Beruflich Dokumente
Kultur Dokumente
B Found
Forensic Expertise Profiling Laboratory, School of Human Biosciences, La Trobe University,Victoria, 3086, Australia and Document Examination Team, Victoria
Police, Victoria Forensic Science Centre, Forensic Drive, Macleod. Victoria, 3085, Australia
DK Rogers'
Forensic Expertise Profiling Laboratory, Science of Human Bioasciences,La Trobe University,Victoria, 3086, Australia
Science & Justice 2005 45 199 - 206 Received 13 June 2003 Accepted 20 May 2005
Many forensic document examiners are hesitant to express author- Schriftzuges treffen sollen, da in diesem - bedingt durch
ship opinions on photocopied handwriting as the photocopying den Fotokopier-Prozess - verglichen mit der Original-
process results in less feature information than original writing. Schriftprobe weniger Merkmale erkennbar sind. Diese
This study aimed to test the accuracy of 15 examiners' opin- Studie wurde durchgefiihrt, um die Genauigkeit von 15
ions regarding whether photocopied questioned signatures were Sachversttindigengutachten hinsichtlich ihrer Aussage zu
genuine or simulated. Each examiner received the same set of iiberpriifen, ob fotokopierte fragliche Unterschriften echt oder
original signature exemplars, from one individual, and a set of gefalscht waren. Jeder der Gutachter erhielt den gleichen Satz
eighty questioned photocopied signatures comprising of genuine von Unterschriftsproben einer Person und einen Satz von 80
and simulated signatures. The overall misleading (error) rate for zu priifenden fotokopierten Signaturen, der aus echten und
the grouped examiners' opinions was 0.9% providing strong ev- nachgeahmten Proben bestand. Die Gesamtfehlerrate fur die
idence that examiners can make accurate observations regarding gruppierten Gutachten von 0,9% fuhrt zu dem Schluss, dass
the authorship of non-original handwriting. die Sachverstandigen mit hoher Wahrscheinlichkeit zutreffende
Beobachtungen bzgl. der Urheberschaft von nicht im Original
Muchos investigadores forenses de documentos se muestran vac- vorliegenden Handschriften machen.
ilantes a la hora de expresar opiniones sobre escrituras fotocopi-
adas ya que el proceso de fotocopiado resulta en menor infor- De nombreux spCcialistesforensiques en documents hCsitent i ex-
maci6n de 10s rasgos que el original. Este estudi6 se dirigi6 a primer une opinion sur 1'identitC de I'auteur sur des Ccritures pho-
comprobar la exactitud de las opiniones de 15 examinadores so- tocopiCes, vu que le processus de copie rCsulte en une diminution
bre si unas firmas dubitadas eran genuinas o simuladas. Cada de l'information. Cette Ctude vise il tester la justesse de l'opinion
examinador recibi6 un conjunto de ejemplares de firmas origi- de 15 spCcialistes i determiner si des signatures photocopiCes
nales de un individuo y un conjunto de ochenta firmas dubitadas, contestees Ctaient authentiques ou simulCes. Chaque spkcialiste
fotocopiadas, compuesto por firmas genuinas y simuladas. EL a r e p le m&mejeu d'exemplaires de signatures originales d'un
resultado medio final para el grupo de examinadores fue de un individu et un jeu de 80 signatures contestkes photocopiCes com-
error del0.9% lo que prueba que se pueden hacer observaciones prenant des signatures authentiques et simulCes. Le taux d'erreur
exactas en cuanto a la auton'a de escritura no original. total pour les opinions groupCes Ctait de 0.9 % montrant une forte
capacitk des spkcialistes i faire des observations correctes sur
Viele Experten fur forensische Schriftuntersuchungen zogem, l'identitk de I'auteur sur des Ccrits non originaux.
wenn sie Aussagen zur Urheberschaft eines fotokopierten
Figure 1 Three of the exemplar signatures. Top signature Figure 3 Signatures a 4 are questioned genuine
is an example of a requested signature and the signatures for which misleading opinions were
two bottom signatures were collected. All given. The labels correspond to signatures a-d
signatures in this and following figures are of Table 6. The bottom right signature is an
reproduced at the same magnification. exemplar signature; provided for ease of
com~arison.
