statistical evaluation of morphological data of japanese head hair and the screening of evidential...

13
Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis Hajime Sato * National Research Institute of Police Science, Kashiwanoha 6-3-1, Kashiwa, Chiba 277-0882, Japan Received 9 October 2001; received in revised form 11 March 2002; accepted 15 March 2002 Abstract Intra-individual and inter-individual variations in Japanese head hair morphology were reevaluated and the screening of evidential hair samples using a statistical method was investigated. A quantification of the morphological data was used as a system in which morphological features obtained by macroscopic and microscopic observations can be objectively processed in the form of numerical information. For confirming the availability of morphological features for hair comparison, the inter- individual comparison of ten male Japanese was investigated by discrimination analysis using 18 variables including five original numerical morphological features (variables) and 13 variables obtained from six morphological features by quantifica- tion. The majority of the comparisons showed a high level of discrimination between two individuals. For morphologically narrowing down the number of evidential hair samples collected from a crime scene, a screening method was investigated by cluster analysis using 18 variables. In an experimental model of evidential hair samples derived from 20 individuals, hair samples from 13 individuals were successfully discriminated using the cluster analysis described in the present investigation. q 2002 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Forensic hair comparison; Morphological examination; Cluster analysis; Screening of evidential hair 1. Introduction The evaluation of hair morphological examination and the development of new technologies for forensic hair investigations have been studied during the 1970’s and 1980’s in an effort to improve their accu- racy [1–11]. However, those studies have not produced satisfactory results with regard to improving the objectivity of forensic hair comparison. Gaudette and coworkers attempted to experimentally evaluate the probability of human hair comparison based on morphological features, using both macroscopic and microscopic methods [12,13]. These studies concluded that the significance of their studies was in the experimental proof of the proposition that macroscopic and microscopic hair comparison is a useful technique and constitutes valid evidence [14]. To provide actual probability numbers in a hair comparison based on morphological examination is not satisfactory. However, hair examiners have continued to use Gaudette’s studies as the background for interpreting the significance of hair evidence. Higuchi et al. reported that nuclear and mitochon- drial DNA (mtDNA) sequences can be detected by use of the polymerase chain reaction (PCR) from a single hair shaft [15]. Since then, the introduction of DNA technologies has been considered for improving the accuracy of a forensic hair comparison [16–25]. However, it is difficult to routinely apply nuclear DNA detection methods to evidential hairs since Legal Medicine 4 (2002) 90–102 1344-6223/02/$ - see front matter q 2002 Elsevier Science Ireland Ltd. All rights reserved. PII: S1344-6223(02)00010-X www.elsevier.com/locate/legalmed * Tel.: 181-471-35-8001; fax: 181-471-33-9153.

Upload: hajime-sato

Post on 05-Jul-2016

214 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

Statistical evaluation of morphological data of Japanese head hairand the screening of evidential hair samples by cluster analysis

Hajime Sato*

National Research Institute of Police Science, Kashiwanoha 6-3-1, Kashiwa, Chiba 277-0882, Japan

Received 9 October 2001; received in revised form 11 March 2002; accepted 15 March 2002

Abstract

Intra-individual and inter-individual variations in Japanese head hair morphology were reevaluated and the screening of

evidential hair samples using a statistical method was investigated. A quantification of the morphological data was used as a

system in which morphological features obtained by macroscopic and microscopic observations can be objectively processed in

the form of numerical information. For confirming the availability of morphological features for hair comparison, the inter-

individual comparison of ten male Japanese was investigated by discrimination analysis using 18 variables including five

original numerical morphological features (variables) and 13 variables obtained from six morphological features by quantifica-

tion. The majority of the comparisons showed a high level of discrimination between two individuals. For morphologically

narrowing down the number of evidential hair samples collected from a crime scene, a screening method was investigated by

cluster analysis using 18 variables. In an experimental model of evidential hair samples derived from 20 individuals, hair

samples from 13 individuals were successfully discriminated using the cluster analysis described in the present investigation.

q 2002 Elsevier Science Ireland Ltd. All rights reserved.

Keywords: Forensic hair comparison; Morphological examination; Cluster analysis; Screening of evidential hair

1. Introduction

The evaluation of hair morphological examination

and the development of new technologies for forensic

hair investigations have been studied during the

1970’s and 1980’s in an effort to improve their accu-

racy [1–11]. However, those studies have not

produced satisfactory results with regard to improving

the objectivity of forensic hair comparison. Gaudette

and coworkers attempted to experimentally evaluate

the probability of human hair comparison based on

morphological features, using both macroscopic and

microscopic methods [12,13]. These studies

concluded that the significance of their studies was

in the experimental proof of the proposition that

macroscopic and microscopic hair comparison is a

useful technique and constitutes valid evidence [14].

To provide actual probability numbers in a hair

comparison based on morphological examination is

not satisfactory. However, hair examiners have

continued to use Gaudette’s studies as the background

for interpreting the significance of hair evidence.

Higuchi et al. reported that nuclear and mitochon-

drial DNA (mtDNA) sequences can be detected by

use of the polymerase chain reaction (PCR) from a

single hair shaft [15]. Since then, the introduction of

DNA technologies has been considered for improving

the accuracy of a forensic hair comparison [16–25].

However, it is difficult to routinely apply nuclear

DNA detection methods to evidential hairs since

Legal Medicine 4 (2002) 90–102

1344-6223/02/$ - see front matter q 2002 Elsevier Science Ireland Ltd. All rights reserved.

PII: S1344-6223(02)00010-X

www.elsevier.com/locate/legalmed

* Tel.: 181-471-35-8001; fax: 181-471-33-9153.

