i. SEATZSECAL ANALYSIS FOR
ENDEN’HFECATIO’N 8F KOREAN HANDWRITING

104
667

 

THS

”Eben: .‘ECX' if“ aezree M: M. S.
ZEECHEGM V“ ’ENE‘IERSETY

4\ . a"

3.. .. f‘lt
. UNG: 8; VE‘TO
‘3”‘15
2&04

 

LIBRARY

Michigan State
University

 

 

 

 

A STATISTICAL ANALYSIS
FOR

IDENTIFICATION OF KOREAN HANDWRITING

BY

Sung Tai Cho

AN ABSTRACT OF A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF SCIENCE
School of Police Administration and Public Safety

1964

APPROVED R l ’ w \ {Mm

Chairman '

(/‘7 0“}: A
. 7 . ‘ '(‘_ I / —-‘-‘,. . l_/'
ilbuugha4S\xt ] ’ -~ "*

(fa
,I—l
Memb%§, /7

zxéftbm

Member ‘

cor1
muc}
samp
stat
prob
and r

be 1e

study

The m9
restrj‘
mEHSUri
thESe 1

cation j

Sung Tai Cho

The research reported in this paper was an exploratory study
of the possibility of using inferential statistical methods in hand-
writing analysis. Two inherent characteristics of handwriting--
internal variation of a single individual and coexistence of dis-
similarities with similarities in writings made by more than two
different individuals--make handwriting identification problematic.

Current methods of identification do not completely solve
the problems presented by these factors, because, while they do con-
sider similarity and difference, they do not provide objective crite-
ria for deciding how much similarity there must be before it can be
concluded that samples of writing were made by the same person or how
much dissimilarity there must be before it can be concluded that
samples were made by different persons. Technique of inferential
statistics was developed in order to deal with the same type of
problem in other areas of inquiry and an examination of the conceptual
and mathematical structure of this technique suggests that it can
be legitimately used in the area of handwriting analysis.

The present study was limited in scope in several ways. The
study was based on the measurement of elements of a single character.
The measuring instrument developed by the writer for the study was
restricted to line and angular measurement and was not capable of
measuring stroke curvature. Small samples were used. In spite of
these limitations the statistical tests used led to correct identifi-

cation in 69 percent of the cases for the least discriminating

Sung Tai Cho

element and in 86 percent of the cases for the most discriminating
element. Furthermore, it has been shown that the accuracy of
identification based on a single element can be improved by in-
creases in sample size and by changing the region of rejection of

the null hypothesis. This, to emphasize, is possible. The findings,
then seem to demonstrate that the techniques of statistical inference

hold great promise for improvements in handwriting identification.

A STATISTICAL ANALYSIS

FOR

IDENTIFICATION OF KOREAN HANDWRITING

 

A Thesis
Presented to
the Faculty of the
School of Police Administration and Public Safety

Michigan State University

In Partial Fulfillment
of the Requirements for the Degree

Master of Science

by
Sung Tai Cho

1964

ACKNOWLEDGEMENT

I wish to thank Professor Joseph P. Nicol for help in
review of the literature in handwriting, pretesting, and photo-
graphic works at the initial phase of the research; Professor
Ralph F. Turner, my adviser, for his constant enthusiasm and en-
couragement throughout the research; Mr. P. Rajeswaran for his
critical reading of various chapters; and Professor Patrick T. Cleaver
of The Ohio State University for statistical advice and extremely

valuable suggestions in revision of the thesis.

ii

TABLE OF CONTENTS
Page
ACKNOWLEDGEMENT . . . . . . . . . . . . . . . . . . . . . . ii

LISTOFTABLES.......................iv

CHAPTER
I INTRODUCTION . . . . . . . . . . . . . . . . . 1
The Problem. . . . . . . . . . . . . . . . 1
The Nature of Handwriting. . . . . . . . . 2
Current Identification Methods . . . . . . 5
The Study. . . . . . . . . . . . . . . . . 8
II THE KOREAN LANGUAGE AND WRITING SYSTEM . . . . 11
' The History and Classification of the
Korean Language. . . . . . . . . . . . . 11
Characteristics of the Language. . . . . . 12
Korean Writing System: Hangul . . . . . . 13
Word Formation in Hangul . . . . . . . . . 15
Korean Handwriting . . . . . . . . . . . . 16
III METHODOLOGY. . . . . . . . . . . . . . . . . 20
Inferential Statistics and Handwriting
Identification . . . . , . , , . . . . . 21
The Preliminary Investigation . . . . . . 23
Selection of Characters for Study. . . ... 24
Types of Measurement . . . . . . . . . , . 26
The Measuring Instrument . . . . . . . . . 27
IV ANALYSIS AND FINDINGS. . . . . . . . . . . . . 30
The Major Analysis . . . . . . . . . . . . 30
Tests Using Small Samples . . . . . . . . 33
Discussion of Findings . . . . . . . . . . 37
v CONCLUSIONS..................43

BIBLIOGRAPHY. . . . . . . . . . . . . . . .

APEND ICES O O O O O O O O O O O O O O O O O O O O O O O O 5 1

iii

TABLE
2.1
2.2
2.3

2.4

3.1

4.1

4.2

4.3
4.4
4.5
4.6

4.7

LIST OF TABLES

Hangul Vowels . . . . . . . . . . . . . . . . . .
Hangul Consonants . . . . . . . . . . . . . . . . .
LP and Distribution of Ratio . . . . . . . . . . .

Syllable and Word Formation . . . . . . . . . . . .

Assignment of Numbers to Letter Positions . . . . .

T-Tests for Pairs of Samples for Different Writers.

Means and Standard Deviations for Pairs of Samples
for the Same Individual . . . . . . . . . . . .

T-Tests for Pairs of Samples for the Same Writers .
Correct Identification . . . . . . . . . . . . . .

Means and Standard Deviations for Small Samples . .

Means and Standard Deviations for Individual Samples.

T-Tests for Pairs of Small Samples for Different
Writers . . . . . . . . . . . . . . . . . . . . .

iv

Page

14
15
16

18

25

34

35
36
36

37

40

CHAPTER I

INTRODUCTION

1. The Problem

 

At present handwriting identification, unlike fingerprint
identification, has not yet reached a level of validity and reliabi-
lity which will permit its full acceptance either by criminological
science or by the courts of law. The difficulty of developing a
scientifically valid and legally acceptable handwriting identifica-
tion lies in the nature of handwriting itself. In terms of con-
sistency and individual uniqueness handwriting occupies the opposite
end of a continuum from fingerprints. Repeated prints made by a
given finger of a particular individual are both consistent, in that
they exhibit no important variation, and unique, in that they are
demonstrably different from prints made by any other individual.

Handwriting, on the other hand, is both internally variable
and not completely unique. Within the writings of a given person there
will be variation in the way that the same letter is formed, while the
writings of two individuals will generally show some dissimilarities

as well as similarities. Present methods of handwriting identifica-

tion are not entirely satisfactory because they do not solve the
problems posed by internal variation and by the coexistence of
similarity and dissimilarity in questioned and standard writings.

The present study was an attempt to develop a method of
handwriting identification based on statistical analysis which it
is hoped will constitute an advance in the precision of present
identification methods. It is also hoped that this technique will
meet the scientific criteria of objectivity, reliability and validity
since any criminal identification method must meet these criteria
before it will be accpeted by courts of law.

The remaining part of the present chapter will be concerned
with a review of present handwriting methods and with an outline of

the method developed in the present study.

2. The Nature of Handwriting

 

As the preceeding discussion pointed out, handwriting has,
for identification purposes, two salient characteristics: internal

variation and a lack of individual uniqueness which leads to the

 

1"Questioned writing" is a handwriting specimen whose author-
ship is disputed or unknown. Osborn uses the term "questioned writing"
in referring to documents in general while Hilton uses the word both
in this sense and in the sense of disputed writing. "Standard writing"
or "sample writing" refers to a handwriting specimen which may be taken
from documents known to be written by the suspect or which may be re-
quested from the suspect. See Ordway Hilton, Scientific Examination
of Documents, Chicago: Callaghan and Company, 1956, pp. 10-11, 141-142;
Wilson R. Harrison, Suspect Documents: Their Scientific Examination,
New York: Frederick A. Praeger Inc., 1958, pp. 292, 297-307; Albert S.
Osborn and Albert D. Osborn, Qgestioned Document Problems: The Dis-
covery and Proof of the Facts, Sixth Printing, Albany, N.Y.: Boyd
Printing Company, 1947, pp. 14, 22, 205-206, 352.

 

 

 

 

3

overlapping of characteristics in writing made by two individuals.

It is generally accepted that no individual writes in a com-
pletely uniform manner--within a sample of handwriting made by a given
individual there will be variations in the way that a given letter
is made.2 This variation results from variation in writing condi-
tions,3 writing materials and writing instruments and from the fact
that there is a lack of machine-like precision on the part of the
writer.4 This last factor is particularly important since variation
occurs in a given individual's writing even when writing conditions
and materials are held constant.

