J Forensic Sci, July 2016, Vol. 61, No.

4
doi: 10.1111/1556-4029.13091
PAPER Available online at: onlinelibrary.wiley.com

CRIMINALISTICS

Michael C. Taylor,1 Ph.D.; Terry L. Laber,2 B.S.; Paul E. Kish,3 M.Sc.; Glynn Owens,4 Ph.D.; and
Nikola K. P. Osborne,5 Ph.D.

The Reliability of Pattern Classification in

Bloodstain Pattern Analysis, Part 1: Bloodstain
Patterns on Rigid Non-absorbent Surfaces*

ABSTRACT: This study was designed to produce the first baseline measure of reliability in bloodstain pattern classification. A panel of
experienced bloodstain pattern analysts examined over 400 spatter patterns on three rigid non-absorbent surfaces. The patterns varied in spatter
type and extent. A case summary accompanied each pattern that either contained neutral information, information to suggest the correct pattern
(i.e., was positively biasing), or information to suggest an incorrect pattern (i.e., was negatively biasing). Across the variables under examina-
tion, 13% of classifications were erroneous. Generally speaking, where the pattern was more difficult to recognize (e.g., limited staining extent
or a patterned substrate), analysts became more conservative in their judgment, opting to be inconclusive. Incorrect classifications increased as
a function of the negatively biasing contextual information. The implications of the findings for practice are discussed.

KEYWORDS: forensic science, bloodstain pattern analysis, error rate, reliability, contextual bias, cognitive science

Bloodstains are a common by-product of violent crime, and heart of bloodstain pattern analysis (BPA) is the recognition and
analysis of these stains is a vital part of a crime scene investiga- classification of the bloodstain pattern.
tion. Despite the fact that DNA analysis can now routinely iden- BPA has been used in criminal investigations since the 1800s.
tify the individuals that have bled at a scene, other important Like many other disciplines from the early days of forensic
questions can remain unanswered. For example, it is not uncom- science, its use and acceptance occurred without rigorous valida-
mon for a suspect to claim that the blood found on his/her cloth- tion. The Organization of Scientific Area Committees (OSAC)
ing was deposited when he/she was trying to aid the victim. In BPA Subcommittee (previously the Scientific Working Group
these situations, understanding the mechanism by which the on Bloodstain Pattern Analysis (SWGSTAIN)) has made valu-
stains were deposited onto an article of clothing could be more able progress in setting standards for training and education, ter-
informative than knowing from whom the blood originated. This minology, quality assurance, and validation of new procedures
is where the analysis of bloodstain patterns can often give valu- for the discipline (1–3). Like other forensic practitioners, how-
able clues as to how the blood came to be where it was found. ever, bloodstain pattern analysts are grappling with the problem
Although the dynamics of the formation of a bloodstain pat- of assessing the reliability of the methodology that they use. At
tern appear to be infinitely variable, it is nevertheless true that this time, very little is known about this beyond the instincts of
bloodstain patterns have reproducible characteristics that allow a experienced instructors and investigators who have observed the
connection to be made between the distribution of bloodstains reproducibility of bloodstain patterns over many crime scenes
and the underlying mechanism of their formation. Thus, at the and practical sessions in the classroom. While such experience
has served as the main basis for assessing the reliability of BPA
testimony in the past, the courts now rightly demand much
1
Institute of Environmental Science and Research Limited (ESR), PO
more. Indeed, establishing accuracy and reliability measures in
Box 29-181, Christchurch, New Zealand. forensic disciplines was a key recommendation in the National
2
Minnesota Bureau of Criminal Apprehension (BCA), 1430 Maryland Ave Research Council’s 2009 (4) report on the state of forensic
East, St Paul, MN 55106. science.
3
Forensic Consultant & Associates, PO Box 814, Corning, NY 14830. There are several factors that may influence the reliability of
4
School of Psychology, University of Auckland, Private Bag 92019,
Auckland 1142, New Zealand. bloodstain pattern classifications. First, different bloodletting
5
Institute of Environmental Science and Research Limited (ESR), PO mechanisms can give rise to bloodstain patterns that possess
Box 29-181, Christchurch, New Zealand. similar or indistinguishable characteristics. For example blunt
*Supported by a research grant from the National Institute of Justice force impact and expiration patterns (2) can both feature small
(Award # 2010-DN-BX-K213). Presented at the International Association of
Bloodstain Pattern Analysts (IABPA) Training Conference, October 1-4,
bloodstains and as such may be confused with one another. Sec-
2013, in San Diego, CA. ond, at times, a pattern might only comprise one, or a small
Received 8 June 2015; and in revised form 6 Oct. 2015; accepted 17 Oct. number of stains, meaning an analyst must decide if he/she has
2015. sufficient data to make a reliable classification. Finally, the sur-