All questioned signatures were compiled into a single booklet, Statistical tests involving the Mann-Whitney U test and the Spear-
one signature per page. man correlation coefficient were performed using the Statistical
Package for the Social Sciences (SPSS, version 10) to determine
Procedurefor obtaining FDEs' opinions whether there was a relationship between misleading opinions
The booklet containing the questioned signatures, along with and inconclusive opinions provided by examiners. These anal-
the original signature exemplars were circulated, in turn, around yses were performed for opinions on the full set of questioned
the participating FDEs. In addition, each participant received an signatures, on genuine signatures considered alone and simu-
instruction sheet and answer booklet. lated signatures alone. The Mann-Whitney U Test was used to
test whether there was a difference in the number of inconclusive
FDEs were told that all questioned signatures were non-original, opinions given by examiners who gave misleading opinions com-
that it is to be assumed that the date range of the exemplar material pared to examiners with no misleading opinions. An alpha level
is appropriate to the questioned material and that each questioned of 0.05 was used, thus a significant difference between the two
signature was simply a generation of photocopy taken from the groups of examiners was accepted if the two-tailed probability
original signature. @ value), determined from the U value, was less than 0.05. In
addition, Spearman's correlation coefficient ( p ) was calculated
FDEs were asked to express an opinion, for each of the 80 ques- to evaluate the correlation between FDEs' inconclusive rate and
tioned signatures, as to whether they were genuine signatures their number of misleading opinions. The correlation was ac-
(written normally by the exemplar writer), simulations or whether cepted as significant if the associated two-tailed p value was less
they were unable to say (inconclusive). The participants reported than 0.05.
their opinions using the multiple choice answer booklets. They
were instructed that opinions should not proceed through quality
Results
assurance processes.
Fifteen FDEs provided an opinion for each of the 80 questioned
signatures (80 opinion units per subject) which resulted in 1200
Analysis of FDEs' opinions
opinion units for the group. Of these, there were 240 opinion units
FDEs' opinion responses (opinion units) were marked as correct,
for signatures which were in fact genuine and 960 opinion units
misleading or inconclusive. These marks were then analysed to
for signatures which were in fact simulations.
produce scores for the questioned signatures. The scores are pre-
sented as numbers of opinions or as percentages. The following
are definitions of the score categories used. Grouped results
The opinion scores for the full group of 15 subjects on the ques-
# Correct tioned simulated and genuine signatures are shown in Table 1. As
The number of opinions that were correct. can be observed, examiners provided an opinion that the ques-
#Misleading tioned signatures were either genuine or simulated on 78% of
The number of opinions that were misleading. occasions (22% inconclusive opinions) with an overall mislead-
# Inconclusive ing rate for the group being 1.2%. The data indicates higher
The number of opinions that were inconclusive. conservatism (inconclusive opinions) regarding the expression of
% Correct opinions associated with signatures that were in fact genuine than
The number of correct opinions divided by the total number of for simulated signatures (inconclusive rates of 35.4 and 18.6% re-
opinions (expressed as a percentage). spectively). There was also found to be a marked difference in
% Misleading the percentage misleading of called rate between the genuine and
The number of misleading opinions divided by the total number simulated signatures (5.2 and 0.4% respectively), reflecting pre-
of opinions (expressed as a percentage). vious findings [7-91 for comparisons using original signatures.
% Inconclusive
The number of inconclusive opinions divided by the total number Table 2 provides the group correct, misleading and inconclusive
of opinions (expressed as a percentage). scores for each of the three types of questioned simulations (one-
% Correct called off, practised and auto-simulations). As can be observed, only
The number of correct opinions divided by the sum of the correct three of the 960 opinions expressed by the group relating to ques-
and misleading opinions (expressed as a percentage). tioned simulated signatures were misleading. This corresponds to
% Misleading called a misleading called rate of 0.3% for one-off simulations and 0.5%
The number of misleading opinions divided by the sum of the for practised simulations. No errors were made regarding opin-
correct and misleading opinions (expressed as a percentage). ions on auto-simulated signatures. The inconclusive rates across
all of the questioned signatures that were simulated was found to
The 'called' scores do not include inconclusive opinions. They be similar (between 17 and 20%). The slightly higher misleading
equate to a number that reflects the opinion rate when an FDE was rate associated with practised signatures is not a surprising find-
prepared to express an opinion that was other than inconclusive. ing given that simulators were able to refine their forging attempts
Such positive opinions are likely to have greater impact than and that no cognitive stress was added to the task by having to sign
inconclusive opinions in a casework setting and therefore the a 'one-off' attempt. If anything it is surprising that the misleading
called rates are arguably the rates with most significance for legal and inconclusive rates for the one-off and practised groups are so
determinations. similar in magnitude.
Table 1 Opinion scores of FDE group for the questioned simulated and genuine signatures.