The second characteristic of handwriting--the lack of complete
uniqueness in a given individual's writing which leads to a coexis-

tence of similarity between writings made by different individuals--

 

2To mention only a few John J. Harris, "How much do people
write alike: a study of signatures," Journal of Criminal Law, Crimi-
nology and Police Science, 48(1), Vol. 6, March-April, 1958, pp. 647-
651; Wilson R. Harrison, Suspect Documents: Their Scientific Examina-
tion, New York: Frederick A. Praeger Inc., 1958; Ordway Hilton,
Scientific Examination of Documents, Chicago: Callaghan and Company,
1956: Idem., "Proper Evaluation of Dissimilarities in Handwriting,"
International Criminal Police Review, 104, January, 1957, pp. 48-51;
Albert S. Osborn and Albert D. Osborn, Questioned Document Problems:
The Discovery and Proof of the Facts, Albany, N.Y.: Boyd Printing
Company, 1944; Idem., The Problem of Proof, Sixth Printing, Albany,
N.Y.: Boyd Printing Company, 1947.

 

 

 

 

3Harrison, Ibid., pp. 3, 297, 331-333, 439; Hilton, Ibid.,
pp. 215-218, 246, 247; Osborn, loc. cit.

4Harrison, Ibid., p. 298; Hilton, Ibid., p. 141; Osborn,
Ibid., p. 205.

4

has been discussed by several writers.5 The coexistence of similarity
may result from the fact that individuals whose writings show simi-
larity have learned a similar "style characteristic."6 In some

cases it may be merely the result of random chance.

These factors have, to the present, put rather severe limi-
tations on handwriting identification as a technique in criminal law.
In any case where specimens of writing are compared there are two
possibilities: either the writings were made by the same individual
or by two different individuals. The factors of variation and co-
existence mean that, regardless of the element of combination of
elements of the writing used in the identification process, there
will usually be differences between the writings made by the same
individual and similarities in writings made by different individuals.
This problem of ambiguity which makes a positive decision about the

authorship of a specimen of writing inherently difficult, has not

 

5Wilson R. Harrison, Suspect Documents: Their Scialtific
Examination, New York: Frederick A. Praeger Inc., 1958; Ordway Hilton,
Scientific Examination of Documents, Chicago: Callaghan and Company,
1956; Albert S. Osborn, and Albert D. Osborn, Questioned Document
Problems, The Discovery and Proof of the Facts, Albany, N.Y.: Boyd
Printing Company, 1944. However, the lack of uniqueness in hand-
writing does not necessarily invalidate individuality of handwriting
of a given individual. See Hilton, Ibid., p. 136, 141; Osborn,
Ibid., p. 231, 270. Nor does internal variation necessarily preclude
identification of individuality. In fact, this is the basis of the
whole identification effort.

 

 

 

 

 

6Harrison distinguished "style" and "personal" characteristics,
and the first step in handwriting identification ought to be the dis-
tinction between them. See Harrison, Ibid., pp. 288-289. To borrow
Harrison's distinction, identification is pursuit of "master pattern,"
which may be defined as "personal" characteristics.

been entirely solved by existing identification methods which depend
largely on finding similarities or dissimilarities in questioned and
standard writings. Osborn, who was aware of this fact, warned against
identification based solely on either similarity or dissimilarity:

(By this same method) of ignoring differences (dissimila-
rities) and looking only for similarities almost any two
things not altogether unlike, can be proved to be the same.
This is the basis of the common error of the incompetent
witness in identifying the writing in anonymous letters.
Similarities can always be found in two writings in the same
language or in two writings not utterly unlike. Mere simi-
larities do not necessarily prove genuiness any more than
mere superficial differences necessarily prove lack of ge-
nuiness. The incompetent witness, notwithstanding this fact,
by dependence upon similarities alone reaches the conclusion
of genuiness, or by dependence upon differences alone reaches
the conclusion of lack of genuiness. . . .

It seems evident, then, that a legally and scientifically
acceptable handwriting identification system must be based on a tech-
nique which can control for variation and coexistence of similarity.
A review of the literature on handwriting identification, presented
in the following section, indicates that current methods have not

yet completely developed such a technique.

3. Current Identification Methods

 

Handwriting identification methods can be classified as micro-

examination or the examination of writing elements--stroke lengths,

 

7Albert S. Osborn and Albert D. Osborn, Questioned Document

Problems: The Discovery and Proof of the Facts, Albany, N.Y.: Boyd
Printing Company, 1944, pp. 240-241. See also Idem., pp. 237, 244.

 

6

angles between strokes, etc., and macro-examination which is a method
which classifies handwriting into styles on the basis of letter design.

Current handwriting identification analysis usually combines
macro-examination with a form of micro-examination called the "com-
parative method."8 Better terms for "comparative method" would be
the "one-to-one" method or the "juxtaposition" method since, in the
final analysis, all handwriting examination is comparative.9

In the one-to-one method an element of the questioned writing
is juxtaposed with a similar element in the suspected writing and the
two elements are compared for similarity or dissimilarity. In this
type of comparison the average or modal pattern of the two writings
is not considered. Such a method is inadequate since it does not take
into consideration the internal variation of the two writings and the
likelihood of similarity between at least some elements of the ques-
tioned and suspected writing. Random selection of elements to be
compared may lead to similar elements being selected from dissimilar
writings or dissimilar elements being selected from Similarlwritings.

In recent studies, Sjoegren and Smith have attempted to over-
come the problems caused by variation and coexistence of similarity

by the use of a system of weighting.10 In this system of evaluation

 

8For instance, Tore Sjoegren combined measured character-
istics with general features of handwriting, such as arrangement,
spacing, connections, etc. Tore Sjoegren, "Handwriting Comparison
and Probability," International Criminal Police Review, Vol. 92,
Nov. 1955, pp. 274-283. Stanley Smith suggested use of the latter
group characteristics in the "Secondary Examination." Stanley S.
Smith, "A Method of Comparing Written Documents," ibid., Aug.-Sept.
1954, pp. 205-215. Others such as Harrison, Hilton and Osborn are
of the same opinion.

 

9Harrison, Hilton, Osborn used juxtaposition method for
illustrations in their texts.

lolbid.

7
each similarity in a particular element is given a plus rating and
each dissimilarity is given a minus rating. These weights are summed
algebraically in order to determine the overall tendency to similarity

or dissimilarity.
Hilton has argued that this method is not entirely adequate

and has suggested that the major emphasis in evaluation be placed on

the factor of dissimilarity:11

Rather it is an analysis of the true meaning of the dissimi-
larities and if they are found to be basic and without
logical explanation, the realization that these differences
are the controlling factors which establish that the known
(standard) and unknown (questioned) writings are by two dis-
tinct persons.

Harrison is in general agreement with Hilton on the importance
of dissimilarity.12 Osborn, although agreeing at some points with
Hilton, suggests that similarity should be given equal weight:13

The process of comparison in any field is reasoning regarding
similarities and differences, and necessarily the subject

has an important place in all kinds of investigations. Errors
in identification problems are due not only to the failure

to see the outside things but to the failure to recognize
their real differences and their fundamental similarities

and to understand them and interpret them when they are

seen. Much of what is called science is merely accurate
classification resulting from intelligent observation and

reasoning leading to a correct recognition of similarities
and differences.

 

 

11Ordway Hilton, "Proper Evaluation of Dissimilarities in
Handwriting," International Criminal Police Review, No. 104, January
1957, p. 49. Hilton has kept this view consistently in other places:
Ordway Hilton, Scientific Examination of Documents, Chicago: Callaghan
and Company, 1956, pp. 51, 136-137, 144.

12Harrison, ibid., pp. 343-345.
13Albert S. Osborn and Albert D. Osborn, Questioned Document
Problems: The Discovery and Proof of the Facts, Albany, N.Y.: Boyd

Printing Company, 1944, p. 237. See also Hilton, Scientific Examina-
tion of Documents, p. 143.

 

 

8

Three weaknesses seem apparent in current techniques. The
method of juxtaposition is obviously inadequate since it does not
take into consideration the variability factor. Second, the emphasis
placed by some writers on the factor of dissimilarity has no adequate
theoretical ground since it stresses only one aSpect of variability.
Third, the weighting method developed by Sjoegren and Smith, while
avoiding the first two weaknesses, can be criticized on the ground
that it fails to provide a method for determining the degree of
positive or negative weighting necessary to permit acceptable infer-
ence about authorship.

It seems evident that some fresh approach will be necessary
in order to develop a handwriting identification method which can cope

with the problems of variation and coexistence of similarity.

4. The Study

The study reported in this paper was an exploratory attempt
to apply the technique of statistical inference to the problem of
handwriting identification.14 A comparison of the problems confronting
the handwriting identification specialist and the problems typically
encountered by biological or social scientists attempting to make
inferences about samples of variable material show that in many essen-

tial respects they are remarkably similar.