922 © 2016 American Academy of Forensic Sciences

 15564029, 2016, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/1556-4029.13091 by University Of Northumbria At Newcastle, Wiley Online Library on [21/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TAYLOR ET AL. . CLASSIFICATION RELIABILITY IN BPA 923

face characteristics of the substrate on which the bloodstain is acknowledged that training and experience are important ingredi-
created, whether a rigid smooth surface or an absorbent fabric, ents in the accurate conclusions reached by forensic analysts, the
might add another level of complexity to the pattern recognition approach used here was designed to help define the upper limit
task. These issues have been highlighted in several high profile of pattern classification reliability by focusing attention on
homicide trials in which bloodstain pattern analysts have found method reliability rather than analyst competency.
themselves at the center of controversial arguments relating to The type of pattern, pattern extent, the nature of the substrate,
the explanation of the mechanisms that produce very small and the direction of contextual cues were varied in a balanced
bloodstains on clothing (5,6). experiment designed to determine the effect of these variables
Although the size and distribution of individual bloodstains on pattern classification accuracy.
are often measured, pattern recognition methods rely primarily
on a qualitative assessment of the appearance of the pattern. It is
essential therefore that these methods are well understood, their Method
reliability is demonstrable and that any bloodstain pattern con-
Participants
clusion proffered can be supported by statements that guide the
courts in their assessment of the extent of that reliability. Participants were 27 bloodstain pattern analysts from North
One of the complications for any study of method reliability America, Australasia (New Zealand and Australia), and Europe.
is the variability in the methods used. At this time, there is no All were invited based on their experience and standing within
discipline standard in the methodology employed by bloodstain the BPA community and were required to meet the following
pattern analysts. Two suggestions have been made to meet this three criteria: (i) must have completed at least 80 h training in
need (7,8), but so far no significant effort has been made to BPA, (ii) have been active in BPA casework for a minimum of
establish these or any other approaches as standards. In fact, it is 5 years, (iii) and be qualified by a court as an expert in BPA
possible that some bloodstain pattern analysts would struggle to and have provided expert testimony.
articulate the methodology that they employ.
Testing the reliability of BPA methods is not a straightforward
Materials
task. For example, it is generally impossible to know with cer-
tainty the “true” mechanistic cause of a bloodstain pattern at a Bloodstain pattern targets (40 cm 9 40 cm) were prepared in
crime scene. For this reason, procedures to test method reliability a controlled laboratory setting at the Minnesota Bureau of Crimi-
are limited to artificially created scenarios for which the ground nal Apprehension (BCA). This size was chosen to aid with ship-
truth can be known. This approach has the risk of under or over ping the materials to study participants. The completed targets
estimating identification errors because some of the dynamics were coated with a clear lacquer to prevent deterioration and to
present in a real case investigation are lacking. assist with biohazard safety. This coating meant that no chemical
When considering the reliability of any forensic method, it is tests to identify blood or saliva were possible. Participants were
important to assess whether factors outside of the evidence can expected to assume that any visible red-brown stains were
influence the reliability of interpretations. In particular, a grow- indeed bloodstains.
ing body of research has demonstrated that expert interpretations Three variables were manipulated when preparing the blood-
may be influenced by the presentation of contextual information stain pattern targets: type of pattern, pattern extent, and target
(9–18). substrate.
Perhaps more so than in any other forensic discipline, contex-
tual information is a necessary part of BPA. That is because Type of Pattern—One of four different bloodstain pattern types
bloodstain patterns are analyzed in the context of a case with the were prepared for each target: blunt force impact spatter, firearms-
objective to assist with the reconstruction of events. This means related (back or forward) spatter, cast-off pattern, and expirated
that once the pattern is classified, its relevance to the case inves- spatter. These spatter pattern types were chosen to reflect the poten-
tigation must be considered. These two processes (pattern classi- tial for overlap in pattern characteristics between pattern types,
fication and scene reconstruction) frequently overlap. At present, which can be problematic for bloodstain pattern analysts, and were
there is no rigorous protocol for BPA that distinguishes these designed to represent those typically encountered at crime scenes.
processes. This means that, at an early stage of the analysis, ana- Cast-off pattern and blunt force impact spatter targets were
lysts may consider additional case-specific information, such as made with fresh human blood, donated by project volunteers.
medical findings, case circumstances, and even witness testi- Blood was drawn into tubes containing EDTA anticoagulant and
mony. If the BPA methodology is not reliable, this integration was used within 7 days of drawing. Blood for the firearms-
of case information may create a fertile breeding ground for con- related spatter was purchased from Memorial Blood Center, 737
textual bias (19). Pelham Boulevard, Saint Paul, MN 55114, and was used within
30 days of drawing. Blood for the expirated patterns was drawn
from an experimenter on the same day it was used and was used
The Present Study
unrefrigerated. To represent the variability found in crime scenes
The research hypothesis at the heart of this study was as fol- and to create varied pattern extents, two methods were used to
lows: Pattern recognition methods employed in BPA are reliable create each pattern type (Tables 1–4). Targets were mounted on
when used by fully competent analysts. a vertical surface during preparation.
To test this, a panel of experienced bloodstain pattern analysts
classified a series of bloodstain spatter patterns. These patterns Pattern Extent—The extent of pattern on each target was cate-
included stains made under a variety of conditions relevant to a gorized into three levels: minimum, medium, and maximum. Cat-
crime scene and included some sets of stains produced under egory membership was determined by an approximation of the
“ideal” conditions. That is to say, patterns produced to provide total number of stains in the pattern and the number of stains
the maximum chance of accurate classification. While it is larger than 1 mm in diameter (see Table 5).
 15564029, 2016, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/1556-4029.13091 by University Of Northumbria At Newcastle, Wiley Online Library on [21/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
924 JOURNAL OF FORENSIC SCIENCES