Page 2: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

hairs encountered at crime scenes are nearly always in

the telogen phase and contain limited quantities of

DNA [15,17–19,21]. Since the detection of sequence

polymorphism of mtDNA is now possible by using

PCR amplification and direct sequence methods

[26,27], hair examiners have concluded that mtDNA

analysis is the most suitable DNA typing method for

hair samples and the best approach to the genetic

typing of hair samples [17,19,20,23–25]. However,

an examination of the mtDNA sequences of all

evidential hairs is not realistic, since this analysis is

expensive and requires considerable time. For intro-

ducing mtDNA analysis to evidential hairs in case-

works, it is necessary that hairs collected from crime

scenes are narrowed down via an initial morphologi-

cal screening.

For this purpose a quantification of the morpholo-

gical examination data classified and a screening

method by cluster analysis were investigated as a

possible system in which morphological features can

be objectively processed in the form of numerical

information. Moreover, intra-individual and inter-

individual variations of Japanese head hair morphol-

ogy were statistically reevaluated prior to the statisti-

cal investigation mentioned above.

2. Materials and methods

2.1. Materials

Twenty five hairs were collected from each of ten

adult Japanese males (M1–M10) and females (F1–

F10). These hairs were used in the previous reports

in which intra-individual and inter-individual varia-

tions of Japanese head hairs were investigated [28–

34]. Five hairs were taken from each of five anatomi-

cal locations on the head of each person, that is, front,

top, back, left temple and right temple.

2.2. Morphological examination

Eleven features were adapted for statistical evalua-

tion of Japanese head hair. Five numerical features

were re-examined using methods in previous reports

[29,31,33].

1. The length was measured with a ruler by straigh-

tening a hair sample.

2. The diameter was measured using a microscope

equipped with a micrometer.

3. The medulla index is a ratio of medullary diameter

to hair diameter.

4. The scale count is the number of free edges of hair

scale per unit length (100 mm).

5. The hair index is a ratio of minimum diameter to

maximum diameter in cross section.

Six morphological features were re-examined using

methods in previous reports [28,30–32,34].

1. The general form was classified into six character-

istics; straight, arch, hemi-circle, slight wave, wave

and other (complicated wave, etc.).

2. The macroscopic color was classified into four

characteristics; black, brown-black, brown and

white.

3. The tip form was classified into four characteris-

tics; needle-shape, transverse cut, oblique cut and

other (split, crushed, frayed, etc.).

4. The medulla appearance was classified into five

characteristics; absent, dotted, fragmental, contin-

uous and unobservable.

5. The scale pattern was classified into four charac-

teristics; slight flat wave, flat wave, irregular wave

and extremely irregular wave.

6. The cross section form was classified into five

characteristics; circle, oval, ellipse, pear-shape

(kidney-shape) and triangle (quadrangle).

2.3. Evaluation of intra-individual variation

Eleven morphological features were evaluated as

follows. Intra-individual variations of six morpholo-

gical features classified were evaluated for three

grades. [2] indicates that the ratio of the main char-

acteristic to the same individual is more than 70%,

signifying no variation. [1] indicates that the ratio

of the main characteristic to the same individual is

from 50 to 70%, indicating a small variation. [11 ]

indicates that the ratio of the main characteristic to an

individual is below 50%, indicating a large variation.

Based on the coefficient of variation (CV), the intra-

individual variation of five numerical features was

evaluated as follows; [2]: CV % 10.00, [1]: 10.00

,CV % 20.00, [11 ]: 20.00 ,CV. The variation

H. Sato / Legal Medicine 4 (2002) 90–102 91

Page 3: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

among five anatomical locations on the head of the

same individual was evaluated by an F-test.

2.4. Statistical evaluation of morphological

examination

2.4.1. Procedure of quantification of morphological

feature

Each characteristic of general form, macroscopic

color, tip form, medulla appearance, scale pattern

and cross section form was replaced by the combina-

tion of 4, 4, 4, 5, 4 and 3 categories (C1 , C5). Each

characteristic, categorized as 0/1 data, was replaced

by values obtained by the quantification method of the

third type. The characteristic of each morphological

feature was changed to 2 from 4 values. The six

morphological features obtained by macroscopic and

microscopic observations were transformed into 18

variables obtained by the quantification. As a result,

11 morphological features were changed to 23 vari-

ables including five numerical morphological

features. In five of 23 variables there were cases,

which showed that the standard deviation calculated

from the data from an individual became ‘zero’. The

discrimination analysis used for investigating the

inter-individual variation applies variance as a factor

for discriminating between the data groups. There-

fore, of 23 variables these five variables were

excluded prior to the statistical evaluation. The vari-

ables used were five numerical morphological

features (variables) and 13 variables obtained by the

quantification analysis.

2.4.2. Statistical method

The availability of 18 variables for a comparison of

hair morphology was investigated by a two-way

comparison using a stepwise linear discrimination

analysis [35]. Eighteen variables obtained from 11

morphological features of 25 hairs in each of ten

males were used as a variable for discrimination analy-

sis. The values of Fin and Fout used for the stepwise

discrimination analysis were 4.08 [F401 (0.05)].