 

14The Theory and application of inferential statistics will

be discussed in more detail in Chapter III. See Helen M. walker and
Joseph Lev, Statistical Inference, New York: Holt, Rinehart and Wins-
ton, 1953; W. Allen Wallis and Harry V. Roberts, Statistics: A New
Approach, Glencoe: The Free Press, 1956.

 

 

9

Many research hypotheses in the biological and social sciences
require that the investigator determine whether two samples, alike
in some respects and different in others, were drawn from the same
universe of measurement. The problem cannot be solved by a simple
examination of the samples since differences may have occurred as a
result of the sampling error inherent when samples are drawn from
a heterogeneous universe of measurement. Inferential statistics allow
the researcher to reject or accept the hypothesis that the samples
were drawn from the same universe, not with absolute certainty, but
with a specified small margin of error.

Handwriting identification can be conceptualized as a problem
in sampling. The examiner has two (or more) samples of writing which
will usually show both similarities and dissimilarities. These may
be drawn from two different universes (i.e., made by different individ-
uals) or they may have been drawn from the same universe (i.e.,
made by the same individual) and show differences because of sampling
error.

In the present study specimens of Korean handwriting collected
by the researcher were treated as statistical samples. Measurements
of certain micro-characteristics were made and these measurements
were subjected to statistical analysis in order to determine whether
inferential statistical methods could differentiate between samples
whose authorship was known a priori.

The present study was limited in three respects. First, only
Korean writing was used in the present study because of the writer's

familiarity with this writing system. Second, only a limited set of

10

measurements were used. No attempt was made to evaluate macro-
characteristics statistically, although macro-characteristics were
used in determining the final sample, and only a small set of micro-
measurements were used. Finally, no attempt was made in the present
study to analyze disguised writing. Although the present study was
limited there are obviously rich potentialities for statistical in-
vestigations of handwriting using other languages and more refined
methods.

In the following chapter there is a discussion of the Korean
language and handwriting system, with particular attention to problems
in identification inherent in this writing system. Chapter III dis-
cusses in detail the statistical methods, samples and measuring tech-

niques used in the study and presents an analysis of findings.

CHAPTER II

THE KOREAN LANGUAGE AND WRITING SYSTEM1

Because the technique of handwriting identification reported
in this paper was based on samples of Korean writing a brief dis-

cussion of the Korean language and writing system is appropriate.

The History and Classification of the Korean Language

Most linguists classify Korean in the Altaic language group
which in turn is considered to be closely related to the Ural group.
If there is such a relationship,then Korean is related to such Euro-
pean languages as Finnish and Magyar. The opinion that the Ural and
Altaic languages are related is based on the fact that both are
agglutinative languages and the belief that the origins of both can be
traced to central Asia.

The Altaic group is divided into three branches: Turkic, Men-
golian and Tungustic. Korean belongs to the Tungustic branch. Tungus-
tic variants are spoken in Siberia, in Manchuria and by some 35 million

Koreans, both in Korea and in Japan. Korea shares similarities in its

 

1This part of the paper is mainly taken from the book: Korea--
Its Land, People, and Culture of Allggges, Seoul, Korea, Hakwon-Sa,
Ltd., Part III, People, Language, Chapter 2, Language, pp. 117-124,
1960. Some modifications and innovations were made, however, to meet
the purpose of this research--especially design.

11

12
agglutinative structure with Japanese, but not with Chinese, al-
though modified Chinese characters are used in Korean writing. The
relationship of Korean and Japanese is not surprising since it is
believed that Japan was settled in part by immigrants from Korea and

from other areas where Tungustic languages were spoken.

Characteristics of the Language

 

Although the systematic study of the Korean language has not
yet been completed certain characteristics have been identified.

1. Korean vowels are divided into three groups and vowel com-
binatiOns tend to be made within these groups and not between them.
(Such vowel harmony is a general characteristic of the Ural-Altaic
languages.) The three groups are:

a. Hard Vowels: ’. (a)J_(o))-’ (ae)J,) (oe)

b. Medium Vowels:" (i)

c. Soft Vowels: ‘1 (6)1- (u).—(E)4] (e)
The vowels of groups a and c tend to combine with others within their
group but to resist combination with the other group. The single vowel
of the second group may combine with vowels from either of the other
groups.

2. Korean words, unlike words in Indo-European languages,
never begin with more than one consonant. Such.English.words as
"strike" or "break", for example, would not occur in Korean. Further,
Korean words do not begin with liguid consonants such as the English
"r" or "1". Finally, Korean lacks the consonants "f" and "v". These

sounds are approximated in Korean by the consonant a (p or b).

13

3. The most distinctive characteristic which separates
Korean and other Ural-Altaic languages from Chinese or from Indo-
European languages is the agglutinative structure of its grammar.
.In Indo-European languages grammar is indicated by modifications in
'words in the forms of tenses, bases andunumbers. In Chinese "full"
words or denotative words are never inflected and grammar is indicated
by word position and by the use of word particles or "empty" words
which help convey meaning. Agglutinative languages such as Korean
fall between these two types of languages. Words are fixed, as in
Chinese, but the "empty" words of particles become "glued" or attached
to the fixed words in somewhat the same way that inflected endings

are attached to word roots in Indo-European languages.

Korean Writing System: Hangul

 

Hangul, the Korean writing system, is phonetic, like English,
rather than ideographic, like Chinese, although the characters or
letters were adopted from Chinese.

Hangul was developed in Korea's "Golden Age" Which was ush-
ered in with the reign of the fourth Yi king, Sejong, in the 15th
century. King Sejong, who believed that the function of written
language was communication with the common people, developed an alpha-
betic language in order to facilitate this communication.

There are 24 letters in the Hangul alphabet--10 vowels and
14 consonants. The total number of letters in the Korean alphabet

has been reduced since the period of King Sejong, with the elimination

of such vowel as o and such consonants as A , é .

l4
Hagul vowels can be formed into dipthongs and can be classi-
fied into two categories:
Simple: ‘- (a). I: (ya). .1 (8). :1 <y8).J_<o),iL(yo>
'T <u).;r(yu)...—-(fi>. I (1)
Compound: y, (ae). )1 <yae>. 4] (e). a] (yet), (oe).

7' (ui),J,', (wa), 7’ (wo),‘w, (wae),fi (we)

If the simple and compound vowels are considered together,

there are 21 vowels. Modern grammarians classify them as simple and

dipthong vowels:

Table 2.1. Hangul Vowels
Simple--. k 4, J“ T r, H A, Jr,

Dipthongs"): :’ JL 7 1:, a], J, /’ )1"
14 w a!

This distribution is based on the following triangular chart

for the simple vowels:

Figure 2.1

I 4T

 

Dipthongs are formed by the following principle of combination:

 

2
MeGune-Reisschuer system is the best known system to
romanize Korean pronounciation. Example is shown in Table 2.2.

15

1+.»—
)+-}]

#:144444-
7r 1+Tsfi=l+fldl
71 T+l win-+1 sJ+=—L+I-
7"='T+" ;—91=-L-Ff);fi)='T+-JI

The consonants of Hangul now in use are either simple or

double:

Table 2.2. Hangul Consonants
Simple--7 (k or g), \_ (n),C (t or d), a (r or 1),
\J (no.6 (p or b).A (s). o (silent or ng).

not: or j>.%, (ch'). >7 (k'),j’_<p'),
90-01), '5. (t)

Double--77(kk) ,EC (tt) ,3& (pp) ,A (ss)% (tch)

Word Formation in Hangul

 

Hangul is an alphabet but its function in word formation is
not the same as the alphabet in English. In English,characters of
the alphabet are used to form separate words while characterS‘ in
Hangul are used to build up separate syllables. Hence Hangul may be
considered a half-way point between an alphabetic and a syllabic sys-
tem of writing.

A Korean word consists of one or more "boxed-in"isyllables. A
If such a system were used in English the words "Lansing" and "Jackson"

would be written as E and g: :0. This "boxed-in" syllable

formation and subsequent word formation causes a special problem in

 

 

 

 

 

 

l6

identification in Korean handwriting because variation or "natural
transformation" is caused by changing the position of a character.
(See Design for Natural Transformations and Letter Position).
Korean words are generally composed of from one to four
syllables, although there are a few words of seven or eight sylla-
bles. Seventy percent of the vocabulary is made up of words of two

and three syllables.

Korean Handwriting

 

When Korean is written by hand the size and shape of individ- ‘
ual characters within a syllable "box" vary according to letter posi-
tion and other combining factors. An examination of the handwriting
samples used in the present study verified the fact that 1etter_posi-
tion influences the shape and size of characters and their elements.