TABLE 1––Methods used to create blunt force impact spatter. For both pat- TABLE 6––Example of contextual information.
tern-creating methods, multiple targets were positioned 50 cm from the front,
side, and back of a striking zone. Example scenario containing contextual information* to suggest the
pattern is the result of expirated blood
Method 1 One drop of blood was placed on a wooden block in the center Police were called to a late-night disturbance outside an inner-city club. On
of the striking zone. A hammer was propelled by rubber bands arrival, they found the body of a 23-year-old man in a dark alleyway, with
and gravity onto the blood pool. This method tended to produce a crowd of youths standing nearby watching the paramedics, who comment
fine horizontally directed spatter that they thought the man died from severe internal injuries following a
Method 2 Six drops of blood were placed on a wooden block to the right beating. The man’s external bloodletting injuries were confined to his nose
side of the center of the striking zone and extending outside of and mouth. An officer noticed bloodstains on the side door of the club. The
that zone. The hammer was allowed to fall under gravity alone club owner was interviewed and told police that he had not been present
onto the blood pool. This method tended to produce larger during the disturbance but had heard the noise and had come outside and
spatter stains that travelled higher on the adjacent vertical saw the deceased lying in the alleyway and several youths running off. He
surface says the bloodstaining occurred when the victim was alive and was
coughing up blood. Police bring you a section of the door for your
bloodstain pattern examination. DNA tests confirm the blood on the door of
TABLE 2––Methods used to create cast-off bloodstain patterns. the club was from the 23-year-old victim found in the alleyway. You are
requested to determine the nature of the bloodstaining to confirm the club
Method 1 A wrench was liberally coated in blood and then swung a few owner’s account.
times to remove excess blood. An experimenter stood
approximately 130 cm from the left and front walls that held the *If the target pattern was an expirated bloodstain pattern, this contextual
targets and swung the wrench forcibly from left to right, on an information would be considered positively biasing. If the target pattern was
angle, and overhead and downwards. This method tended to a pattern other than expirated blood, then this contextual information would
produce large stains with a broadly linear distribution be considered negatively biasing.
Method 2 A small knife was dipped a few millimeters into a beaker of
blood and tapped one to two times to remove excess blood. Example scenario containing neutral contextual information
An experimenter then stood approximately 85 cm from the left Police get a call from a hotel manager who reports that one of her cleaning
wall and 130 cm from the front wall and swung the knife staff found bloodstains on the wall in one of the hotel rooms during a
forcibly toward the targets. This method tended to give smaller routine room service. She reports that the guest who occupied the room had
spatter stains in a tightly linear distribution checked out. The police locate the guest who denies all knowledge of the
blood and appears to be able to account for all his movements during his
stay. A DNA test shows the blood in the room is not from him. Police have
brought you this sample from the room and asked you to examine it to help
TABLE 3––Methods used to create expirated bloodstain patterns. determine the significance of the pattern present.