The screening method of evidential hair samples

was experimentally investigated using cluster analysis

H. Sato / Legal Medicine 4 (2002) 90–10292

Table 1

Variation in numerical morphological features within the same individuala

Subject Length (cm) Diameter (mm) Medulla index Scale count Hair index

1 11 * (3.5 ^ 1.2) 1 (100.9 ^ 18.5) 1 (17.2 ^ 1.8) 2 (11.8 ^ 0.6) 1 (67.2 ^ 8.4)

2 11 (5.7 ^ 1.7) 1 (89.3 ^ 13.1) 1 (22.0 ^ 3.0) 2 (12.4 ^ 0.5) 2 (86.8 ^ 8.4)

3 11 * (5.2 ^ 1.8) 1 (78.7 ^ 13.3) 1 (19.7 ^ 3.1) 2 (11.3 ^ 0.9) 1 (79.5 ^ 11.8)

M 4 1 * (8.8 ^ 1.7) 1 (92.0 ^ 14.3) 1 (20.4 ^ 2.6) 2 (12.2 ^ 0.6) 1 (85.4 ^ 8.9)

A 5 11 * (6.9 ^ 2.3) 1 (82.7 ^ 16.5) 1 (18.9 ^ 2.8) 2 (12.1 ^ 0.8) 1 (80.2 ^ 12.0)

L 6 11 (5.0 ^ 1.6) 1 (85.7 ^ 15.4) 1 (19.4 ^ 2.0) 2 (13.3 ^ 0.7) 1 (78.6 ^ 11.4)

E 7 11 (9.8 ^ 2.6) 1 (93.0 ^ 10.6) 1 (19.9 ^ 2.1) 2 (12.7 ^ 0.5) 1 (81.0 ^ 9.3)

8 11 * (7.4 ^ 3.0) 11 (80.7 ^ 19.6) 1 (19.0 ^ 2.0) 2 (12.1 ^ 0.7) 1 (76.5 ^ 13.7)

9 11 * (6.9 ^ 2.9) 2 (86.3 ^ 7.9) 1 (23.0 ^ 4.0) 2 (12.8 ^ 0.5) 1 (81.8 ^ 9.3)

10 11 * (7.9 ^ 1.8) 11 (95.0 ^ 21.5) 1 (20.2 ^ 2.4) 2 (13.2 ^ 0.7) 11 (71.3 ^ 14.3)

1 1 (12.7 ^ 2.3) 1 (90.4 ^ 14.7) 1 (20.9 ^ 2.3) 2 (13.4 ^ 0.8) 1 (80.0 ^ 9.9)

2 1 (9.9 ^ 1.8) 1 (100.6 ^ 10.5) 1 (21.5 ^ 2.7) 2 (13.8 ^ 0.5) 1 (78.1 ^ 10.3)

F 3 11 (14.4 ^ 4.4) 11 (76.2 ^ 16.3) 1 (22.4 ^ 3.1) 2 (13.6 ^ 0.6) 2 (89.4 ^ 6.2)

E 4 1 (11.5 ^ 2.3) 1 (88.4 ^ 11.0) 1 (24.0 ^ 4.1) 2 (12.8 ^ 0.9) 1 (83.7 ^ 10.2)

M 5 11 (13.4 ^ 3.7) 1 (86.5 ^ 14.2) 1 (21.8 ^ 3.0) 2 (12.4 ^ 0.7) 1 (74.1 ^ 13.1)

A 6 11 * (12.5 ^ 2.6) 1 (99.7 ^ 14.6) 1 (20.6 ^ 3.2) 2 (13.9 ^ 0.5) 1 (75.4 ^ 9.6)

L 7 11 (35.6 ^ 20.2) 1 (95.6 ^ 11.8) 1 (21.7 ^ 2.6) 2 (12.9 ^ 0.5) 1 (71.5 ^ 10.8)

E 8 1 (8.8 ^ 1.3) 1 (97.3 ^ 15.5) 1 (23.9 ^ 3.5) 2 (12.7 ^ 0.6) 1 (86.6 ^ 9.9)

9 1 * (9.4 ^ 1.0) 2 (86.9 ^ 7.0) 1 (19.5 ^ 2.9) 2 (12.9 ^ 0.5) 1 (83.2 ^ 8.4)

10 1 * (5.3 ^ 1.0) 1 (89.8 ^ 10.3) 11 (20.0 ^ 5.1) 2 (12.3 ^ 0.6) 1 (83.6 ^ 8.4)

a 2 : CV , ¼ 10.00 (CV: coefficient of variation within the same individual); 1: 10.00 ,CV , ¼ 20.00; 11 : 20.00 ,CV; *: significant

difference at P , 0.01 by F-test of regional variation within the individual; and (): mean ^ standard deviation calculated from 25 hairs of each

individual.

Page 4: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

[35,36]. Three hairs were randomly sampled from 25

hairs of each individual and a total of 60 hairs were

produced for use as trial hair samples (experimental

evidential hair samples). The remaining 22 hairs of

each individual were used as known control hair

samples of each individual. In each trial 82 hair

samples, comprising 60 evidential hair samples (Nos

1 , 60) and 22 control hair samples (Nos 61 , 82),

were screened by cluster analysis using the square

Euclidean distance as a dissimilarity and group aver-

age as a clustering method.

3. Results

3.1. Intra-individual variation of morphological

features

Intra-individual variations in macroscopic and

microscopic morphological features are shown in

Tables 1 and 2.

The length showed an extremely large variation on

the same head and three numerical features such as

diameter, medulla index and hair index showed a

small variation for the same head. No intra-individual

variation was detected in the scale count. The range of

length varied from one individual to another. The

diameter, medulla index, scale count and hair index

showed a small variation between different indivi-

duals as well as within the same individual.

The scale pattern showed a small variation, but

other morphological features such as general form,

macroscopic color, tip form, medulla appearance

and cross section form, respectively, showed moder-

ate variations. Intra-individual variations observed in

medulla appearance and cross section form were

larger than those observed in the general form, macro-

scopic color and tip form. The general form and

macroscopic color showed moderate variations

between different individuals. In these two morpholo-

gical features there were some cases, which showed

large inter-individual variations that exceeded intra-

individual variations. The other four morphological

features showed simply small variations between

different individuals. The range of characteristics of

medulla appearance and cross section form tended to

overlap between different individuals.