The characters in the Korean alphabet can be classified into

three groups according to the type and number of letter positions:

Table 2.3. LP and Distribution of Ratio

 

Groups Letter Position Letter Position
(consonants) (vowels)

1 initial (7...: ) inner ( 'J ’# )

2 initial (7-? ) outer ( I-I ’: ) 1

2 upper ( 7-?) medial (.leth .—

3 lower ( '7'?) all V0W9-1S 0"”)

 

Note: 1. Vowels of the same group can be combined; vowel and
consonant take opposite position in combination.
2. This rule applies to double consonants, compound con-
sonants and dipthongs.
3. Consonants of the group 3 may be called terminal or
"batchim." This is used in this study on examination
of the vowel ’- (a).

17

In addition to variation introduced as a result of position
in the syllable "box" there is also variation resulting from other
factors. For example, when a syllable is at the end of a sentence
or paragraph, its characters tend to be either increased or reduced
in size.

As indicated above, writing is done in an imaginary "box"
for each syllable; syllables are combined into rectangular words.
Elements of a character are either predominantly horizontal or pre-
dominantly vertical; vertical strokes are made from top to bottom and

horizontal strokes are made from left to right.

Figure 2.2. Illustration for Korean Writing Step

 

N

4--—--b---—

 

v

51

3
'/\‘ «17“
ax.

L______

v

—-———-—(

 

 

 

 

The position of a vowel in a syllable can cause modification in these
basic strokes. For example, the writing steps for a consonant consist
first of a vertical movement and then of a horizontal movement if the
consonant is combined in a syllable with vowels in inner or outer
positions; no horizontal movement is needed if it is combined with
vowels in the medial position. The relationships between consonants

and vowels in syllable and word formation are shown below.

18
Table 2.4. Syllable and Word Formation

 

 

 

 

 

 

Groups Examples Specifications
1 I 1. 3 i- 1=initial consonant
’JZE 2=outer vowel
tit) 3=initia1 consonant
4-outer vowel
1 J.
2 (3 L: 1=initial consonant
2-outer vowel
‘7
3 y ‘3 l=upper consonant
O 7 2=media1 vowel
-—' 3=upper consonant
,1.
1 41 4=medial vowel
5 z l=initial consonant
1 4 ;' 2=outer vowel
:2 3=lower consonant
\p A 4=initia1 consonant
\_3 o 6 5=outer vowel

6=lower consonant

 

The same writing steps are used in the formation of sentences
and paragraphs. Traditional style has a top-to-bottom direction for
vertical movements and a right-to-left direction for horizontal move-
ments; there is seldom any top-to-bottom direction of vertical move-
ment combined with left-to-right horizontal movement. The Western
style is adopted in the copybook style of writing; only 2 out of 78
standard writings used in the present study were written in the
traditional way.

As in Chinese and Japanese, Hangul script is modelled after
the brush-stroke tradition--the initial and terminal edges of each

stroke are usually pointed while the middle part is rounded. Adherence

TAII_

 

19

to this method depends not only on practice but also on the skill of
the writer. (Individuals who are skilled in writing Chinese charac-
ters usually write Korean characters in the traditional style). Hence,
presence or absence of the traditional style has some evidential value.

Connecting strokes are more evident in vertical writing than
in horizontal writing. However, in the standard writing pattern no
connecting strokes are to be made in either case and, if they are
present, they are the result of either speed of writing or individual
habit. (It should be noted that some of the old style brush writing
have connecting strokes as a standard form, particularly in vertical
writing.)

In the present study individual "elements" which include
strokes and angles between strokes, are the basic units of study.
Neither stroke widths or connecting strokes are examined, although

both might be used in further study.

CHAPTER III

METHODOLOGY

As indicated in Chapter I, the present study was designed
to explore the usefulness of inferential statistics in the area of
handwriting identification. More specifically, the object of the
study was to determine whether samples of handwriting made by differ-
ent individuals whose identity was known could be distinguished
quantitatively and objectively through the use of statistical analysis
and, as a corrolary, to determine if two samples of writing known to
be made by the same individual could be shown by similar methods, to
have been made by the same writer. If these similarities and
differences could be demonstrated statistically then the applica-
tion of statistical methods in handwriting analysis could fruitfully
be made the object of further study and refinement.

The present study was patterned after typical research in
the biological and social sciences.1 Random samples of handwriting
were used, objective methods of measurement were developed and
'appropriate statistical tests were applied to the measurements. The
present chapter discussed these aspects of the study while the follow-
ing chapter presents the analysis of the data and discusses the find-

ings of the study.

 

1For a discussion of social science research methodology see
Claire Selltiz et a1, Research Methods in Social Relations, 2nd ed.
New York: Henry Holt, 1959; Pauline Young, Scientific Social Survgys
and Research, Englewood Cliffs, N.J.: Prentice-Hall, 1956.

20

 

 

 

21

Inferential Statistics and Handwriting Identification

The present section presents a more detailed examination of
the relevance of inferential statistics to handwriting identification,
a question which was discussed briefly in the first chapter.

In the biological and social sciences (and in other fields
as well) researchers are often confronted with the necessity, be-
cause of limited time and resources, of conducting research with
samples rather than with universes.2 Such an approach poses problems
in making generalizations about the universe from which the sample
was drawn since samples drawn from heterogeneous universes (i.e.,
universes in which the eleuent being measured exhibits variability)
are unlikely to coincide in any summary measurement, such as the
mean, with the universe. A series of samples drawn from a particular
universe will exhibit "sampling variation" or "sampling error"; a
descriptive statistic, such as the mean, will vary from sample to
sample and the means of most of the samples will not be the same as

the mean of the universe.

 

In statistical usage the term "universe" refers to all individ-
uals of a particular category (or more specifically, to all measure-
ments of a particular class, such as the heights of all residents of
Lansing, Michigan). If all individuals in a universe are measured,
the resulting set of measurements is called a "census." A smaller
set of individuals selected from a universe is called a "sample."
Walker and Lev, _p.‘gl£., p. 1.

3See Walker and Lev, _p. 235., Chapter I; Hubert M. Blalock,
Social Statistics, New York: MeGraw-Hill, 1960, Chapter 8; Celeste
MeCollough and Loche Van Atta, Statistical Concepts, New York: McGraw-
Hill, 1963, Chaps. 11 and 12.

 

22
The factor of sampling variability gives rise to a common problem
in which the investigator has two (or more) samples and wishes to
determine whether they were drawn from the same universe. In all
likelihood the samples will not be the same in any of the descriptive
statistics (e.g., mean, standard deviation or median) which might be
computed for them. Inferential statistics allow the investigator
to determine, within a specified probability of error, whether the
two samples were drawn from different universes, with different
characteristics, or from the same universe. In the latter case
the differences in the samples can be attributed to sampling variabi-
lity. The investigator has a choice of a variety of measures--chi
square, t-test, F-test and others--which are applicable to different
forms of data but which are similar in that they permit inferences
to be made with known probabilities of error.

As indicated previously, the problem of handwriting identifi-
cation closely parallels this type of problem. All of the writings
of a given individual can be conceived of as a universe of measure-
ment while any particular piece of writing can be conceived of as
a sample from this universe. Because of the factor of variation in
writing two samples of writing made by a particular individual will
probably show differences in some measurable characteristic. such
as the mean length of an element of some character. 0n the other
hand writings made by different individuals often show similarities
as well as differences. The problem confronting the examiner is
the question of determining an appropriate weighting of similarities

and differences in order to make a decision about authorship. Such

 

23
a problem can be approached within the logical framework of infer-
ential statistics since the problem is one of determining whether
the writings were drawn from the same universe (that is, made by the
same individual) and are different because of sampling variability
or whether they were drawn from different universes (i.e., made by
different writers). While inferential statistics does not permit
such a decision to be made with certainty it does allow it to be
made with a specified probability of error. If this technique can
be applied to handwriting analysis it should be a useful method in
objectively evaluating similarity and dissimilarity.

The present study was based on the use of two inferential sta-
tistical tests--the F test and the t-test--app1ied to measurements of
several micro-elements of handwritings made by a sample of Korean
students. The present chapter discusses the sample and the techniques
of measurement. The analysis of the measurements is discussed in the

following chapter.

The Preliminary Investigation

 

One branch of handwriting identification involves the classi-
fication of writing types, on the basis of qualitative patterns or
configurations, into broad categories or "class patterns." It can
be seen that the analysis of wrflting within a particular "class
pattern" provides the most rigorous test of the power of inferential
statistics since the differences between writings within the same
class pattern will be smaller than differences between writings in

different classes. Since the present study was intended as a test

24

of the relevance of inferential statistics in handwriting identifi-
cation it was decided to use samples drawn from the same class
pattern for analysis.

In order to identify the class patterns within Korean writing,
a sample text, which was to be copied in order to provide samples
of writing, was developed. (Appendix 1) The text for the letter was
designed to meet two criteria: inclusion of as many characters as
possible and repeated use of high frequency characters. The sample
texts were sent to a sample of 200 Korean college students studying
in the United States. The sample was drawn randomly from two direc-
tories of students supplied by the Consulates General in New York
and San Francisco who have jurisdiction over the Korean student popu-
lation in the United States. The 78 students who returned the
letters included both graduate and undergraduate students and had a
mean age of 25.7.