Method 1 An experimenter transferred blood to his lips with a finger and

blew air gently through tightly pursed lips directly toward the substrate. The wallpaper was white Brewster Easy Texture pain-
targets, which were mounted 15–20 cm from him. This method
tended to produce smaller stains and generally lacked mucus table wallpaper (STRIA Pattern 99417F), with one coat of Zins-
strands ser 1-2-3 primer. The target was rotated during pattern
Method 2 An experimenter took a small volume of blood (<1 mL) and construction so that the wallpaper texture ran vertically. The
mixed it gently with saliva before coughing from the front of chipboard surface was made from oriented strand board (OSB),
the mouth directly toward the targets, which were mounted
50–90 cm from him. This method tended to give a larger
which comprises wood logs (e.g., pines and aspen) that are
range of spatter sizes and frequently contained mucus strands chipped and oriented in random directions.

Contextual Information—In addition to varying the type of

TABLE 4––Methods used to create firearms-related bloodstain patterns. For bloodstain pattern, pattern extent, and target substrate, case sum-
both methods, a 0.22 caliber bullet was fired from a pistol through a blood- maries were created to manipulate the contextual information asso-
soaked sponge. Some targets required multiple shots to obtain the desired ciated with each target. Contextual information was presented in
amount of pattern.
the form of a short vignette that provided background information
Method 1 Back spatter was collected from targets positioned 120 cm in about how the bloodstain pattern was found and what was known
front of a blood-soaked sponge about the case. This included information such as eyewitness
Method 2 Forward spatter was collected from targets positioned 120 cm to reports, the position of the pattern/victim, injuries sustained by the
the rear of the sponge victim, and any weapons that were found or assumed to have been
used. All scenarios indicted a single bleeder and included a DNA
TABLE 5––Thresholds for determining pattern extent. result. The scenario either contained information that supported the
correct classification (positively biasing), was misleading toward a
Total N Stains N Stains > 1 mm Extent particular incorrect classification (negatively biasing), or contained
<50* <10 Minimum no directional information (neutral). No analyst received the same
<50 >10 Medium vignette twice. Two examples are given in Table 6.
50–500 <50 Medium
50–500 >50 Maximum
>500 Maximum Procedure
*For cast-off patterns, a minimum pattern was deemed to be fewer than Analysts were initially invited via email. In both the invitation
four tightly linear stains, or fewer than 10 broadly linear stains. If there were email and in a letter to the analysts after they had agreed to par-
more than 10 linear stains, the pattern was considered to have medium extent.
ticipate, they were informed that the aim of the study was not to
test competency, but rather the reliability of BPA methodology.
Substrate—Targets were one of three different rigid-surface Furthermore, they were informed that all responses would
substrates designed to represent varying levels of anticipated remain anonymous and could in no way be linked to any speci-
identification difficulty; paint, wallpaper, and chipboard. Two fic analyst. To further ensure anonymity, an independent third-
coats of white Zinsser 1-2-3 primer were used for the painted party liaised with study participants.
 15564029, 2016, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/1556-4029.13091 by University Of Northumbria At Newcastle, Wiley Online Library on [21/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TAYLOR ET AL. . CLASSIFICATION RELIABILITY IN BPA 925