The length, general form, tip form, medulla appear-

ance and scale pattern showed a regional variation on

ten, eight, three, three and one individuals, respec-

tively. General form and length showed a large differ-

ence among the five anatomical locations on the same

head. However, regional variations were not observed

in five morphological features such as cross section

form, diameter, medulla index, scale count and hair

index.

3.2. Statistical evaluation of morphological

examination

The inter-individual comparison by discrimination

analysis and the screening of experimental evidential

hair samples by cluster analysis were carried out. The

procedure used for expressing the six morphological

features classified numerically is shown in Table 3.

Each characteristic of general form, macroscopic

color, tip form, medulla appearance, scale pattern

and cross section form was replaced by the combina-

tion of four, four, four, five, four and three categories

(C1 , C5), respectively. Each characteristic categor-

ized as 0/1 data was replaced by some values obtained

by the quantification method of the third type. For

example, six characteristics of general form are

expressed as combinations of four categories as

shown in Table 3 and a response pattern of 0/1 data

was obtained using a category as a column and a

characteristic as a row. Since, in this response pattern,

the number of category is smaller than that of the

characteristic, an eigenvalue problem of symmetric

matrix in category is solved using the Jacobi method.

As a result, four eigenvalues can be obtained as

follows; 1, 0.75, 0.75 and 0.25. The eigenvector was

calculated for the maximum of eigenvalues except ‘1’

and values given for each characteristic were calcu-

lated using the eigenvetor. Three values shown in

Table 3 were obtained for each characteristic, respec-

tively.

The process for quantifying morphological data of

six hairs, that one hair per individual was selected

from six individuals, was shown in Table 4. The

first step involved morphological data obtained by

macroscopic and microscopic observations. In the

second step, six morphological features were trans-

formed into 18 variables obtained by the quantifica-

tion shown in Table 3. As a result, 11 morphological

features were changed to 23 variables including five

H. Sato / Legal Medicine 4 (2002) 90–102 93

Page 5: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

H.

Sa

to/

Leg

al

Med

icine

4(2

00

2)

90

–1

02

94

Table 2

Variation in morphological features within the same individuala

Subject Macroscopic features Microscopic features

General form Color Tip form Medulla appearance Scale pattern Cross section form

1 1 (Arch) 1 (Black) 1 * (Oblique) 11 11 11

2 2 (Arch) 2 (Brown-black) - (Transverse) 11 11 * 1 (Circle)

3 11 2 (Black) 1 (Transverse) 11 1 (Flat wave) 1 (Ellipse)

M 4 1 * (Hemi-circle) 2 (Black) 1 (Transverse) 11 2 (Flat wave) 1 (Ellipse)

A 5 2 (Arch) 1 (Brown-black) 1 (Transverse) 11 2 (Flat wave) 11

L 6 1 (Arch) 11 11 11 1 (Flat wave) 11

E 7 11 * 1 (Black) 11 * 11 11 11

8 1 * (Arch) 1 (Brown-black) 1 * (Transverse) 1 * (Absent) 2 (Flat wave) 1 (Ellipse)

9 1 * (Arch) 1 (Brown-black) 1 (Transverse) 11 2 (Flat wave) 11

10 11 11 1 (Oblique) 1 (Continuous) 2 (Flat wave) 11

1 1 * (Slight wave) 11 2 (Transverse) 1 (Dotted) 2 (Flat wave) 11

2 11 11 1 (Oblique) 11 * 1 (Flat wave) 11

F 3 11 1 (Brown) 11 11 * 2 (Flat wave) 1 (Circle)

E 4 11 * 11 11 11 1 (Flat wave) 11

M 5 11 11 1 (Needle) 1 (Fragmental) 11 11

A 6 1 (Arch) 1 (Brown-black) 11 11 2 (Flat wave) 1 (Ellipse)

L 7 1 (Complicated wave) 1 (Change of color)** 2 (Needle) 11 2 (Irregular wave) 11

E 8 1 (Slight wave) 11 2 (Oblique) 11 1 (Flat wave) 11

9 1 * (Slight wave) 11 1 (Oblique) 11 2 (Flat wave) 11

10 1 * (Arch) 1 (Black) 1 (Transverse) 1 (Dotted) 1 (Flat wave) 11

a 2 : no variation (ratio of the main feature to an individual is more than 70%); 1: small variation (ratio of the main feature to an individual is from 50 to 70%); 11 : large

variation (ratio of the main feature to an individual is below 50%); (): main feature in the individual (ratio is more than 50%); *: significant difference at P , 0.01 by F-test of regional

variation within the individual using variables obtained by quantification; and **: change of macroscopic color was observed along a single hair shaft.

Page 6: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

numerical morphological features. In five variables

such as color-V1, color-V3, tip form-V3, scale

pattern-V3 and cross section form-V2, there were

cases, which showed that the standard deviation

calculated from data from an individual became

‘zero’. The discrimination analysis used for investi-

gating the inter-individual variation applies variance

as a factor for discriminating between data groups.

Therefore, of 23 variables these five variables were

excluded prior to the statistical evaluation. The vari-

ables used were five numerical morphological

features (variables) and 13 variables obtained by the

quantification analysis.

The availability of 18 variables for forensic hair

comparison was investigated by two-way comparison

in ten Japanese males using a stepwise linear discri-

mination analysis. The results of the two-way compar-

ison are shown in Table 5. The approximate F values

signify that all results obtained by the two-way

comparison using the stepwise linear discrimination

analysis were significant at P , 0:01 because these F

values were greater than 4.31 [F403 (0.01)].

