7

7

to 8

0.051

0.058

-0.007

4t0.002

8

8

to 9

0.049

0.051

-0.002

?t0.002

9

9 to 10

0.047

0.046

+0.001

?t0.003

There is a qualificationto be noted immediately,for Table I was compiledfromnumberscomposedin generalof four,five-and six digits. It will be shownlater that Eq. (1) is a distributionlaw forlargenumbers,and thereis a more general equation that applies when consideringnumbersof one, two significantdigits. If we may assumethe accuracyofEq. (1), we thenhave a probabilitylaw ofthemostgeneralnature,forit is a probabil- ity derivedfrom"events" throughthe mediumof theirde- scriptivenumbers;it is not a law of numbersin themselves. The range of subjects studied and tabulated was as wide as timeand energypermitted;and as no definiteexceptionshave ever been observed among true variables, the logarithmic law forlarge numbersevidentlygoes deeperamong the roots ofprimalcauses thanournumbersystemunaidedcan explain.

FrequencyofDigitsin theqthPosition The second-placedigits are ten in number,for here we musttake 0 intoaccount. Also,in consideringthe frequency

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

THE LAW OF ANOMALOUS NUMBERS

555

Fb of a second-placedigit b we must take into account the digit a that preceded it. The logarithmicintervalbetween two digitsis nowto be dividedintotenpartscorrespondingto

the ten digits0, 1, 2,

berand b be theseconddigit;thenusingthecustomarymean- ing of positionand orderin our decimal systema two-digit numberis writtenab, and the nextgreaternumberis written ab + 1.

9. Let a be the firstdigitofa num-

The logarithmic interval between ab and ab

+

1

is

log (ab + 1) - log ab, whilethe intervalcoveredby the ten possible second-place digits is log (a + 1) - log a. There-

forethe frequencyFb of a second-placedigit b followinga first-placedigita is

F Fb== log Og

(

ab?+ )/l1

ab

,1og

+

a

(2)

As an example,theprobabilityFb ofa 0 followinga first-place 5 in a randomnumberis the quotient

Fb = log0/

log .

It followsthat the probabilityfora

digitin theqthpositionis

lg Fb = -3abc log

1

abc

abc

abc

abc

p (q +1)

pq

o (p+1))

oIp

p

Here the frequencyof q depends upon all the digitsthat precedeit, but when all possible combinationsof these digits are takenintoaccountFq approachesequalityforall thedigits

0, 1, 2, *

9, or

Fq

0.1.

(4)

As a resultofthisapproachto uniformityin the qthplace the distributionof digitsin all places in an extensivetabula- tion of multi-digitnumberswill be also nearlyuniform.

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

556

FREQUENCY

Digit

FRANK BENFORD

TABLE III

OF DIGITS

IN FIRST

FirstPlace

AND SECOND

PLACES

Second Place

0.

0.000

0.120

1.

0.301

0.114

2.

0.176

0.108

3.

0.125

0.104

4.

0.097

0.100

5.

0.079

0.097

6.

0.067

0.093

7

0.058

0.090

8

0.051

0.088

9.

0.046

0.085

Reciprocals Some tabulations of engineeringand scientificdata are givenin reciprocalform,such as candles per watt,and watts per candle. If one formof tabulationfollowsa logarithmic distribution,thenthe reciprocaltabulationwill also have the same distribution. A littleconsiderationwill show that this mustfollowfordividingunityby a givenset of numbersby means oflogarithmsleads to identicallogarithmswithmerely a negativesignprefixed.