Simulated Genuine
Score type signatures signatures Total
# Correct 778 147
# Misleading 3 8
# lnconclusive 179 85
% Correct 81 .O 61.3
% Misleading 0.3 3.3
% lnconclusive 18.6 35.4
% Correct called 99.6 94.8
% Misleading called 0.4 5.2
Summary of individual subjects' opinions scores given must be read taking into account the small number
A summary of individuals' correct, misleading and inconclusive of opinion units used in the percentage calculation. The mean
opinions is given in Table 3. Eight subjects expressed no erro- number of inconclusive opinions for examiners who made errors
neous opinions, four subjects expressed one misleading opinion, was 3.7 and for examiners without error it was 7.0. The difference
two subjects expressed two misleading opinions and one subject failed to reach significance (U = 11.5, p = 0.066) probably due
expressed three misleading opinions. The percentage mislead- to the reduced number of opinions. A trend towards an inverse cor-
ing of called score for examiners ranged from 0.0 to 3.9%. The relation between the number of errors examiners made and their
percentage inconclusive rate varied widely from 1.3 to 75.0%. inconclusive rate was not significant (p = 0.414, p = 0.125).
For those examiners who made no errors the % inconclusive rate
varied from 11.3 to 75%. Examiners who made errors, had on av- Table 5 provides further information on the questioned simu-
erage, fewer inconclusive opinions (mean = 9.6) than examiners lated signature group only. There were 64 simulated signatures
without error (mean = 24.6). A Mann-Whitney U Test indicated (29 one-off, 30 practised and five auto-simulations) in the ques-
that the difference was significant (U = 10.0, p = 0.037). This tioned signature group. The difference between individuals in de-
effect was also reflected by a small but significant inverse cor- termining whether a signature was produced using a simulation
relation between the number of errors examiners made and their process, can be observed from the variation in the inconclusive
inconclusive rate (p = 0.60, p = 0.018). rates (1.6 - 81.3% which represents between 0 and 52 incon-
clusive opinions). This inter-individual variation is particularly
In all, seven of the 15 examiners made at least one error. Six interesting given that only two individuals of the 15 formed mis-
individuals, of the 15 participating, were responsible for the eight leading opinions. For those individuals not expressing misleading
errors regarding questioned signatures that were genuine and two opinions, the inconclusive rate also varied from 1.6 - 81.3%.
individuals were responsible for the three errors regarding the Again it should be noted that these percentages must be read tak-
questioned signatures that were simulated. ing into consideration that only small amounts of data are being
used for individual percentage calculations. The small number of
Table 4 provides further information on the questioned genuine misleading opinions precludes the demonstration of a significant
signatures only. Further evidence of the difference between in- relationship between error and inconclusive rate.
dividuals can be observed from the variation in the inconclusive
rates (0.0-8 1.3% which represents between 0 and 13 inconclusive Observations regarding the signatures where
opinions) and the misleading of called rate (M6.7%). Note that misleading opinions were expressed
because we are now reducing the number of individual opinions, The errors made by the examiners were made on only six of the 80
by dividing up the questioned opinion unit group, the percentage questioned signatures. The eight misleading opinions on genuine
Table 2 Opinion scores of FDE group for the questioned simulation types.
Table 3 Scores for individual FDEs' opinions regarding the group of questioned signatures.
Subject # Correct # Misleading # Inc. % Correct % Misleading % Inc. % Correct called % Misleading called
1 71 0 9 88.8 0.0 11.3 100.0 0.0
2 50 0 30 62.5 0.0 37.5 100.0 0.0
3 74 3 3 92.5 3.8 3.8 96.1 3.9
4 20 0 60 25.0 0.0 75.0 100.0 0.0
5 58 1 21 72.5 1.3 26.3 98.3 1.7
6 69 0 11 86.3 0.0 13.8 100.0 0.0
7 77 2 1 96.3 2.5 1.3 97.5 2.5
8 71 1 8 88.8 1.3 10.0 98.6 1.4
9 75 1 4 93.8 1.3 5.0 98.7 1.3
10 70 0 10 87.5 0.0 12.5 100.0 0.0
11 47 0 33 58.8 0.0 41.3 100.0 0.0
12 56 0 24 70.0 0.0 30.0 100.0 0.0
13 61 2 17 76.3 2.5 21.3 96.8 3.2
14 60 0 20 75.0 0.0 25.0 100.0 0.0
15 66 1 13 82.5 1.3 16.3 98.5 1.5
signatures related to the four questioned signatures shown in Fig- sumably, correctable error in perception. Signature 'd' exhibits
ure 3 and the three misleading opinions on simulated signatures a mixed response. This signature must have contained feature
related to the two questioned signatures shown in Figure 4. Table characteristics that supported both the simulation and genuine
6 provides the raw opinion data for all response types for each propositions.