On five of 45 comparisons, hairs from two indivi-

duals (groups 1 and 2: each individual shown on left

side column and upper row of Table 5) could be

completely distinguished. On three comparisons, 25

hairs of one individual (group 1) were completely

distinguished from hairs from another individual

(group 2), but a few hairs from group 2 individuals

were incorrectly assigned to group 1 individuals.

Conversely, on ten comparisons, 25 hairs of one indi-

vidual (group 2) were completely distinguished from

H. Sato / Legal Medicine 4 (2002) 90–102 95

Table 3

Quantification of morphological features obtained by macroscopic and microscopic observation

Page 7: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

H.

Sa

to/

Leg

al

Med

icine

4(2

00

2)

90

–1

02

96Table 4

Procedure for converting morphological data into numerical dataa

a From six individuals one hair per individual as indicated as an example.

Page 8: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

hairs from another individual (group 1), but a few

hairs from group 1 individuals were incorrectly clas-

sified as group 2 individuals. On the other hand, on 27

comparisons, a few hairs from each individual were

incorrectly grouped with another individual, respec-

tively. The most difficult combination on the two-way

comparison was a comparison between M5 and M9.

In this comparison, of 25 hairs of M5, 18 were

correctly identified as being from M5 but the other

seven hairs of M5 were incorrectly discriminated to

be from M9. In addition, of 25 hairs of M9, 20 were

correctly assigned to M9 but the other five hairs of M9

were incorrectly assigned to M5. The percent of

correct discrimination for M5 and M9 were 72%

and 80%, respectively. The majority of the compari-

sons showed a high discrimination power between the

two individuals.

Fig. 1 is the result of the experimental screening

where individual M1 was used as a known control.

The clusters, which showed a distance of 100 or less,

are shown in the dendrogram of Fig. 1. The order of

samples was amenable to the last result of the cluster

analysis. The sample number enclosed by a square

(A) indicates a control hair sample (M1). The sample

number enclosed by a circle (W) indicates hairs,

randomly collected from individual M1 and which

should morphologically correspond to control hair

samples. The standard used in the clustering of this

investigation were two points, as follows. One stan-

dard was that the distance for clustering was 100 or

less. On the level of this distance about half the

number of hair samples used for the cluster analysis

are merged in the clusters formed. Another standard

was that more than 50% of the hairs included in the

cluster were control hairs. Under this condition the

main morphological feature of hairs grouped into

the cluster reflects the characteristic of control hair

samples. In Fig. 1, five clusters including cluster 2-

H. Sato / Legal Medicine 4 (2002) 90–102 97

Table 5

Discrimination of morphological features in head hairs of ten male Japanesea

a Upper half: results of a two-way comparison using the stepwise linear discrimination analysis. The four values shown in each box of the

upper half were as follows. A value at the upper left represents the number of hairs of group 1 (individual shown on left side column) that were

correctly judged as belonging to group 1. A value at the lower right represents the number of hairs of group 2 (individual shown on upper row)

that were correctly judged as belonging to group 2. On the other hand, the values at the upper right and lower left represent the numbers of hairs

of group 1 or 2 that were incorrectly judged as belonging to of group 2 or 1, respectively. Lower half: approximate F values for the step in which

the stepwise linear discrimination analysis was finished.

Page 9: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

80-61, cluster 3-71-75-76, cluster 9-70-67, cluster 62-

64 and cluster 24-65-74 were consistent with the

above-mentioned standards. These clusters indicated

that hairs grouped into these cluster are morphological

similar. As a result, out of three hairs (Nos 1–3),

which must have morphologically corresponded

with control hair samples of individual M1, two

hairs of Nos 2 and 3 could be included in the cluster

and correctly discriminated as hairs originating from

individual M1. On the one hand, two hairs of Nos 9

and 24, which should not morphologically correspond

to control hair samples, were incorrectly associated as

hairs originating from individual M1. One hair (No.

1), which must have morphologically corresponded

with control hair samples, was contained in one clus-

ter with three hairs of sample Nos 72, 82 and 30. Since

only two hairs (Nos 72 and 82) were included in this

cluster as control hairs, this cluster does not satisfy the

above-mentioned standards and was not used for hair

morphological comparison. As a result, this one hair

(No. 1) was incorrectly excluded as a hair originating

from the individuals, except for individual M1. When

the obtained cluster was evaluated based on only two

of the above-mentioned standards, only one control

hair was included in the cluster consisting of two

hairs. In this case the hair morphological comparison

became very difficult. In addition, in the case

mentioned above the cluster with a distance of 50

and under, was added as a standard for judging cluster

analysis.

Based on the three above-mentioned standards,

experimental evidential hair samples were evaluated

against each control hair samples from 20 individuals

by cluster analysis (Table 6). The numbers of hair

samples, which were judged to morphologically

correspond to control hair samples of each individual,

were 0–4. In four individuals (M4, M8, F3 and F10) it

was judged that all of the hair samples selected by

cluster analysis corresponded morphologically with

control hair samples of each individual. In the other

16 individuals a few hairs, which should not morpho-

logically correspond to control hair samples, were

incorrectly associated as a hair originating from

each individual. In three individuals (M9, F1 and

F4), all three hairs selected by the cluster analysis

were incorrectly associated as a hair originating

from the corresponding individual. By 20 cluster

analyzes, 45 hairs were totally selected from 60

H. Sato / Legal Medicine 4 (2002) 90–10298

Fig. 1. Dendrogram of a cluster analysis using hairs of individual M1 as a known control hair sample. Twenty two head hairs from individual M1

were used as a control sample. The results of a cluster analysis of 60 crime scene hair samples made experimentally from 20 persons are

indicated. The sample number enclosed by a circle indicates hairs, randomly collected from individual M1 and should morphologically

correspond to control hair samples. *: Hairs selected as ones morphologically corresponded with hairs of individual M1.