The Law ofAnomalousNumbers

A studyofthe itemsofTable I showsa distincttendency

forthoseofa randomnatureto agreebetterwiththelogarith- mic law than those of a formalor mathematicalnature.The best agreementwas foundin the arabic numbers(not spelled out) of consecutivefrontpage news items of a newspaper. Dates werebarredas not beingvariable,and the omissionof spelled-outnumbersrestrictedthe counteddigitsto numbers 10 and over. The first342 streetaddressesgivenin the cur- rentAmericanMen ofScience(Item R, Table IV) gave excel- lent agreement,and a completecount (except fordates and page numbers)of an issue of the Readers'Digestwas also in

agreement. On the other hand, the greatest variations from the logarithmicrelationwere foundin the firstdigitsof mathe-

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

THE LAW OF ANOMALOUS NUMBERS

557

maticaltablesfromengineeringhandbooks,and in tabulations ofsuchcloselyknitdata as MolecularWeights,SpecificHeats, Physical Constantsand AtomicWeights.

SUMMATION

OF DIFFERENCES

Nature

TABLE IV

BETWEEN

FREQUENCIES

OBSERVED

AND THEORETICAL

Nature

1 NewspaperItems

D

2.8

11

N

Cost Data, Concrete

12.4

2 PressureLost,AirFlow

F

3.2

12 S

n

n8,n!

13.8

3 H.P. Lost in AirFlow

4 StreetAddresses,A.M.S.

5 Am.League,1936

G

R

P

4.8

5.4

6.6

13 L

DesignData Generators 16.6

16.6

15 I DrainageRate ofRivers 21.6

14 B

Population,U. S. A.

6 Black BodyRadiation 7.2

Q

16 K

n-1,-Vfn .*-

22.8

7 0

X-RayVoltage

7.4

17 H

MolecularWgts.

SpecificHeats

23.2

8 M

Readers'Digest

8.4

18 E

24.2

9 A

AreaRivers

9.8

19 C

PhysicalConstants

34.9

10 T

Death Rates

11.2 20

J

AtomicWeights

35.4

These factslead to theconclusionthatthelogarithmiclaw appliesparticularlyto thoseoutlawnumbersthat are without known relationshiprather than to those that individually followan orderlycourse;and thereforethelogarithmicrelation is essentiallya Law ofAnomalousNumbers.

PART II: GEOMETRIC BASIS OF THE LAW

The data so farconsideredhave been composedentirelyof used numbers;that is, numbersas theyare used in everyday

affairs. There must be some underlyingcauses that distort whatwe call the "natural" numbersystemintoa logarithmic distribution,and perhapswe can best get at these causes by firstexaminingbrieflythe frequencyof the natural numbers

themselveswhen arranged in the

infinitearithmeticseries

1, 2, 3,

n, wheren is as largeas any numberencountered

in use. Let us assume that each individualnumberin the natural numbersystemup to n is used exactlyas oftenas everyother individual number. Starting with 1, and counting up to

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

558

FRANK BENFORD

10,000,forexample,1 would have been used 1,112 times,or 11.12 per centofall uses. If the countis extendedto 19,999 thereare 9,999 l's added, and firstl's occurin 55.55 per cent of the 19,999 numbers. When number 20,000 is reached there is a temporarystopping of the addition of firstl's and 90,000 of the otherdigitsare added to the seriesbefore

0.30

FRfEQfJENVCY

P/,Sr

Or

PLACC

0 OBSERVED

/

2

3

4

a

5

6

D/G/r3

7

8

3

FIG. 1. Comparisonofobservedand computedfrequenciesformulti-digit numbers.

l's are again broughtintotheseries,at100,000. At thispoint the percentageof l's is again reducedto 11.112 per cent as illustratedin curveA of Fig. 2. This curveis Fn and log n plotted to a semi-logarithmicscale. If the equations forA are writtenforthethreediscontinuousbut connectedsections 10,000-20,000,20,000-99,999 and 99,999-100,000 the area underthe curvewillbe veryclosely0.30103,wherethe entire area ofthe frameof coordinateshas an area 1. But an inte- grationby the methodsofthe calculusis merelya quick way ofaddingup an infinitenumberofequallyspaced ordinatesto the curveand fromthisadditionfindingthe averageheightof

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

THE LAW OF ANOMALOUS NUMBERS

559

the ordinatesand hencethe area underthe curve. But ifwe are satisfiedwitha resultsomewhatshortofthe perfectionof the integralcalculus we may take a finitenumberof equally spaced ordinatesand by plain arithmeticcome to practically the same answer. By definitioneach point of A represents