For the final study writings from the personal correspondence
file of the researcher collected from 1954-1962 were used. The
sample included ten letters, each chosen randomly from a total of

490 letters, for ten individuals.
Selection of Characters for Study

Since it was beyond the scope of the present study to analyze
all of the characters found in the sample writings it was necessary
to select certain characters for analysis. For the present research
a consonant 7‘ (j) and a vowel ’- (a) were chosen for a study of
class patterns and the vowel was selected for measurement and statis-

tical analysis. In choosing the two characters four criteria were used:

’25
high frequency of usage in ordinary language, maximum number of
strokes, common origin and persistence of class patterns. The pre-
test showed that the consonant had 25 class patterns and the vowel
18. (Appendix 2)
As indicated previously a letter in Korean writing will vary
according to its position in a syllable and according to the letters

with which it combines. These variations are shown in Table 3.1.

Table 3.1. Assignment of Numbers to Letter Positions

 

 

 

 

Consonants Vowels
1 Letter Letter Letter
Numbers Positions Numbers Positions Numbers Positions
1 initial 4 inner 7 inner
2 upper 5 outer 8 outer
3 lower 6 medial 9 medial

 

1. These numbers appear in the classification code.

2. Letter Positions, 7, 8, 9 are similar to 4, 5 and 6
respectively with an exception that 4, 5 and 6 have
lower consonants or "Batchim" additionally.

3. The consonants may be combined with any one of the
letter positions specified hereon.

4. For the exemplary study in this paper, Letter Posi-
tion 9 of the vowel along with Class Pattern 1 was
used for measurement.

5. See Korean Writing System for Syllable and Word Forma-
tion.

In order to further limit the variation in the letter studied
analysis was limited to letters occurring within a single position

as well as within a single class pattern. In the present study, then,

thw vowel }-(a) occurring in position 9 of class pattern 1 was measured

and analyzed.

26
Types of Measurement

Since Korean writing is linear, with the exception of a few
circular characters, two types of measurement can be made: measure-
ment of stroke length and measurement of the angle between strokes.
In addition the writing of many individuals show some curvature in
the normally linear strokes. The measurement of this type of cur-
vature was beyond the technical means available to the researcher
so measurement was confined to length, angle and ratio between
lengths.

In the statistical analysis the vowel ‘}-(a) was measured
' and analyzed. This vowel has three elements which can be measured:
the vertical stroke (X1), the horizontal stroke (X3) and the angle

between them (X2). (See Appendix 3)

Figure 3.1

/
X.
\e

 
     

)‘a

    
  

x;
The ratio (X1/X3) can also be analyzed. All of these measures were
used in the analysis.

Although the operation of measuring straight lines and
angles would appear to be simple the factor of linear irregularity
found in handwritten characters requires the adoption and systematic

4
use of measurement rules.

 

4 O O 0
See Appendix 3. Application of the measurement plate shows
how linear irregularity is compensated by consistent use of measure-
ment rules.

27

The Measuring Instrument

A special measuring instrument was developed by the writer
for use in the study. In order to make the instrument useful for
further studies of Korean writing it was decided to design it in
such a way as to make possible the measurement of any of the Korean
characters. An examination of Korean letters indicated that they

can be classified into a few general shapes as indicated in Figure

 

 

3.2.
Figure 3.2. Basic Pattern Chart
1 Basic Basic
Groups Consonants Pattern Vowels Pattern

 

I Kiwi A H) (MAP ----|

H 0,».- @ mom “‘4—
1” 79); \C(-'—' 43.)} [E] _..,| ----5

IV D ,:(H) (I) E

l. Grouping based on common origin of the stroke and
basic design.

2. - - - - denotes movable line, and
stable line.

3. ( ) sub-grouping.

 

 

denotes basic

The basic patterns shown in Figure 3.2, square, triangle, and circle,
were included in the measuring instrument. In order to insure flexi-
bility in measurement these were incorporated in the instrument as
concentric squares, wedges and circles. In order to avoid crowding,
some modification of these patterns was made. Concentric circles were
used as the basic unit of measurement. In the upper half of the plate

concentric squares were modified into parallel lines which corresponded

28

to the units of the concentric circles. Concentric wedges, as such,
were not used but the parallel lines and the lines dissecting the
concentric circles formed wedges which could be used for measurement.
The lower half of the plate was divided into 16 parts, with each
part equaling 11.25 degrees. The entire circle was divided into
eight sections, each of 45 degrees, so that wedge-shaped consonants
and vowels could be measured. (Appendix 4)

The instrument described above was first drawn in enlarged
size with the largest circle approximately 43 cm in diameter. The
drawing was then reduced to the smallest size possible without
blurring of lines. Such a reduction in size was necessary in order
to insure the maximum accuracy in measurement. In the reduced draw-
ing the largest circle was 3 cm in diameter; hence, the drawing was
reduced to l/14 of its original size.

Kodak microfilm was used in preparing the reduced plate.
Kodak lantern slides were found to be the best for prints, since they
are durable, transparent and easy to use. It should be noted that
the thickness of the slide can cause visual distortion unless the
slide is viewed directly from above.

In use the slide was placed over the element to be measured
and measurement was made in the arbitrary units on the slide. In
the development of the instrument no effort was made to use absolute
units such as inches or centimeters. While the use of absolute

units would be desirable for police records and similar uses it was

29

felt that it was not necessary in this study because the same
instrument was used for all observations and, as a result, the

unit of measurement was held constant.

CHAPTER IV
ANAEYSIS AND FINDINGS

The statistical analysis discussed in this chapter was made
using measurements of the vertical stroke (X1), the horizontal
stroke (X2) and the angle between strokes (X3) of the Korean vowel

F.(a). The writings of eight Korean students were the source of
the characters measured. Two sets of samples from these writings
were analyzed. In the first set 50 characters were selected at
random from each of these writings; in the second set 10 characters
were selected randomly. All of the characters used in the analysis
were originally classified in class pattern 1. The purpose of the
analysis was to determine whether the authorship of these samples

could be determined statistically.

The Major Analysis

 

When pairs of writings are compared in handwriting analysis
there are two possibilities: either the writings were made by the
same individual or they were made by different individuals. The
problem confronting the handwriting examiner is to determine which
of these alternatives is true. If statistical analysis is to be use-

ful in problems of this sort it must be equally capable of determining

3O

 

t.

31
similarity or difference. In the part of the statistical analysis
discussed in this section both situations were tested. For the test
of difference each individual was compared with every other individual
for a total of 28 comparisons. For this tesn samples of 50 charac-
ters were used. For the test of similarity,each sample was divided
randomly into two samples of 25 and comparisons were made between
these pairs of samples for each individual.

The statistical test used was the t-test for differences
between means. 'This test was selected since the data used in the
analysis meets the necessary conditions of interval scale, normal
distribution and equal variance. Providing that these conditions can be
met, the t-test is the most sensitive statistical test which can be used
for comparison of pairs of samples.1

In statistical tests used to determine whether two samples
were drawn from the same uniwerse it is the usual practice to phrase
this question in the form of a "null hypothesis" (symbolized as HO).

Ho: There is no significant difference between the means
of the two samples. The null hypothesis can be paraphrased in the
following way: "Any difference between the means of the two samples
is the result of sampling variability and not the result of the
fact that the samples were drawn from different universes." After
stating the null hypothesis the statistician establishes a "region
of rejection" or "critical region" which is stated in probability

terms. If the researcher selects a probability of .05 as the critical

 

1Blaylock, _En cit., Chapt. 13.

32

region, for example, he will reject the null hypothesis if the
difference between his samples could have occurred by chance less
than 5 times out of 100, as a result of sampling variability. If
he rejects the null hypothesis he accepts the alternative hypothesis
that the samples were drawn from different universes. If the dif-
ference between his samples could have occurred more than 5 times out
of 100 as the result of sampling variability he accepts the null
hypothesis and assumes that sampling variability did, in fact, cause
the difference. The choice of a critical region depends on the
nature of the study (and sometimes on how conservative the researcher
is). A statistical test, such as the t-test, is used to determine
the probability of the difference occurring as a result of sampling
variability.

In the various analyses discussed in this chapter the following
null hypothesis was used:

Ho: Pairs of writing samples were made by the same individual.
The critical region was set at .05 and two-tailed tests were used in
all cases.2 In terms of the present analysis the acceptance or rejec-
tion of the null hypothesis must be interpreted in two ways. In
tests where the samples being compared were made by different individ-
uals, a probability which allows the null hypothesis to be rejected
means that the identification has been made correctly--that the

writings are from different universes. If the null hypothesis must

 

2A discussion of one and two-tailed tests will be presented
later in the chapter.

33

be accepted an incorrect identification has been made. Where the
samples being compared were made by the same individual the reverse
is true--the rejection of the null hypothesis means an incorrect
identification and the acceptance of the null hypothesis means that
the identification has been made correctly since the samples were,
in fact, drawn from the same universe.