Page 10: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

evidential hair samples. Of these 45 hairs 19 were

correctly assigned as a hair originating from each

individual used as a control individual in each cluster

analysis. On the other hand, 26 hairs were incorrectly

associated as a hair originating from the correspond-

ing individual. In all the cluster analyzes, 41 hairs

were incorrectly excluded.

4. Discussion

In a Japanese forensic laboratory, a forensic

comparison of evidential hair samples collected

from the crime scene has been carried out principally

based on three examinations, which include a

morphological comparison of the hair, ABO blood

grouping and elemental analysis with energy disper-

sive X-ray microanalysis (EDX) [11]. In most actual

caseworks a morphological examination represents

the main comparison procedure because of variable

lengths, damage and ageing of hair samples. There-

fore, the objectivity of the final judgment in a forensic

hair comparison is always influenced by the profi-

ciency and experience of the hair examiners. A high

objectivity and precision in personal identification has

continued to make demands on the evaluation of hair

comparison results in court. At present, the applica-

tion of DNA analysis to forensic hair comparison is

expected to serve as a new technique for improving

the precision of forensic comparison in the Japanese

forensic science field. However, some hesitation

continues to exist in applying DNA analysis to a

routine forensic hair comparison, even today.

DNA extracted from hair samples is decomposed

and typically short in size, and very little DNA can be

obtained from a single hair. Therefore, a hair sample

is a difficult evidential sample for use in DNA analysis

[15,17–19,21] and DNA analysis has not been

H. Sato / Legal Medicine 4 (2002) 90–102 99

Table 6

Morphological comparison between known control hair samples of an individual and crime scene hair samples by screening using cluster

analysis

Names of individual

whose hair should be

discriminated

Number of hair

samples selected

by the screening

Number of hair samples categorized

Correct inclusiona Incorrect inclusionb Incorrect exclusionc

M1 4 2 2 1

M2 1 0 1 3

M3 2 1 1 2

M4 1 1 0 2

M5 0 0 0 3

M6 4 1 3 2

M7 0 0 0 3

M8 2 2 0 1

M9 3 0 3 3

M10 3 2 1 1

F1 3 0 3 3

F2 4 1 3 2

F3 2 2 0 1

F4 3 0 3 3

F5 1 0 1 3

F6 4 2 2 1

F7 3 2 1 1

F8 2 1 1 2

F9 2 1 1 2

F10 1 1 0 2

a Correctly associated as a hair originating from a target individual.b Incorrectly associated as a hair originating from a target individual.c Incorrectly excluded as a hair originating from the individual except for the target individual.

Page 11: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

employed in routine cases of forensic hair compari-

son. It is thought that mtDNA analysis might be suita-

ble for hair samples in contrast to nuclear DNA

because of the high copy number per cell. Since a

method using PCR amplification and direct sequen-

cing can be introduced to the mtDNA sequence, the

use of mtDNA analysis of hair samples in the future is

expected. However, it is not reasonable to examine

mtDNA sequences of all evidential hairs collected

from a crime scene, since this analysis is expensive

and time consuming. Consequently, it is necessary

that evidential hair samples collected from a crime

scene are narrowed down by the screening of morpho-

logical hair examination data for the effective use of

mtDNA analysis.

For morphologically narrowing down evidential

hair samples, a quantification of the classified

morphological examination data and a screening

method by cluster analysis was investigated. All

morphological features classified by macroscopic

and microscopic observations were objectively

processed in the form of numerical data. Morphologi-

cal features were classified using simple characteris-

tics as much as possible for excluding inter-individual

differences of hair examiners in their judging the

morphological data. Prior to the investigation of a

screening method, the actual validity of various

morphological features was evaluated from the point

of view of intra-individual and inter-individual varia-

tions. It was noted that medulla appearance, cross

section form and length generally showed relatively

large variations within the same individual. Particu-

larly, the range of length varied from one individual to

another. The range of characteristics of medulla

appearance and cross section form tended to overlap

between different individuals. The general form,

macroscopic color, tip form, scale pattern and count,

diameter, medulla index and hair index showed small

or moderate variations within the same individuals.

Except for scale pattern, the range of their character-

istics showed moderate variations between different

individuals. Scale pattern showed a very small varia-

tion between different individuals as well as within the

same individual.

As a result, morphological features of Japanese

head hairs can be appraised as follows.

† Morphological features, which show small intra-

individual variations and additionally show large

inter-individual variations, were not observed in

Japanese head hairs.

† There were cases in which morphological features,

which show moderate intra-individual variations,

showed larger inter-individual variations than

intra-individual ones.

From these results, we conclude that morphological

comparisons between hair samples are possible, based

on the simple characteristics used in the present inves-

tigation.

In order to evaluate hair morphological examina-

tions objectively, the probability involved in human

hair comparison has been reported by Gaudette et al.

and other forensic hair researchers [12–14,37]. In a

later publication Gaudette concluded that the signifi-

cance of the research was not in the actual probability

numbers found but in the experimental proof of the

proposition that the macroscopic and microscopic hair

comparison is a useful technique and that hair consti-

tutes valid evidence [14]. On the other hand, it is

considered by statisticians that the use of Bayesian

theorem is the best approach to the evaluation of

matching probability between evidential hairs and

known control hairs [38,39]. However, hair examiners

have continued to use classical morphological

comparison among hairs because of the inconsistency

between results obtained from practical morphologi-

cal examinations and results statistically inferred by

the Bayesian theorem [40,41].