0.4

0O3

0

LINEAR

FREqU/NC/ES

FROM /Q000TO /0,000

1

2

3

4 5

67

89

ArFOR/

8FOR

9

/0

NA7VRAA NUMBER

FIG. 2. Linearfrequenciesof the naturalnumbersystembetween10,000and

100,000.

the frequencyoffirstl's from1 up to that point,and an inte- gration (by calculus or arithmetic)under curve A gives the averagefrequencyoffirstl's up to 100,000. The finitenumber correspondingto equally spaced ordinatesnow representsa geometricseriesof numbersfrom10,000to 100,000,and it is substantiallythisseriesofnumbers,in thisand otherordersof

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

560

FRANK BENFORD

the naturalnumberscale that lead to the,numericalfrequen- cies alreadypresented. Curve B of Fig. 2 is for9 as a firstdigit. The frequency of9's decreasesin thenumberrangefrom10,000to 89,999and thenincreasesas 9's are added from90,000to 99,999,and an integrationundercurveB leads to a good numericalapproxi- mationto thelogarithmicintervallog 10 - log 9, as called for by the previousstatisticalstudy.

Geometricand Logarithmic Series The close relationshipofa geometricseriesand a logarith-

micseriesis easilyseenand hardlyneedsformaldemonstration. The uniformlyspaced ordinatesof Fig. 2 forma geometric

series of numbersforthese numbershave

betweenadjacentterms,and thisconstantfactoris determined in size by the constantlogarithmicincrement.

a constantfactor

Semi-LogCurves

A geometricseriesofnumbersplottedto a semi-logarithmic scale gives a straightline. In the originaltabulationof ob- served numbersthe line of data marked "R" is designated simplyas "street addresses." These are the streetaddresses ofthefirst342 peoplementionedin thecurrentAmericanMen ofScience. The randomnessofsuch a listis hardlyto be dis- puted, and it should thereforebe usefulforillustrativepur- poses. In Fig. 3 these addressesare firstindicatedby the height ofthe linesat the base ofthe diagram. The heightof a line,

measured on

the scale at the left,indicates the

numberof

addressesat, or near, that streetnumber. Thus

therewere

fiveaddressesat No. 29 on variousstreets. In orderto make the trendclearer,the heightsoftheselinesweresummed,be- ginningat theleftand proceedingacrossto theright. It was foundthat fourstraightlines could be drawn among these summationpoints with fairfidelityof trend,and these four lines representfour geometricseries, each with a different factorbetweenterms. Each line will give the observedfre-

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

- ~~D/UTRl8ur/ON AND 3UMMA7YON

/0

/0

t_Q

I

WIll

OF

F/Rsr 34 2

5IMER/CAN MEN OF SCIENCE"

/9-34

-

-

-

-

-

 

~

~

~

~

~

~

_08_C

FIG. 3. Distributionand summationoffirst342streetaddresses,American

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

562

FRANK BENFORD

quencyoverthe numericalrangeit covers,and hencesatisfies the logarithmicrelationship.

The Natural Numbersand Nature'sNumbers In natural events and in events of which man considers himselfan originatorthereare plentyofexamplesofgeometric

orlogarithmicprogressions. We areso accustomedto labeling things1, 2, 3, 4, *** and thensayingtheyare in naturalorder

that the idea of 1,

mentis not easily accepted. Yet it is in this lattermanner

that a surprisinglylargenumberofphenomenaoccur,and the evidenceforthisis available to everyone. First,let us considerthe physiologicaland psychological reactionto externalstimuli.