The results for pairs of samples made by different individ-
uals are found in Table 4.1 and the results for pairs of samples made
by the same individual are found in Tables 4.2 and 4.3. In these
tables incorrect identifications are indicated by asterisks. Table

4.4 summarizes the results of both sets of tests.

Tests Using Small Samples

 

In practical handwriting analysis the examiner may have
available only a small piece of writing with a relatively limited
number of characters. In order to determine the efficiency of sta-
tistical analysis in situations of this kind statistical tests were
made between pairs of samples, made by different individuals, with
10 characters in each sample. The null hypothesis, region of rejec-
tion and statistical test were the same as for the larger samples.
Table 417 presents the results of this test. For the smaller samples
60.71 percent were correct for X1, 35.71 percent for X2 and 39.28

percent for X3.3

 

3When groups of small samples are analyzed it is considered
good statistical practice to compute an F-test for the entire set of
samples. For the samples in this part of the study;the samples for
all three elements were significantly different at the .01 level.

 

I;

34

Table 4.1. T-Tests for Pairs of Samples
for Different Writers (N=50)

 

Pair

 

 

1 2 3
1-2 5.55 7.16 .65*
1-3 l.00*(a) 4.84 1.30*
1-4 .65* 5.62 2.40
1-5 14.86 26.66 7.36
1-6 22.88 2.17 .23*
1-7 2.15 1.38* .70*
1-8 7.92 10.69 3.20
2-3 26.20 11.59 2.78
2-4 5.76 .93* 4.35
2-5 9.81 16.07 8.55
2-6 18.80 9.06 1.21*
2—7 1.68* 4.27 .05*
2-8 3.96 3.38 2.55
3-4 1.57* 8.66 1.57*
3-5 15.77 1.74* 10.50
3-6 25.64 2.71 1.57*
3-7 2.78 5.19 2.35
3-8 9.00 14.97 5.29
4-5 13.08 10.45 11.60
4-6 22.51 6.96 3.14
4-7 1.75* 4.01 3.64
4-8 7.17 1.34* 8.00
5-6 14.42 5.19 9.40
5-7 6.97 7.27 7.72
5-8 3.89 20.25 5.45
6-7 15.33 3.04 1.05*
6-8 12.48 12.83 4.85
7-8 4.02 7.03 2.93
d.f.= 49 t.05= 2.02
(a) = incorrect t.Ol= 2.70
identifications t.001=3.55

and marked with
an asterisk.

 

35

 

 

 

mm. N~.m mm. ow.m mm. #N.m wo.H «5.5 ow.H qm.m om.H mo.m w
mo. ow.~ do.H Nn.N qH.H w¢.o mm.H NH.© no.0 oh.m oo.N oo.¢ n
so. NB.N ow. mm.N mo. om.m am. ow.m 0N.H on.oH mm. do.w o
mo.a om.¢ we. om.¢ om. mm.¢ mo. qw.¢ wo.H mm.n mN.H «N.m m
on. om.N ad. mw.~ Nw.H do.o om.H «N.w mo.H wm.m HN.H ow.m q
ow. qm.m om. wo.~ mo. NH.m mm. oo.m mm. NN.m mo.H om.m m
mo. mm.~ an. ww.~ mo.H Nn.n qw.a mm.o Hm.H mo.q OH.H mw.¢ N
¢¢.H om.~ mm. wm.~ ma. oo.o mm. om.m mm. q¢.m mH.H w¢.m H
m M m M m M m w m M m M Hmapfi>fich
N

 

 

Ammuzv HmsvH>HwGH 68mm osu Mom
monEmm mo mummm mom mnoauofi>on vuwwawum was memo:

.N.¢ oHnoH

36

Table 4.3. T-Tests for Pairs of Samples

for the Same Writers (N=25)

 

 

Individual

 

1 2 3
1 .13 .15 1.19
2 2.23*(a) 1.63 1.73
3 .29 .43 2.19*
4 1.31 2.70* 2.334
5 2.80* .45 .79
6 6.65* 1.69 2.02
7 1.61 .95 .07
8 .27 2.00 .08

 

d.f.= .24; t.05= 2.06; t.01= 2.80; t.001= 3.55

(a) incorrect identifications are indicated by an

 

 

asterisk
Table 4.4. Correct Identification
X1 X3
Type N Z N Z N Z Total N
Different
Individuals 23 82.14 24 85.71 19 67.85 28
Same
Individual 5 62.50 7 87.50 6 75.00 8
Both Types 28 77.77 31 86.11 25 69.44 36

 

37

Table 4.5. Means and Standard Deviations
for Small Samples (N=10)

 

l _ 2 _ 3
Individual X s X s X s
l 3.55 1.19 2.05 .47 1.80 .72
2 4.40 1.09 2.30 1.03 2.40 1.19
3 2.80 .95 2.10 .89 1.68 .96
4 4.95 1.69 1.95 .57 2.80 1.49
5 5.93 1.02 3.60 1.09 1.81 .58
6 9.45 1.71 2.15 .59 4.64 1.17
7 3.95 1.89 3.75 2.45 1.73 1.26
8 2.85 1.24 2.25 .51 1.35 .62

 

Discussion of Findings

 

An examination of the results of the analysis points to two
general conclusions about the use of statistical inference in hand-
writing analysis. First, it is evident that, in the case of samples
of writing which are fairly large, a rather -high degree of accuracy
in determining similarity and difference is possible. As the sample
size decreases accuracy also decreases.

Second, it must be stressed that the technique is not com-
pletely accurate for any of the elements measured even for the
larger samples. Because a high degree of accuracy is vital in the
legal application of handwriting analysis,methods for improvement
must be developed. Accuracy may be increased in part by combining
macro-analysis with micro-analysis, in part by improvements in the
micro-analysis. A detailed analysis of the statistical findings suggest

several ways in which micro-analysis can be improved.

38

1. The use of more than one character as the basis for
analysis. An examination of the mean measurements in Table 4.6
shows that some of the individuals are rather close in their mean
measurements for a particular element. Normal handwriting falls
within a restricted size range and it can be expected that a num-
ber of individuals will have the same mean size and range of va-
riability in a given letter element. The figures in Table 4.1
indicate, however, that there is no pair of individuals which is
not significantly different in at least one of the measurements.
This fact indicates that individuals may be more unique in their over-
all writing than they are in terms of a single character and suggests
that more than one character should be analyzed in applied hand-
writing examination.

Table 4.6. Means and Standard Deviations
for Individual Samples (N=50)

 

X X X
Individual X 1 s X 2 s X 3 s
l 3.46 1.06 5.98 .91 2.57 1.17
2 4.57 .94 7.32 .97 2.70 .76
3 3.26 .97 5.06 .99 2.31 .79
4 3.59 1.17 7.59 1.81 2.09 .66
5 6.73 1.26 4.78 .55 4.41 1.28
6 9.67 1.48 5.58 .94 2.53 .61
7 4.15 2.01 6.30 1.37 2.71 1.01
8 5.60 1.57 7.98 .97 3.21 .84

 

39

2. More accurate measurement. Some of the apparent similarity

 

between individuals may be the result of inaccuracy or insufficient
refinement in measurement. It can be suggested that improvements
should be made in the instrument in order to permit more precise
measurement and in order to permit the measurement of curvature which
is an important part of the characteristics of a stroke.

3. Control of Type I and Type II errors.4 Because statistical

 

inference is based on probability rather than certainty it is possible
that a decision about the null hypothesis may be in error. There
are two kinds of error which can be made:

(a) Type I error - the rejection of the null hypothesis when
it should be accepted. In terms of handwriting analysis this means
that the examiner concludes that the writing was made by different
individuals when it was actually made by the same individual. The
errors found in Table 4.3 are of this type.

(b) Type II error - the acceptance of the null hypothesis
when it should be rejected. Again, in terms of handwriting analysis
this means that the examiner concludes that the writing was made
by the same individual when, in fact, it was made by different in-
dividuals. Errors in Tables 4.1 and 4.7 are of this sort.

There are several approaches to increased identification
accuracy based on an analysis of the two types of errors. Type I
error can be reduced simply by raising the region of rejection. If
the region of rejection were set at .01 instead of at .05 (which was

used in this study) only one error instead of 6 would have occurred

 

For a discussion of types of error and the power efficiency
of statistical tests see Blaylock, 22, cit., Chapter 14.