The morphological examination method considered

in this study is a method in which hair morphological

features are classified by means of a simple classifica-

tion standard, the results are treated statistically and

evidential hairs are screened in comparison with

known control hairs. From this method the matching

probability among hairs is not calculated and this

examination is performed only as a screening step

prior to more objective methods such as DNA analysis

with nuclear DNA and mtDNA. Once a classification

standard of morphological features is quantified by

the quantification method of the third type, it is possi-

ble that the values obtained can be always applied to

the expression of hair morphological features and

used in a morphological comparison among hair

samples as far as the same classification standard is

used. Since hair examiners commonly perform a

H. Sato / Legal Medicine 4 (2002) 90–102100

Page 12: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

morphological examination and only replace morpho-

logical features with numerical data in a rigid manner,

they can objectively execute a nearly similar screen-

ing with an actual judgment of morphological exam-

inations. Simple characteristics were used as a

classification standard in this study. In two indivi-

duals, evidential hairs could not be selected at all by

this screening. Less than two hairs per person were

correctly selected as hair samples that morphologi-

cally corresponded to control hair samples. There

was not an individual in which all of three hairs

included experimentally were selected correctly.

Further, 26 hair samples were associated incorrectly

with 20 individuals, and half of the hair samples

extracted by the screening were associated incor-

rectly. In six of 20 individuals, an incorrect inclusion

was observed. However, in an experimental model of

evidential hair samples derived from 20 individuals,

hair samples from 13 individuals were successfully

discriminated by the cluster analysis used in the

present investigation. This result suggests that the

screening of evidential hair samples is possible by

cluster analysis using variables obtained by the quan-

tification of morphological features.

It is not clear from these correct and incorrect asso-

ciation results in the range of this study that the results

are meaningful for forensic comparison of hair

morphology. As a general tendency, an incorrect asso-

ciation is limited in terms of setting up fine character-

istics for classifying morphological features. However,

a correct association also consequently decreases.

Thus, skill is required in the use of fine characteristics

for classifying hair morphology and morphological

comparisons of hair samples need to be carried out

more precisely. In order to improve the precision of

the method designed, the issue of how fine a character-

istic for classifying morphological features is arises.

Ideally, using simple characteristics, categories are

combined in a complicated manner for the quantifica-

tion of morphological features, and it is important that

values obtained by the quantification are as significant

as possible. Therefore, it is necessary to develop a

classification standard of hair morphology which is

useful in morphological examinations, requiring

many hair examiners to arrive at a similar conclusion

at all times. From the point of view of simplifying a

method and improving screening probability it is

necessary that the characteristics used for morphologi-

cal observation are unified into a system, that can be

used by many examiners.

Hair itself is typically not of any evidential value.

Hair evidence increases its significance as an exhibit

item as the result of a comparison between an eviden-

tial hair and a known hair sample. Hair may have

higher evidential value than other biological materials

in the case where a morphologically narrowed eviden-

tial hair is examined by mtDNA analysis since

mtDNA analysis is a good tool for comparing hairs.

References

[1] 10th Meeting of International Association of Forensic

Science, Oxford, England. Characterization of human hair. J

Forensic Sci Soc 1984;24:247–249.

[2] 11th Meeting of the International Association of Forensic

Science, Vancouver, Canada. Human hair. Cad Soc Forensic

Sci 1987;20:239–250.

[3] Federal Bureau of Investigation, editor. Proceedings of the

International Symposium on Forensic Hair Comparisons

Quantico: FBI Academy, 1985.

[4] Porter J, Fouweather C. An appraisal of human head hair as

forensic evidence. J Soc Cosmet Chem 1975;26:299–313.

[5] Robertson J. An appraisal of the use of microscopic data in the

examination of human head hair. J Forensic Sci Soc

1982;22:390–395.

[6] Bisbing RE. The forensic identification and association of

human hair. In: Saferstein R, editor. Forensic science hand-

book, New Jersey: Prentice-hall Inc, 1982. pp. 184–221.

[7] Shaffer SA. A protocol for the examination of hair evidence.

Microscope 1982;30:151–161.

[8] Clement JL. Recent progress in identifying hair. Int Criminal

Police Rev 1983;38:22–25.

[9] Committee on Forensic Hair Comparison. Preliminary report.

Crime Lab Digest 1985;12:50–57.

[10] Bisbing RE, Human hair in a forensic perspective. Proc Int

Sym Forensic Hair Comparison, FBI Academy, 1985. pp. 35–

44.

[11] Seta S, Sato H, Miyake B. Forensic hair investigation. Foren-

sic Sci Prog 1988;2:47–166.

[12] Gaudette BD, Keeping ES. An attempt at determining prob-

abilities in human scalp hair comparison. J Forensic Sci

1974;19:599–606.

[13] Gaudette BD. Probabilities and human pubic hair compari-

sons. J Forensic Sci 1976;21:514–517.

[14] Gaudette BD. Some further thoughts on probabilities and

human hair comparisons. J Forensic Sci 1978;23:758–763.

[15] Higuchi R, von Beroldingen CH, Sensabaugh GF, Erlich HA.

DNA typing from single hairs. Nature 1988;332:543–546.

[16] Uchihi R, Tamaki K, Kojima T, Yamamoto T, Katsumata Y.

Deoxyribonucleic acid (DNA) typing of human leukocyte

antigen (HLA)-DQA1 from single hairs in Japanese. J Foren-

sic Sci 1992;37:853–859.

H. Sato / Legal Medicine 4 (2002) 90–102 101

Page 13: Statistical evaluation of morphological data of Japanese head hair and the screening of evidential hair samples by cluster analysis

[17] Wilson MR, Polanskey D, Butler J, DiZinno JA, Replogle J,

Budowle B. Extraction PCR amplification and sequencing of

mitochondrial DNA from human hair shafts. BioTechniques

1995;18:662–669.