The growthofthe sensationofbrightnesswithincreasing illumination is a logarithmicfunction,as illustrated by

Fechner's Law.

whiletherodsoftheretinaare aloneresponsive,and a straight line on semi-logarithmicpaper (the stimulus being on the logarithmicscale) can representthe intensity-brightnessfunc-

tionin thisregion. When the cones come into action there is a sharpchangein the rate of growth,and anotherstraight line representsour workingrangeofvision. Whenover-exci-

tationand fatigueset in,a thirdlineis needed; and thusthree

geometricseries could be

illuminationand the sensation of brightness. If the litera- ture containedsufficientnumericalreferences,the brightness functionshouldgive an extremelyclose approximationof the logarithmiclaw of distribution. The sense of loudnessfollowsthe same rules,as does the sense of weight;and perhapsthe same laws operateto make thesenseofelapsedtimeseemso differentat ages tenand fifty. Our musicscales are irregulargeometricseriesthat repeat rigidlyeveryoctave. In the fieldofmedicine,the responseofthe body to medi- cine or radiationis oftenlogarithmic,as are the killingcurves

2, 4, 8, *.

beinga morenaturalarrange-

The growthof sensation is slow at first

used to state the relationbetween

undertoxinsand radiation.

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

THE LAW OF ANOMALOUS NUAIBERS

563

In the mechanicalarts, wherestandard sizes have arisen fromyears of practicalexperience,the finalresultsare often geometricseries,as witnessour standardsof wire diameters and drillsizes, and the issued listsof "preferrednumbers." The astronomerlistsstarson a geometricbrightnessscale that multipliesby 100 everyfivesteps and the illuminating engineeradoptsthesametypeofseriesin choosingthewattage ofincandescentlamps. In the field of experimentalatomic physics,where the resultsrepresentwhat occurs among groups of the building units of nature,and wherethe unit itselfis knownonly by mass action,thetestdata are statisticalaverages. The action of a single atom or electronis a random and unpredictable event; and a statistical average of a group of such events would show a statisticalrelationshipto the resultsand laws herepresented. That this is so is evidencedby the frequent use made ofsemi-logpaper in plottingthe test data, and the test points often fall on one or more straightlines. The analogyis complete,and one is temptedto thinkthatthe 1, 2,

3, *

base e ofthe naturallogarithms,Naturecounts

scale is not the natural scale; but that, invokingthe

e0, ex,

e2x

e3x

and builds and functionsaccordingly.

PART III.

DIGITAL

ORDERS OF NUMBERS

The natural numbersystem is an array of numbersin simplearithmeticseries,but on top of this we have imposed an idea taken froma geometricseries. Numbers composed of many digitsare ordinarilyseparated into groupsof three digitsby interposingcommas,and herewe unknowinglygive evidenceofthe use ofthesenumberson a geometricscale. For convenienceofdescriptionthenaturalnumbers1 to 10 are called thefirstdigitalordernumbers,thosefrom10 to 100 the seconddigitalorder,etc. It willbe notedthat 10 is both the last numberofthe firstorderand the firstnumberofthe second order,and when an integrationis carriedout, as will

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

564

FRANK BENFORD

be done later,10 appears as both an upperand a lowerlimit, and it is thusused in thiscase as a boundaryline ratherthan

a unit zone in the natural numbersystem. In Fig. 4 the curves show the frequencywith whichthe naturalnumbersoccurin the Natural NumberSystem,begin- ningat theleftedge,where1 is theonlynumber,itsfrequency is 1; that is, until a second numberis added 1 is the entire numbersystem. When2 is reachedthefrequencyis 0.50 for1