40

Table 4.7. T Tests for Pairs of Small Samples
for Different Writers (N=10)

 

 

 

t
Pair X1 X2 X3
1-2 1.66* .70* 1.37*
1-3 1.55* .16* .31*
1-4 2.14* .43* 1.91*
1-5 4.80 4.13 .34*
1-6 8.95 .42* 6.54
1-7 .56* 2.15 .12*
1-8 1.28* .91* 1.50*
2-3 3.49 .47* 1.48*
2-4 .86* .99* .66*
2-5 3.24 2.74 1.43*
2-6 7.88 .40* 4.25
2-7 .65* 1.72* 1.22*
2-8 2.96 1.37* 2.48
3-4 3.49 .45* 1.99*
3-5 7.08 3.38 .34*
3-6 10.74 .15* 6.19
3-7 1.72* 2.00* 1.00*
3-8 .010* 4.62 .91*
4-5 1.57* 4.24 1.96*
4-6 5.91 .76* 3.07
4-7 1.35* 2.26 4.73
4-8 3.15 1.25* 4.84
5-6 5.60 3.69 3.69
5-7 2.92 .18* 1.76*
5-8 6.05 .35* 3.50
6-7 6.82 3.54 3.70
6-8 9.87 3.70 3.54
7-8 2.45 1.89* 1.89*
d.f.= 9 t.05= 2.26
t.01= 3.25

(a) asterisks indiggfg4ihéorrect identification

41
in comparisons of writings made by the same individual as shown
in Table 4.3. This would, however, have increased the number of
Type II errors in Table 4.1 from 18 to 25.

One method of reducing both types of error is to raise the
region of rejection in order to reduce a Type I error and increase
the sample size in order to reduce a Type II error. The reduction of
Type II error by increasing sample size can be seen in the improve-
ment in accuracy of samples of 50, Table 4.1, over samples of 10
in Table 4.7. It can be suggested then that the examiner use a high
region of rejection (at least .01) and use the largest sample possible.

An additional reduction in Type II error can be made by
using a one-tailed test. In a two tailed test, the type used in this
study, the direction of differences between means was not specified.
More specifically, when two samples were compared the test was made
to include both the probability that the sample mean was below the
universe mean and the probability that the sample mean was above
the universe mean. If the universe mean is known, then a one-tailed
test in which the direction of the sample mean is specified. Such
a test will have a smaller probability of Type II error than a one-
tailed test for the same sample size. The use of a one-tailed test
might be possible if a sample of questioned writing is being compared
with a large standard writing of known authorship. This situation
might be possible if criminal investigation bureaus kept quantita-
tive records of writing in the same way that fingerprint records are

kept.5

 

51bid., chapter 14.

42

It is apparent that statistical inference can determine the
identity of writing with a relatively high degree of accuracy al-
though not with absolute certainty. The probability of error in
identification can be greatly reduced by the use of methods suggested

above.

CHAPTER V

CONCLUSIONS

The research reported in this paper was an exploratory
study of the possibility of using inferential statistical methods in
handwriting analysis. Two inherent characteristics of handwriting--
variation within the writings of a single individual and overlap in
writings made by different individuals--make handwriting identifica-
tion problematic. Current methods of analysis do not completely
solve the problems presented by these factors because, while they
do consider similarity and difference, they do not provide objective
criteria for deciding how much similarity there must be before it
can be concluded that samples of writing were made by the same per-
son or how much dissimilarity there must be before it can be con-
cluded that samples were made by different persons. The technique
of inferential statistics was developed in order to solve the same
type of problem in other areas of inquiry and an examination of the
conceptual and mathematical structure of this technique suggests
that it can be legitimately used in the area of handwriting analysis.

The present study was limited in scope in several ways. The
study was based on the measurement of elements of a single character.
The measuring instrument developed by the writer for the study was

restricted to line and angular measurement and was not capable of

43

44

measuring stroke curvature. Small samples were used. In spite of
these limitations the statistical tests used led to correct iden-
tifications in 69 percent of the cases for the least discriminating
element and in 86 percent of the cases for the most discriminating
element. Further, it has been shown that the accuracy of identifica-
tion based on a single element can be improved by increases in sample
size and by changing the region of rejection of the null hypothesis.
The findings, then, seem to demonstrate that the techniques of
statistical inference hold great promise for improvements in hand-
writing identification.

The present study was exploratory and was not designed to
develop a finished method of identification. The development of a
finished method which can be used in criminal investigation can only
result from continued research in this area. On the basis of the
preliminary research made for the present study and the findings
reported in the preceeding chapter the writer would suggest that
research for the development of a mature statistical handwriting
identification technique should include the following things:

1. The development of better measuring instruments. It would be

 

particularly desirable to develop instruments which could measure
curvature and stroke width and which were calibrated in standard
linear measurements such as fractions of a millimeter. In the area
of measurement,methods for controlling the measurement error,which

would vary from examiner to examiner should be developed.

45

2. The combination of micro-analysis and macro-analysis.

 

The analysis of class-patterns--the overall stylistic characteris-
tics of writing--should be combined with the measurement of charac-
ter elements. The development of a technique for quantifying macro-
characteristics would be an important step in the development of a
precise handwriting system.

3. The study of writing in various languages. The present

 

study was based on the Korean alphabet. If the technique of statis-
tical analysis is to have general applicability, the factors of varia-
bility and coexistence of similarity must be studied in other languages.
Although it can be assumed that the methods used in the present study
are applicable for Other languages such an assumption must be tested
before a universal system of identification can be developed.

4. The investigation of statistical analysis in disguised
writing. The present study made no attempt to examine disguised
writing. It is apparent that a great number of the cases processed
by handwriting examiners may include disguised writing. For this
reason the usefulness of statistical analysis in disguised writing
must be thoroughly investigated.

5. The development of techniques for multiple character
analysis. One of the findings of the present study suggested that
identification can be more accurate if it is based on the examination
of several characters. A mature identification system should probably

include techniques for quantitatively combining various elements.

46

6. Exhaustive study of writing variation and coexistence.

 

Before statistical identification methods can be completely developed
more data concerning patterns of variability, overlap of character-
istics and internal combinations must be collected and analyzed. In
other words, a great deal more must be known about the universe of
handwriting before a mature technique can be developed. This could
be facilitated if criminal investigation departments in various
countries would colleCt and classify handwriting specimens in the same
way that they collect fingerprints.

. The writer feels that the use of statistical methods holds
great promise for handwriting identification and is hopeful that
more research in this area can lead to the development of a scientif-

ically and legally acceptable handwriting identification system.

BIBLIOGRAPHY

A. Books

Blalock, Hubert M. Jr., Social Statistics, New York: McGraw-Hill
Book Company, Inc., 1960.

 

Dixon, W. J. and F. J. Massey, Jr., Introduction to Statistical
Analysis, New York: MeGraw-Hill Book Company, Inc., 1951.

 

Eisenhart, C., M. W. Hastay, and W. A. Wallis, Techniques of Sta-
tistical Analysis, New York: MeGraw-Hill Book Company, Inc.,
1947.

 

 

Harrison, Wilson R., Suspect Documents: Their Scientific Examination,
New York, Frederick A. Praeger Inc., 1958.

 

Hilton, Ordway, Scientific Examination of Documents, Chicago: Callaghan
& Company, 1956.

 

McCarthy, P. J., Introduction to Statistical Reasoning, New York:
MeGraw-Hill Book Company, Inc., 1957.

 

McCollough, Celeste and Loche Van Atta, Statistical Concepts: A
Program for Self-Instruction, New York: MeGraw-Hill Book
Company, 1963.

 

 

Mood, A. M., Introduction to the Theory of Statistics, New York:
MeGraw-Hill Book Company, Inc., 1950.

 

Osborn, Albert S., The Problem of Proof, Sixth Printing, Albany, N.Y.:
Boyd Printing Co., 1947.

 

 

, Questioned Documents, Second Edition, Albany, N.Y.:
Boyd Printing Co., 1946.

 

 

& Albert D. Osborn, Questioned Document Problems,
The Discovery and Proof of the Facts, Albany, N.Y.: Boyd
Printing Company, 1944.

 

 

Savage, L. J., The Foundations of Statistics, New York: John Wiley
and Sons, Inc., 1954.

47

48

Sellitz, Claire, et a1, Research Methods in Social Relations,
2nd Ed., New York: Henry Holt, 1959.

Somprasongk, Prathnadi, "Identification of Thai Handwriting: A
Classification of Common Letter and Numerical Forms in Four
Hundred Standards," An unpublished M.S. thesis, University
of California, Berkeley, Cal., Oct. 13, 1960.

Walker, Helen M., and Joseph Lev, Statistical Inference, New York:
Holt, Rinehart and Winston, 1953.

 

Wallis, W. Allen, and Harry V. Roberts, Statistics: A New Approach,
Glencoe: The Free Press, 1956.

Young, Pauline, Scientific Social Surveys and Research, Englewood
Cliffs, N.J.: Prentice-Hall, 1956.

 

B. Articles

Albarracin, Roberto, Peritajes Judiciales (Elements of document
examination), Revista de la Mutualidad de la Policia Federal
(Argentine), An 28, no 322, mars 1953, p. 40.

Bischoff, M.A., "The Possibilities to Improve Writing Comparison
Evidence," Revue Internationale de Criminologie et de Police
Technique (Suisse), Vol. 9, No. 4, Oct.-Dec. 1955, p. 273
(14,1).

Clark, P. F., "Classification and identification of criminal hand-
writings," The Australian Police Journal, Vol. 6, No. 4, Oct.
1952, p. 276.