[18] Matsuda H, Imaizumi K, Kubota S, Miyasaka S, Yoshino M,

Seta S. Technical investigation of DNA extraction from single

hair shaft (in Japanese). Rept Nat Res Inst Police Sci

1997;50:23–28.

[19] Matsuda H, Sekiguchi K, Kasai K, Yoshino M, Seta S.

Evaluation methods of DNA analysis for hair samples. Jpn J

Sci Tech Iden 1997;2:79–84.

[20] Allen M, Engstrom AS, Meyers S, Handt O, Saldeen T, von

Haeseler A, Paabo S, Gyllensten U. Mitochondrial DNA

sequencing of shed hairs and saliva on robbery caps: sensitiv-

ity and matching probabilities. J Forensic Sci 1998;43:453–

464.

[21] Linch CA, Smith SL, Prahlow JA. Evaluation of the human

hair root for DNA typing subsequent to microscopic compar-

ison. J Forensic Sci 1998;43:305–314.

[22] Nozawa H, Yamamoto T, Uchihi R, Yoshimoto T, Tamaki K,

Hayashi S, Ozawa T, Katsumata Y. Purification of nuclear

DNA from single hair shafts for DNA analysis in forensic

sciences. Legal Med 1999;1:61–67.

[23] Yoshino M, Sato H, Seta S. Deoxyribonucleotide acid typing.

In: Siegel JA, Saukko PJ, Knupfer GC, editors. Encyclopedia

of forensic sciences, 3. San Diego: Academic Press, 1999. pp.

1025–1031.

[24] Dizinno JA, Wilson MR, Budowle B. Typing of DNA derived

from hairs. In: Robertson J, editor. Forensic examination of

hair, London: Taylor & Francis, 1999. pp. 155–173.

[25] Baker LE, McCormick WF, Matteson KJ. A silica-based mito-

chondrial DNA extraction method applied to forensic hair

shafts and teeth. J Forensic Sci 2001;46:126–130.

[26] Hopgood R, Sullivan KM, Gill P. Strategies for automated

sequencing of human mitochondrial DNA directly from

PCR products. BioTechniques 1992;13:82–92.

[27] Wilson MR, Holland MM, Stoneking M, DiZinno JA,

Budowle B. Guideline for the use of mitochondrial DNA

sequencing in forensic science. Crime Lab Digest

1993;20:68–77.

[28] Sato H, Yoshino M, Miyake B, Seta S. An investigation on the

macroscopical morphology in identification of human head

hairs (in Japanese). Rept Nat Res Inst Police Sci

1981;34:101–103.

[29] Sato H, Yoshino M, Miyasaka S, Miyake B, Seta S. An inves-

tigation on the numerical morphology in identification of

human head hair (in Japanese). Rept Nat Res Inst Police Sci

1983;36:131–136.

[30] Sato H, Tuzuki Y, Miyasaka S, Yoshino M, Miyake B, Seta S.

The evaluation of the validity of medulla appearance in foren-

sic comparison of Japanese head hair (in Japanese). Rept Nat

Res Inst Police Sci 1988;41:1–12.

[31] Sato H, Kimijima T, Miyasaka S, Yoshino M, Miyake B, Seta

S. Scale pattern and scale count in forensic comparison of

Japanese head hair (in Japanese). Rept Nat Res Inst Police

Sci 1990;43:1–16.

[32] Sato H, Kimijima T, Miyasaka S, Yoshino M, Miyake B, Seta

S. Evaluation of cross section morphology in forensic compar-

ison of Japanese head hair. 1. The cross section form and

pigment morphology (in Japanese). Rept Nat Res Inst Police

Sci 1990;43:73–88.

[33] Sato H, Kimijima T, Miyasaka S, Yoshino M, Miyake B, Seta

S. Evaluation of cross section morphology in forensic compar-

ison of Japanese head hair. 2. The numerical morphology (in

Japanese). Rept Nat Res Inst Police Sci 1990;43:89–100.

[34] Sato H, Seta S, An appraisal of the use of macroscopic and

microscopic data in Japanese head hair comparison. Proc Int

Sym Forensic Hair Comparison, FBI Academy, 1985. pp. 143.

[35] Tanaka Y, Tarumi T, Wakimoto K, editors. Pasokon-Toukei-

kaiseki-handobukku. II Tahenryoukaiseki-hen (in Japanese)

Tokyo: Kyouritushuppan, 1984. pp. 112–137, 226–257, 296–

313.

[36] Nakamura S. Reikai Tahenryou-kaiseki-nyuumon (in Japa-

nese), Tokyo: Nikkan-Kougyousinbunsha, 1980. pp. 153–171.

[37] Wickenheiser RA, Hepworth DG. Further evaluation of prob-

abilities in human scalp hair comparison. J Forensic Sci

1990;35:1323–1329.

[38] Aitken CGG, Robertson J. A contribution to the discussion of

probabilities and human hair comparisons. J Forensic Sci

1987;32:684–689.

[39] Hoffmann K. Statistical evaluation of the evidential value of

human hairs possibly coming from multiple sources. J Foren-

sic Sci 1991;36:1053–1058.

[40] Gaudette BD. Evidential value of hair examination. In:

Robertson J, editor. Forensic examination of hair, London:

Taylor & Francis, 1999. pp. 243–260.

[41] Gaudette BD. Comparison: significance of hair evidence. In:

Siegel JA, Saukko PJ, Knupfer GC, editors. Encyclopedia of

forensic sciences, 3. San Diego: Academic Press, 1999. pp.

1018–1024.

H. Sato / Legal Medicine 4 (2002) 90–102102