Materials were only sent to analysts after they had indicated a

willingness to participate. Each analyst received 15 or 16 targets,
and a response sheet associated with each target. A number was
placed at the top of each target; this identified which target cor-
responded to each response sheet and indicated the pattern align-
ment during pattern construction. No two analysts received the
same targets.
Each of the manipulated variables was counterbalanced across
the other variables and across the entire sample.
The response sheet included the following instructions: “You
are preparing your final report for investigators, which could
end up being presented in court. Please give your opinion as to
the pattern type or types that could account for the stains on the
sample target.”
A list of pattern types, along with SWGSTAIN definitions, FIG. 1––Proportion of response outcomes as a function of pattern type.
was provided as follows: cast-off, drip trail, impact, saturation,
splash, transfer, drip, expiration, pool, spatter from gunshot
trauma (forward or back spatter), swipe, insect stain, drip stain,
flow, projected (e.g., arterial), and wipe.
Analysts could select any number of patterns, or indicate that
“I can’t state that any of the above patterns would account for
the stains on the sample target.”
After classifying all target patterns, analysts emailed or posted
their responses to the independent third-party.

Results
Error and Accuracy Rate Determination
Analysts’ responses were considered correct if they selected
the pattern type that represented the true mechanistic cause for FIG. 2––Proportion of response outcomes as a function of pattern extent.
pattern (or selected this pattern type among multiple selections).
Analysts’ responses were considered incorrect if the true pattern pattern extent, v2 (4, N = 413) = 26.74, p < 0.0001. As might
type was not selected in their choice(s). If analysts could not state be expected, analysts’ accuracy improved as the extent of blood-
that any of the listed patterns could account for the target pattern, staining in the pattern (essentially the number of stains present
their responses were considered inconclusive. If analysts had clas- on the target) increased (Fig. 2). Rather than the decrease in
sified the pattern by indicating all listed types as possible, these staining extent increasing analysts’ propensity to make errors,
responses would have been considered inconclusive; none did so. the decrease in correct classifications was matched with an
increase in inconclusive classifications.
Pattern Classification Accuracy
Effect of Target Substrate
Twenty-seven analysts made judgements on 15–16 target pat-
terns each, yielding 416 unique assessments. Three assessments There was a significant overall difference in the frequency of
were removed from the analysis because no response was provided; correct, incorrect, and inconclusive responses as a function of
413 responses remained. In total, analysts were correct for 69.5% the substrate on which the bloodstain patterns were deposited, v2
of classifications and made errors for 13.1% of classifications. The (4, N = 413) = 38.64, p < 0.0001. Analysts made the greatest
remaining 17.4% of responses were inconclusive. Chi-squared (v2) number of correct classifications on the white-painted surface,
tests were performed to determine whether the frequency of correct, followed by the wallpaper surface (Fig. 3). The chipboard sur-
incorrect, and inconclusive responses varied significantly face resulted in the lowest proportion of correct responses and
(p < 0.05) as a function of each of the variables under examination. greatest proportion of inconclusive responses. The chipboard sur-
face had a highly patterned finish with many surface features
that could have been confused with small bloodstains. Analysts
Effect of Pattern Type
had little or no scope for using any enhancement methods that
There was a significant overall difference in the frequency of ana- might have assisted them with distinguishing bloodstains from
lysts’ correct, incorrect, and inconclusive responses as a function of artefacts. Analysts’ decrease in correct classifications was
pattern type, v2 (6, N = 413) = 13.871, p = 0.031. This was matched by an increase in inconclusive classifications, rather
mainly due to analysts’ success in identifying expirated patterns than an increase in erroneous classifications.
(81% correct compared to 64–69% for the remaining types; Fig. 1).
Effect of Contextual Information
Effect of Pattern Extent
There was a significant overall difference in the number of
There was a significant overall difference in the frequency of correct, incorrect, and inconclusive responses as a function of
correct, incorrect, and inconclusive responses as a function of the direction of the contextual cues, v2 (4, N = 413) = 12.39,
 15564029, 2016, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/1556-4029.13091 by University Of Northumbria At Newcastle, Wiley Online Library on [21/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
926 JOURNAL OF FORENSIC SCIENCES