 

Conway, J. V. P., "The Identification of Handwriting," JCLC and PS,
Vol. 45, No. 5, January-February, 1953.

 

Domenici, Folco, Valore et limiti dell' esame peritale dei mano-
scriti (The Value and limits of examination of written docu-
ments), La Giustizia Penale (Italie), An 58, no. 4, avril
1953, I, p. 163.

 

Dvorak, A., "Zur Frage der Beweiskraft individueller Merkmale,"
(The Probative Value of Individual Characteristics), Arch,
f. Krim., 122 (3/4:90-100, Sept.-Oct.

Filho, Jose Del Picchia, Metodo grafoscopico universal ("Une methode
graphoscopque universelle--A universal method of handwriting
examination"), Inverstigacoes(Brazil), An 4, no 41, mai
1952, p. 75.

 

49

 

, "Que es documentoscopia (La documentoscopie-
Examination of documents)," Investigacion(Espagne), An 20,
no 294, Oct. 1952, p. 77.

 

 

, Resenha bibliografica da perxcia de docu-
mentos (A bibliography of document examination), Investigacoes,
An 4, no 43, juillet 1952, p. 95.

Goyle, D. N., "Identification of Handwritings," The Penal Reformer,
(Inde), Vol. 8, No. 4/5, Juin-Juillet 1952, p. 86, et no
6/7, Aout- Septembre 1952, p. 147.

Harris, J., "Disguised Handwriting," J. Criminal Law Criminol. Police
Science, Vol. 43, No. 5, January-February, 1953.

 

Harris, John J., "How much do people write alike," J. Criminal Law
Criminol. Police Science, Vol. 48, No. 6(1), pp. 647-651.

 

 

Hilton, Ordway, "Can the Forger be Identified from His Handwriting?"
J. Criminal Law Criminol. Police Science, Vol. 43, No. 4,
Nov.-Dec., 1952, pp. 547-555.

 

 

, "The Collection of Writing Standards in Criminal In-
vestigation," J. Criminal Law Criminol. Police Science, Vol. 32,
No° 2, July-August, 1942, pp. 241-256.

 

 

, "Der Beweis fur die Echtheit eines Schriftstuecks,"
(Proof of Genuineness), Kriminalistik, 12(11):459-462, Nov.

 

 

, "Proper Evaluation of Dissimilarities in Handwriting,"
International Criminal Police Review, 104, Feb. 1957, pp. 48-
51.

Livingston, Orville, B., "A Handwriting and Pen-printing Classification
System for Identifying Law Violators," J. Criminal Law Criminol.
Police Science, Vol. 49, No. 5, Jan.-Feb., 1959 pp. 487-506.

 

Malley, R., "Identification and Handwriting," International Criminal
Police Review, No. 94, Jan., 1956.

 

 

Serrano, Pedro, "Grafistica (Elements de documentoscopie(Essentials
of document examination)," Policia(ESpagne), No. 128, Oct. 1952,
supplementa et No. 129, Nov. 1952.

 

, Grafistica(The elements of document examination),
Policia(Espagne), an 12, no 131, janvier 1953, supplement
et no 132, fevrier, supplement.

 

, Grafistica, (The elements of document examination),
Policia, An 12, no 132, avril 1953.

50

 

, Grafistica, "Elements of document examination,
Policia, an 12, no 137, juillet 1953, no. 138, aout 1953,
no. 139, septembre 1953.

SjBegren, Tore, "Handwriting Comparison and Probability," International

 

Criminal Police Review, Vol. No. 92, Nov. 1955, pp. 274-283.

Smith, Leh Theodora, "Six Basic Factors in Handwriting Classifica-
tion," J. Criminal Law Criminol. Police Science, 44, 1954,
pp. 810-816.

Smith, Stanley 8., "A Method of Comparing Written Documents,"
International Criminal Police Review, Aug.-Sept., 1954, pp.
205-215 0

Wrenshall, A. F. and D. M. Duke, "Statistical Methods and the Examina-
tion of Questioned Documents," Seminar No. 4, The Examination
of Questioned Documents, RCMP Crime Detection Laboratories,
May 10-11, 1956, pp. 89-98.

 

Appendix 1

Sample Text

“7:77 W4! #777? %

74:7

‘747 7'7; 7777 4747.77 7477 37.77
9171/ :1“ 1’]: 112F173. «37.7.4.7 Vol V7373! 74/77 7:13.
#47": 375’"! 7753?} *1/71} 744 it: lab—77%,
17/73: 517,47; r717 «Ia/>7 77,777 77447?)
'4 7 “7 77:37" 777—2, 4777 7777721

37,7}; 77:14.9. 7 71:. 77; “44,44,714

P771717: 44:1 71453473— 3127/7

val-+474 47777“ 711.71 777773717777

J‘f 97.1% «I 774 7717—7 77102.3- 737,17-

7-3! 77 7+ .77 7747 7+7 7 :7 747777
71:: 77‘; 777—77 :17 777 {~77 2% 47777

“7" 57-1477 3 1
”1 “I I 74177:: I774 777:“
3’14“: 774‘ new 77 7: 71’”
7 ., ,
WW7; j' “I 442 #73 “7%, “4143:! figi—d 1:71’7’}
74M; £3741: $775}

777 77777 444' 777 074% 77‘
7777 7 I77 777, rI-a 2.77 77 777777 77
)4; 1'7 "1‘7”, 7’“ : :Z’i‘WJ/Wef” #:1117136
71/7. wag-7771.77 «77./a 47774771

5' sI 71v ‘1” 57427717“) ”I’M—‘- ‘V/‘M'J ‘Pf
%3 vii I ’1’! W777} t’d'fiéw
d 691 [477-3177

.9771: a} 73/4 Ifl/d 6')! to («I it!

7:147 “77777 77.27744
}fl .7“ a"

Appendix 2

Major Divisions of Class Pattern

‘1 'l ‘ II ._ I 4. , . . _ \

. w 1 ,t\ 1 I- . . .
«W-_Q\‘ K§‘\.uulltw0\\1\\hfl\ \\.\.u\\.nm‘\ VAI\..\Q\M|\\AU

54

Q\

a , A.:_,..H%;..

I

. Kim m

\VQ “\Afib \M4DQ&N\» IIIII !-.--l:\ \Qxah. - .1 -:l-

 

A A

"u’s m the (icntimcn-r

.mwlunuag) 014) 0) 3““"1111

 

 

 

 

 

 

 

 

 

 

 

 

 

 

:3“
qvmx

 

122$-

 

27L

 

55-

 

§m M\§\m§§\ Q§§§N
SSK .1 wamxfi £5

22 “11.1

 

 

 

17
\

. V

.
1.
.

 

111121.

1...] .. . :5... .Li. .5. g 1. 9....1HH1L1 24-..}.
2.2+ 1L. L L3 11L. _
LL. .2]... . .2. L L 2.11
L..L.2 ... 12L. l1..11LH2,2 .. .. L1 2 _ L L.2L1 . m 2.1.1
L L V. . L . L L L 2.
.1411111.L11~14‘.1~2 LL,1 L .1. I1.1.4 .. .L 11. Ill
L . .1L1.. .I L ,L112L1 . L . .11
x24. £1 23 L 2 :21
”M L .L. . 12.. .. . LHMM
.2 §Rn 1_ m . L .1 .211
IL112.22..LL1L11 1.1 L1111L .1121

. L L

 

 

#2..
.
.L.
L. .
L112
L
L
L
L

 

 

 

 

 

 

 

L L L
.1L 2
2L LAN: .
. 22m- ..L.HL2..2..M 1
2 L m, LL . L

 

 

 

 

RV 1 3»

 

é.

 

 

mi Z 1

 

 

 

 

 

 

fink g ESL 3. \. %

 

 

 

 

3:11,...1

 

 

 

 

Appendix 3

Application of the Mbasurement Plate

57

In the measurement of the lines and angles of the charac-
ter analyzed in this study the following rules have been used:

1. Maximum Extension Rule (MER)--The measuring instrument
is placed on the line to be measured.

2. Inner Line Rule (ILR)--The measuring instrument is placed
inside the line to be measured.

3. Maximum Angle Rule (MAR)--As in the maximum extension rule
the measuring instrument is placed on the two lines which
intersect.

4. Inner Angle Rule (IAR)--the instrument is placed inside
the angle to be measured.

To illustrate the steps for application of the measurement
plate and the aforementioned rules:

   

a.
4- c
Note: (1) LINE-UP (2) LINEAR MEASUREMENT (3)ANGULAR
ab-Xl-Z MEASUREMENT
bc=X2=l abc=X3=8

red lines=1ines
drawn according
to the rules

black lines-lines
on the plate

All three steps may be illustrated:

 

Appendix 4

Measurement Plate CEnlarged Positive)

59

 

*ROOM use out!

i A] N UT;

l’ "90

 

 

    

.girflujnflmr

WIIHWIIII

293 0304

31

IIHIHWIHHIIIUIIH