/INEAIR FREQUEAW/E5

OF

rHE

NATURAL

mum8ERS

/ro /,0O

am

r

~~~~~JECOND

/Gr17L

V0

ORDER-O_

7-I/RD

feL

X

A

I 1

DwX w

I

r 4

DTrAL/

OeRDCR

0

41

11;1

rr/2

;SS

44

42C~~~~~~~XM1.2%-Vtl=SW

FIG. 4. Linear frequencies of the natural

numbersin the firstthree orders.

andO.S0for2. AtS,forexample,thefrequencyforeachofthe firstS digitsis 0.20,andtheequal divisioncontinuesuntil9 is

reached. At 10, the digit

frequencyof 0.20 against0.10 foreach of the othereight digitsthathaveappearedbutonce. It willbe observedthatthecurverisingfrom9 onthescale ofabscissoeis foronlythedigit1, whilethecurvecontinuing

downwardfrom9 is forthedigits2 to 9 inclusive. At 19the

frequencycurvefor2

1 has'appearedtwiceand has a

risestojointhecurvefor1 at 29 and 1

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

THE LAW OF ANOMALOUS NUMBERS

565

and 2 have a commoncurve until 99 is reached and a third first1 is about to be added to theseries. At any ordinatethe curves thereforetell the frequencyof the total numberof naturalnumbersup to that point.

I,

9

III

/0

j

FIG. 5. Continuous and discontinuousfunctionsin the neighborhoodof the digit 9.

The curvesare drawnas ifwe weredealingwithcontinuous functionsin place of a discontinuousnumbersystem. The justificationforusinga continuousform-is that the thingswe use the numbersystemto representare nearlyalways per- fectlycontinuousfunctions,and the number,say 9, givento anyphenomenonwillbe used in somedegreeforall theinfinite

sizes of phenomenabetween8 selvesto singledigitnumbers.

and 10 when we confineour-

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

566

FRANK BENFORD

An enlargedsketchof the linearfrequencycurves at the junction ofthefirstand secondordersis givenin Fig. 5. The

lines h-b and b-j are the computedratios of 1 in this region,

whilethe lines8-b forthe ratioof9 beginsat

size 8 is passed thereis a possibilityofour usinga 9, whilefor

8, foras soon as

rFIW /EUNCY OF S/NGLE / T09

D/G/I3

+

rHEORE7/CAL

 

O

OgSERvVED

FREQUENCY

Or FOOTNOFfS

/N

/o BOOKS

EACH

H/AVING Ar L.EAsr

ONE PAGE WIrH TEN

FoorWores

(2,968

O&RV&ED)

0.50

l

0.40

 

_

0. /O

-

 

/

2

3

4

5

6

7

8a9

FIG. 6. Theoreticaland observedfrequenciesofsingledigits.

size 812 the chancesare about equal forcallingit either8 or 9. The summationof area underthe curve 8-b-c is taken as the probabilityofusinga 9 forphenomenain thisregion. This is about equivalentto knowingaccuratelythesize ofall phenom- ena in thisregionand decidingto call everythingbetween8.5

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

THE LAW OF ANOMALOUS NUMBERS

567

and 9.5 by the number9.

b-j, beginsto risein anticipationofthe phenomenabetween9 and 10 that willbe called 10. It has beennotedthatforhighordersofnumberstheareas underthe curvesofFig. 2 are proportionalto thefrequencyof use of the firstdigit. The same demonstrationwill now be made withthe aid ofthe calculusin regionsthat are markedly discontinuous. Selectingthe thirddigitalorder,Fig. 4, the area underthe 1-curvecan be written

Once 9 is passed the curvefor1,

A1"' -

*199

yddx +j
00

999

lOQO

19

Y2d

+

99

3 dx,

(5)

wherethe ordinatesofthe firstrisingsectionofthe curveare

 

a -

88

Yi

a

(6)

The descendingsectionofthe curvehas ordinates

Y2 =

111

(7)

and thelast risingsectionbetween999 and 1,000has ordinates

a -

a-

888

(8)

The curvesare plottedto semi-logarithmiccoordinatesaDd

x =

log a,

(9)

dx =

dala.

(10)

The integralsaftermakingthese substitutionsgive the value

A1"'= loe

1990

99

8

1000'

A similar operation yields for the 1-curve in the second

digitalorder

A1" A1 =