Where a classification was made, either by choosing a single

pattern or by nominating more than one pattern, 13.1% of these
classifications did not include the correct pattern type. This find-
ing can be considered the first approximation of an overall error
rate for bloodstain pattern classification on rigid non-absorbent
surfaces.
Some patterns were more reliably classified than others. In
particular, the error rate was 4% for expirated patterns compared
with 19% for impact patterns. The characteristic features of expi-
rated patterns (e.g., mucus strands and air bubbles) were gener-
ally evident within these patterns, which could explain why this
pattern type was the most recognizable. The second most recog-
nizable pattern type was cast-off spatter. Many of the distinctive
features of a cast-off pattern, such as a linear or curvilinear dis-
FIG. 3––Proportion of response outcomes as a function of target sub- tribution of stains, were also generally present in the cast-off pat-
strate. terns prepared in this study.
Other than a lack of distinctive features in a pattern, it is not
clear why the proportion of incorrect classifications for impact
and firearms-related spatter patterns was higher. It is possible
that the features that were observed in these patterns were those
shared by other pattern types. For example, many of the impact
patterns had a somewhat narrow distribution of stains, similar to
cast-off. Indeed, on closer observation of the data, many of the
impact patterns were incorrectly classified as cast-off. Firearms-
related spatter patterns in this study did not always have a large
proportion of the very small “mist” stains usually associated with
this pattern type, which may explain why firearms related spatter
was excluded as a mechanism for some of these patterns.
Looking more closely at the other variables used in this study,
it is evident that where the amount of available pattern was lim-
ited or the substrate made stain visualization more difficult (e.g.,
FIG. 4––Proportion of response outcomes as a function of direction of chipboard) the proportion of correctly classified patterns
contextual cues. decreased. However, the reduction in classification accuracy was
generally accompanied by an increase in the proportion of incon-
p = 0.015 (Fig. 4). When a positive context was presented, ana- clusive responses. So where the pattern was more difficult to
lysts were less likely to make an incorrect classification than recognize, analysts became more conservative in their judgment,
when the context was neutral, with the overall error rate drop- which is what is to be expected from a reliable method.
ping from 11% to 8%. Conversely, when a negative context was The error rates estimated from this study are based, in part, on
presented analysts were more likely to make an incorrect classi- the conditions chosen for examination. These conditions were
fication than when the context was neutral, with the overall error designed to be a fair representation of real casework but remain
rate increasing to 20%. The rate of inconclusive decisions was the subjective judgement of the experimenters. In particular, the
highest when the contextual information contained no directional patterns used in the study were restricted to common spatter
cues (21%), relative to when there were positively (15%) or neg- types. Had other more easily recognized pattern types, such as
atively biasing cues (16%). drip stains, drip patterns, transferred blood, and saturation pat-
terns been included, the proportion of correct classifications may
have been higher. In addition, the size of the targets meant that
Discussion
in many cases the analysts received only a portion of the total
The purpose of this study was to assess the reliability of clas- pattern. As well as reducing the number of bloodstains available
sification decisions in bloodstain pattern analysis. To do this, for analysis, the limited pattern size meant that it may not have
well-trained and highly experienced analysts examined over 400 been possible to perform other analyses such as area of origin
spatter patterns on rigid non-absorbent surfaces, representing sur- determinations. Even with a small number of stains and only a
faces frequently encountered at crime scenes. Four spatter pattern portion of the complete pattern available, however, many ana-
types commonly encountered in crime scene investigations were lysts were prepared to make a classification. Finally, the use of a
used. The extent of available pattern, the nature of the substrate, coating of lacquer removed the potential to perform any chemi-
and the type of contextual information were varied in a balanced cal tests on the bloodstains, which may have hindered analysts’
experiment designed to determine the effect of these variables ability to confirm, for example, the presence of saliva to indicate
on pattern classification accuracy. expirated blood. It is worth noting that some of these limitations
Analysts were required to classify the pattern to a standard exist in casework, especially where analysts are working solely
required for court. To do this, analysts were allowed to select from photographs. While any of these limitations could have
one or more pattern classifications that they determined could reduced the classification success rate, none would be expected
not be excluded. Within the chosen range of conditions, the to increase erroneous classifications, because such errors mean
17.4% overall rate of inconclusive decisions gives some measure the correct pattern type has been excluded. Rather, an increase in
of interpretation difficulty for the patterns created in this study. inconclusive classifications would be expected, which was
 15564029, 2016, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/1556-4029.13091 by University Of Northumbria At Newcastle, Wiley Online Library on [21/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TAYLOR ET AL. . CLASSIFICATION RELIABILITY IN BPA 927