of the questioned genuine signatures where misleading opinions
were expressed. Table 7 provides the raw opinion data for all It is clear from Table 7 that for signature 'e' most examiners
response types for each of the questioned simulated signatures were either inconclusive or expressed the correct opinion that
where misleading opinions were expressed. The data of the two the signature was simulated. Clearly, in the case where one mis-
tables provides additional information regarding the likely na- leading opinion is recorded, the response is either a transcription
ture of the errors being made. For example, it is clear that for error on the part of the examiner or a, presumably, correctable er-
signature 'b' all examiners, other than the one registering a mis- ror in perception. Signature 'f' presents a more mixed picture and
leading response, expressed inconclusive opinions. It could be must display features that support both the genuine and simulated
argued therefore that signature 'b' must have displayed feature propositions.
characteristics such that it could neither be confidently grouped
as either genuine or simulated. Likewise for signature 'c' the Discussion
vast majority of examiners expressed the correct opinion and This study provides substantial support for the proposition
only one examiner expressed a misleading opinion. Clearly this that FDEs, taken over the group, possess skills at determining
is either a transcription error on the part of the examiner or a, pre- whether photocopied signatures were genuine signatures (written
Subject # Correct # Misleading # Inc. O h Correct % Misleading % Inc. % Correct called % Misleading called
1 8 0 8 50 0 50 100 0
2 12 0 4 75 0 25 100 0
3 13 1 2 81.3 6.3 12.5 92.9 7.1
4 8 0 8 50 0 50 100 0
5 13 1 2 81.3 6.3 12.5 92.9 7.1
6 7 0 9 43.8 0 56.3 100 0
7 14 2 0 87.5 12.5 0 87.5 12.5
8 14 1 1 87.5 6.3 6.3 93.3 6.7
9 11 1 4 68.8 6.3 25 91.7 8.3
10 9 0 7 56.3 0 43.8 100 0
11 8 0 8 50 0 50 100 0
12 10 0 6 62.5 0 37.5 100 0
13 1 2 13 6.3 12.5 81.3 33.3 66.7
14 5 0 11 31.3 0 68.8 100 0
15 14 0 2 87.5 0 12.5 100 0
Subject # Correct # Misleading # Inc. % Correct % Misleading % Inc. % Correct called % Misleading called
normally by the writer of the exemplars) or were the product of this variation translates to an opinion profile for the examiner
a simulation process. The misleading scores recorded are low in group where, over all questioned signatures, the % correct called
spite of the images under examination bearing filtered informa- approaches 100 and the % misleading called approaches zero.
tion in comparison to their source images. This study provides However, there was a significant tendency for a greater number
further empirical data to that already presented by Dawson and of examiner errors to be associated with a lower inconclusive rate
Lindblom [l] and Found et al. [6], that examiners can translate and examiners who made no errors had, on average, higher in-
observations regarding non-original line quality features and ad- conclusive rates than examiners who gave misleading opinions. It
dress whether they are consistent with a genuine writing act or an appears logical, and appropriate, that individual examiners were
act of simulation. using the inconclusive opinion option to avoid expressing mis-
leading opinions.
The overall average inconclusive rate for the group (22%) may
appear to be high. It is not possible to tell whether this was It was not the purpose of this study to pursue the issue of ex-
due to the nature of the exemplar and questioned signatures aminer expertise, that is, whether the skill of FDEs is signifi-
used or was related to the fact that the questioned signatures cantly greater than that of lay people. Previous studies on sig-
were photocopied. It is possible that the photocopying process nature comparison with original signatures [7-91 have demon-
reduces the confidence of some examiners, perhaps due to reduc- strated such expertise. In the three studies, the error rates were
tion in features evident in the original exemplars. However, in markedly and statistically lower for FDEs than for lay subjects.
a previous study [6], considered in the Introduction, there was It was observed that the higher error rates for the lay sub-
no difference between opinions on original signatures and their jects were due to them providing significantly fewer inconclu-
photocopies. sive opinions than the FDEs. This suggests that an element of
examiners' expertise relates to their conservatism, reflecting an
A striking feature of the current study was the marked inter- acceptance of the examination technique's limitations. As with
examiner variation in the inconclusive rate. It is interesting that many techniques, forensic signature comparison has limitations