g190 +

og.

9

+ 100

8

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

568

FRANK BENFORD

and in the firstorder

10

8

Al' = loge-jj+ -?jj-

From the symmetryrunningthroughthese solutionsand fromthe solutionsfortheeightotherfirstdigits,we can write the generalequation forthe Law of AnomalousNumbers

where

F r =

=[log 10 (2.10r1 r

loge

_1

-

oge [lge

Fa

~~a+

(a

1)

10-

)10r -1

1)

1

8

1

N

+or

loritN11

-

_

whereN = log, 10is thefactorto converttheexpressionsfrom thenaturallogarithmsystem,base e,to thecommonlogarithm system,base 10. If highordersofr are considered,as was unwittinglydone in the originalstatisticalwork,these expressionssimplifyby droppingthe terms - 1 in both numeratorand denominator, and the numericaltermshaving lor in the denominatorbe- come negligible. Hence the generalequations become

Fr =

2

log0lo1,

=

Far

a$l

= log

a

 

(12)

,

(13)

but thesetwo expressionsno longerhave a differencein form, and-theymay be mergedinto

+

Far = log1o a

(14)

whichwas the relationshiporiginallyobservedformulti-digit numbers. In Table V numericalvalues are givenforthe theoretical frequenciesof used numbersforthe first,second, thirdand limitingdigitalorders.

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

THE LAW OF ANOMALOUS NUMBERS

THEORETICAL

TABLE V

FREQUENCIES

IN VARIOUS

DIGITAL

ORDERS

569

First Digit

First Order

1 tO 10

Second Order

10 tO 100

Third Order

100 tO 1000

Limiting Order

1 0.39319

0.31786

0.30276

0.30103

2 0.25760

0.17930

0.17638

0.17609

3 0.13266

0.12432

0.12487

0.12494

4 0.08152

0.09479

0.09669

0.09691

5 0.05348

0.07631

0.07889

0.07918

6 0.03575

0.06366

0.06662

0.06695

7 0.02352

0.05444

0.05764

0.05799

8 0.01456

0.04742

0.05078

0.05115

9 0.00772

0.04190

0.04537

0.04576

The frequenciesofthesingledigits1 to 9 varyenoughfrom the frequenciesofthe limitingorderto allow a statisticaltest ifa sourceof digitsused singlycan be found. The footnotes so commonlyused in technical literatureare an excellent source, consistingof units that are indicated by numbers, lettersor symbols. The procedureofcollectingdata forthefirst-ordernumbers was to make a cursoryexaminationof a volume to see if it containedas many as 10 footnotesto a page, forobviously no test of the range1 to 9 could be made if the maximum numberfell short of the full range. The numbershere re- cordedin Table VI are the numberof footnotesobservedon consecutivepages, beginningon page 1 and continuingto the end of the book, or untilit seemed that a fairsample of the book had been obtained. The books used werethe Standard Handbook for Electrical Engineers, Smithsonian Physical Tables,HandbuchderPhysik and Glazebrook's Dictionaryof Applied Physics. In Table VI the observedpercentagesofsingledigits1 to 9 aregivenalongwiththenumberofpagesused ineach volume and the numberof footnotesobserved. The frequencyfor1 is seen to be 43.2 per centas againstthetheoreticalfrequency of39.3 percent,and forthedigit9 theobservationsagreewith theorywithFg' = 0.8 per cent.

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

570

FRANK BENFORD

In generalthe agreementwith theoryis as good as the computedprobableerrorsofthe observation.

Volume

Volume

1. S. H. E. E.

2.

3.

4.

5.

6.

7.

8.

9.

Sm. Phy.

Ta

II. derPhy

H. der Phy

der Phy

H.

H.

der Phy

H. derPhy

H. derPhy

GlazebrookI

.

10. GlazebrookV

ObservedAve.43.2

PredictedAve.39.3

Difference.

ProbableError.

~~Pages

Used

All

All

TABLE VI

COUNT OF FOOTNOTES

1

2

3

4

5

6

7

8

9

 

Total

 

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

Count

 