indeed the observed outcome in this study for the pattern extent A study of this sort has limitations, not the least of which is
and substrate limitations. the fact that analysts were not making decisions in the context
Of particular interest in this study was whether contextual of a real case, with the associated demands of a court. Within
information can influence classification decisions. Indeed, where the restrictions of the study, however, a useful baseline has been
a scenario was offered that deliberately pointed analysts toward established for the expected error rate in bloodstain pattern clas-
the correct classification, the proportion of misclassifications that sification.
resulted was lower (8%) than that observed for patterns with
neutral scenarios (11%).
References
Of more concern, the proportion of misclassifications that
resulted when the contextual information pointed toward an incor- 1. Scientific Working Group on Bloodstain Pattern Analysis. Guidelines for
rect pattern type was considerably higher (20%) than when the the minimum educational and training requirements for bloodstain pattern
analysts. Forensic Sci Commun 2008;10(1):1–5.
contextual information was neutral (11%). This study, therefore, 2. Scientific Working Group on Bloodstain Pattern Analysis. Recommended
has produced evidence that analysts consider contextual informa- terminology. Forensic Sci Commun 2009;11(2):1–3.
tion when classifying bloodstain patterns and that this information 3. Scientific Working Group on Bloodstain Pattern Analysis. Guidelines for
can influence accuracy. These findings may be evidence for con- a quality assurance program in bloodstain pattern analysis. Forensic Sci
Commun 2008;10(1):1–4.
firmation bias (19), where analysts interpreted the pattern in line
4. National Research Council. Strengthening forensic science in the United
with the expectation created by the contextual cues. States: a path forward. Washington, DC: The National Academy of
In situations where data from a bloodstain pattern is limited, it Sciences, 2009.
may be tempting for analysts to seek out contextual information 5. Aguirre-Jarquin v. State. Supreme Court of Florida, 2014.
in order to help reach a decision. Unfortunately, contextual bias 6. Camm v. State. NE 2d: Ind: Court of Appeals; 2011; 205.
7. Gardner RM. Defining a methodology for bloodstain pattern analysis. J
is most likely to occur when data are ambiguous (20–24), and in Forensic Identif 2006;56(4):549–57.
these situations, efforts to avoid contextual information are par- 8. Saviano J. Articulating a concise scientific methodology for bloodstain
ticularly important. The nature of forensic science is such that pattern analysis. J Forensic Identif 2005;55(4):461–70.
analyses are rarely conducted without a degree of ambiguity and 9. Dror IE, Charlton D, Peron AE. Contextual information renders experts
vulnerable to making erroneous identifications. Forensic Sci Int 2006;156
the need to apply subjective judgements. With a reliable method-
(1):74–8.
ology, however, it should be possible to reach sound conclusions 10. Dror IE, Hampikian G. Subjectivity and bias in forensic DNA mixture
even in the absence of ideal conditions. interpretation. Sci Justice 2011;51(4):204–8.
Analysts’ use of contextual information to inform classifica- 11. Dror IE, Rosenthal R. Meta-analytically quantifying the reliability and
tion decisions could reflect the fact that the boundary between biasability of forensic experts. J Forensic Sci 2008;53(4):900–3.
12. Nakhaeizadeh S, Dror IE, Morgan RM. Cognitive bias in forensic
pattern classification and crime scene reconstruction is often anthropology: visual assessment of skeletal remains is susceptible to con-
blurred. This blurred boundary is not helped by the fact that, at firmation bias. Sci Justice 2014;54(3):208–14.
present, there is not a rigorous protocol for BPA that distin- 13. Kerstholt J, Eikelboom A, Dijkman T, Stoel R, Hermsen R, van Leuven
guishes the two processes. This means that, at the stage of pat- B. Does suggestive information cause a confirmation bias in bullet com-
parisons? Forensic Sci Int 2010;198(1):138–42.
tern classification, additional case-specific information, such as
14. Dror IE, Wertheim K, Fraser-Mackenzie P, Walajtys J. The impact of
medical findings, case circumstances, and eye-witness statements human–technology cooperation and distributed cognition in forensic
could influence pattern interpretation decisions (25). This prob- science: biasing effects of AFIS contextual information on human
lem is compounded by the fact that the BPA terminology used experts. J Forensic Sci 2012;57(2):343–52.
to describe pattern classifications generally describe the mecha- 15. Osborne NK, Woods S, Kieser J, Zajac R. Does contextual information
bias bitemark comparisons? Sci Justice 2014 Jul;54(4):267–73.
nism of pattern formation (e.g., cast-off, expiration). Thus classi- 16. Kerstholt JH, Eikelboom AR. Effects of prior interpretation on situation
fications actually form components of a reconstruction theory, assessment in crime analysis. J Behav Decis Making 2007;20(5):455–65.
rather than purely representing a summary of pattern characteris- 17. Kerstholt JH, Paashuis R, Sjerps M. Shoe print examinations: effects of
tics. Moving forward, it could be advantageous for the BPA expectation, complexity and experience. Forensic Sci Int 2007;165(1):30–4.
18. Fraser-Mackenzie PA, Dror IE, Wertheim K. Cognitive and contextual
community to agree on a standard methodology for the analysis
influences in determination of latent fingerprint suitability for identifica-
of bloodstain patterns that includes a better distinction between tion judgments. Sci Justice 2013 Jun;53(2):144–53.
classification and reconstruction, and on terminology that relies 19. Nickerson RS. Confirmation bias: a ubiquitous phenomenon in many
less on mechanistic descriptions of patterns. guises. Rev Gen Psychol 1998;2(2):175.
20. Andrews S, Bond R. Lexical expertise and reading skill: bottom-up and
top-down processing of lexical ambiguity. Read Writ 2009;22:687–711.
Conclusions 21. Bartels BL. Top-down and bottom-up models of judicial reasoning. In:
Klein DE, Mitchell G, editors. The psychology of judicial decision mak-
One significant implication of this study is the inference that ing. New York, NY: Oxford University Press, 2010;41–56.
contextual information is being integrated into pattern classifica- 22. Ha YW, Hoch SJ. Ambiguity, processing strategy, and advertising-evi-
dence interactions. J Consumer Res 1989;16:354–60.
tion decisions. The assumption made here is that this is not a
23. Giannelli PC. Confirmation bias. Criminal Justice 2007;22:60–1.
competency issue, given that the participants in this study were 24. Osborne NK, Zajac R. An imperfect match? Crime-related context influ-
all experienced bloodstain pattern analysts. This vulnerability to ences fingerprint decisions. Appl Cognitive Psychol 2015;30(1):126–34.
context effects justifies further investigation. Based on the find- 25. Jamieson A. A rational approach to the principles and practice of crime
ings of this study, it seems prudent for practitioners and agencies scene investigation: I. Principles. Sci Justice 2004;44(1):3–7.
to take steps to minimize the effects of contextual information.
Additional information and reprint requests:
The practicalities of this may vary from agency to agency and Michael C. Taylor, Ph.D.
may not be straightforward, as many analysts are immersed in Institute of Environmental Science and Research Ltd
an investigation in ways that make it difficult to control the flow PO Box 29181
of information. However, further research efforts could see valid Christchurch 8540
New Zealand
contextual information management methods emerge.
E-mail: [email protected]