Frequencies, in Per Cent

 

55.1

22.7

12.3

5.0

2.4

1.7

0.3

0.3

0.3

586

56.3

22.1

6.6

6.1

5.0

2.2

1.1

0.6

0.0

181

360

52.8 23.6

8.5

5.5

4.0

3.2

0.8

0.0

1.6

127

360

37.2

25.7

12.1

9.5

4.8

5.2

2.6

2.2

0.9

230

365

29.7

26.6

14.6

11.0

8.0

5.9

1.8

1.0

1.4

287

361

19.5

17.4

17.7

11.9

11.3

9.2

6.1

5.8

1.1

293

360

33.0 27.5

11.8

10.7

4.3

5.9

2.8

2.4

1.6

254

360

56.8 23.2

6.7

7.6

2.4

1.4-

0.5

1.4

0.0

211

All

49.5 22.3

13.7

6.9

2.3

1.5

1.5

1.5

0.8

394

All

41.7 25.2

13.4

9.1

4.7

3.2

1.7

0.5

0.5

405

 

23.6

11.8

8.3

4.9

3.9

1.9

1.6

0.8

2968

25.7

13.3

8.1

5.3

3.6

2.4

1.5

0.8

 

+3.9

-2.1

-1.5

+0.2

-0.4 +0.3

-0.5

+0.1

0.0

3.0 40.6

40.7 ?0.5

?0.6

?0.5

?0.4

?0.4 ?0.4

 

SummationofFrequencies

One oftheconditionsthatmustbe metbytheseexpressions forthe frequenciesof the integersis that, in any one order, the sum ofthe frequenciesmustequal unity;that is, the sum oftheirprobabilitiesmustequal certainty. Selectingthe first-orderdigits,Eq. 11, and remembering thelogarithmicrulethat the sumofthe logarithmsofa group ofnumbersis equal to the logarithmof theircombinedprod- ucts,we have the probabilityP'

P

1023456789

= logio 9102-345678

rs

1

1-010

whichreducesto

1

10

1

1010

1

1

10

Pt =

log1010 + 0

=1.

1

10

1

10

1

10

1

N'

In a similarmannerfromthe completeset of equations

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

THE LAW OF ANOMALOUS NUMBERS

indicatedby Eq. 11 we have

571

P=

1lo

190 29-3949-59-6979 89-99

g10 99 19 29 39-4959-69-79-89

+I

1

1

10

10 10100

1

-

1

1

-

1

 

100

100

100

100

 

1

1

100

N

= log1010 + 0

=1

and similarproofcan be workedout forthe otherorders.

SummaryofPart III

Single digits,regardlessof their relationto the decimal pointand also regardlessofprecedingor followingzeros,have

a specificnatural frequencythat varies sharply fromthe logarithmicratios. The second digital order,whichis com- posed of two adjacent significantdigits,has a specificfre- quency approximatingthe logarithmicfrequency;and for threeor moreassociated digitsthe variationfromthe latter frequencywould be extremelydifficultto findstatistically. The basic operation

F=fF fda

or

a

F_

_a a

in convertingfromthelinearfrequencyofthenaturalnumbers

to thelogarithmicfrequencyofnaturalphenomenaand human

events can be interpretedas meaningthat, on the average, these things proceed on a logarithmicor geometricscale. Anotherway ofinterpretingthisrelationis to say that small thingsare more numerousthan large things,and thereis a tendencyfor the step between sizes to be equal to a fixed fractionofthe last precedingphenomenonorevent. Thereis

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions

572

FRANK BENFORD

no necessityor implicationof limitsat eitherthe upper or the lowerregionsofthe series. If the viewis acceptedthatphenomenafallintogeometric series,thenit followsthat the observedlogarithmicrelation- ship is not a resultof the particularnumericalsystem,with its base, 10, that we have elected to use. Any otherbase, suchas 8, or 12,or20, to selectsomeofthenumbersthathave been suggestedat various times,would lead to similarrela- tionships; for the logarithmicscales of the new numerical systemwouldbe coveredbyequallyspaced stepsbythemarch ofnaturalevents. As has beenpointedout before,thetheory of anomalousnumbersis reallythe theoryofphenomenaand events, and the numbersbut play the poor part of lifeless symbolsforlivingthings.

This content downloaded from 184.174.224.243 on Sat, 8 Feb 2014 11:22:35 AM All use subject to JSTOR Terms and Conditions