|
Stochastics - The real science behind
forensic pattern identification
November 24, 2009 by John M.
Collins, CLR Chief Managing Editor &
Director of the
DuPage County Crime Laboratory in Wheaton, IL
Click here to Print
It is the science in which all of
forensic science will grow newer and deeper roots.
It is a mathematical foundation for
common forensic testing methods that never before could be
quantified in understandable terms.
It provides, without a doubt, a
statistical and philosophical framework that can drive
revolutionary research in all of the forensic sciences,
including DNA.
It is called Stochastics – the
science of randomness. Many forensic
scientists have never heard of it. But it
waits with open arms to welcome a profession that is in
desperate need of a home.
For over a century, many of the
classical forensic disciplines such as latent prints,
toolmarks, and footwear impressions have struggled to
overcome a debilitating identity crisis.
Practitioners in these disciplines rightly insist that their
work is rooted in good science. But when
scrutinized by persistent inquisitors who expect conclusions
of uniqueness to be accompanied by a quantitative
assessment, many forensic scientists are left to simply
present their “training and experience” as the ultimate
foundation upon which their conclusions rest.
Undoubtedly, a forensic scientist’s
training and experience are critical and relevant.
We also know that accurate and reliable forensic
determinations are being made each day even though
mathematical formulas don’t yet exist to back them up.
But it is not enough to make forensic science
determinations. They must be
communicated as well.
That’s where the problem lies.
In today’s scientific and legal
environments, presenting one’s training and experience as
the exclusive, underlying foundation of critically important
conclusions, particularly statements of certainty and
uniqueness, is no longer seen as appropriate and never will
be again. No matter how forensic science
practitioners feel about this trend, many current practices
must eventually give way to something more transparent and
perceptibly objective – both for scientific reasons and to
command a higher level of confidence, which is actually what
science is about anyway.
That time may not have come just yet,
but it will. So if the forensic science
community is to be seen as trustworthy, responsible, and
truly desirous of continual advancement, it must actively
support and encourage the newer kinds of research that will
allow scientists to communicate their results more
effectively.
It seems that within the realm of
stochastics all of the nuances and complexities of forensic
science, which so often frustrate judges and lawyers, make
complete sense. Here’s why.
Take something as seemingly simple as
the flip of a coin. Despite all of the
scientific and mathematical knowledge at our disposal –
knowledge that has found cures to the most devastating
diseases and has safely landed astronauts on the moon –
there is no way to predict with any degree of mathematical
certainty on which side a well-flipped coin will land.
Yes, we can calculate the odds that the
coin will land with heads or tails up.
But we cannot conclusively predict it.
In other words, the flip of a coin is
not so simple after all. The random processes that result in
one side facing upwards are infinitely complex and therefore
escape our predictive reach.
Calculating the odds that a coin will
land either heads up or tails up is a deterministic
problem. Predicting exactly which event
will occur is a stochastic problem.
In the pattern identification
disciplines, the words deterministic and stochastic are
generally not used. The terms class
and individual are used instead but they mean the
same thing. In DNA testing, stochastic
phenomena are dealt with on a regular basis and are
described as such. Random appearances and disappearances of
genetic alleles occur unpredictably during the analysis of
extremely low levels of DNA and can’t be mathematically
explained.
In reality, stochastic problems and
processes are everywhere and are relevant to the most
complex issues in the human experience.
Forensic evidence is no exception and the absence of exact
mathematical models to describe what many practitioners
observe in a forensic testing laboratory does not, and
should not, minimize its scientific standing.
Doctors, for example, cannot predict the
exact number of days that a terminally ill patient will
survive after his or her diagnosis. The
outcome is dependent upon stochastic processes.
Behavioral psychologists cannot predict
the chance that a child will grow up to become a serial
killer. The outcome is dependent upon
stochastic processes.
Geophysicists cannot reliably predict
the chance that an active volcano will erupt in the next
year. The outcome is dependent upon
stochastic processes.
The natural world is dominated by random
variations and infinite possibilities, a problem that cursed
even the great Albert Einstein who believed that
mathematical equations could ultimately explain and predict
everything.
“God does not play dice with the
universe,” he once wrote to a friend.
“But God does play poker,” warned the
famous physicist Stephen Hawking.
The patterns observed in latent
fingerprints, on bullets, in a person’s handwriting or on
the bottom surface of their shoes are also the results of
stochastic processes that cannot be controlled.
Patterns that result from these processes are judged
to be unique mainly because decades of observation and
research have only validated the formulation of such
judgments.
But there is one small problem.
Those who conduct comparisons of
stochastic patterns in forensic science know very well that
there are limits to their ability to assess uniqueness.
With enough smudging or interference, a latent
fingerprint cannot be identified to the person who deposited
it. With enough damage, a bullet cannot
be identified to the gun that fired it.
Not all identifications are created
equally. Some are easy.
Some are difficult. In the most
extreme instances, comparisons can result in identifications
with which not all examiners would agree.
It is precisely these extreme instances
– the gray areas in forensic science – that require
attention. It is also the reason that
identifications, or conclusive statements of origin, demand
the highest levels of scientific justification.
Make no mistake, forensic practitioners
present authoritative and reliable opinions about the
identity and origin of latent prints, toolmarks, footwear
impressions, and other stochastic pattern evidence.
But as questions begin to explore a practitioner’s
opinion about all other sources, other people,
other guns, and other shoes, a more
conservative and cautious approach is now expected until the
research justifies “the exclusion of all others.”
In the meantime, a more responsible and
perhaps a more compelling approach is for forensic
scientists to simply state the truth about identifications.
“I have never seen, nor would I expect to see, this
amount of similarity in [friction ridge patterns, bullet
striations, footwear impressions, or other types of pattern
evidence] that came from different sources.”
This is a statement of fact and is
supported by the practitioner’s education, training, and
experience. Arguing, on the other hand,
that an identification excludes every possible source that
ever did, does, or could exist is probably correct, but it
cannot be entirely defended under the scientific
expectations that our courts now have. And
when identifications fall increasingly close to those
marginal gray areas, the risks naturally increase.
So how are the courts to know if an
identification falls close to or far from these margins of
higher risk? Where are the
margins? How much is enough?
Judges and juries have a right to know.
To get a glimpse of where future
research can take the profession, the following questions
are posed.
Would you believe that the odds of two
different guns leaving the same pattern on the primer of a
cartridge case (the small metal disc that ignites the gun
powder when struck by the gun’s firing pin) were less than 1
in 60.5 sextillion? Would a judge or jury
find this information to be compelling if there was ample
research to support the claim?
Consider the following example.
Imagine a circle. Now
imagine dividing this circle into 360 equally spaced
radial-lines (spokes), and 100 equally spaced concentric
circles (percent of the radius).
 If
you count every location where lines intersect, you wind up
with 36,001 “points” of comparison.
Using this rough model, one can quickly
see how easy it is for nature to create unique stochastic
patterns that can be evaluated by forensic scientists.
If two different people drew one of
these hypothetical circle-grids then randomly and
independently selected five of the 36,001 available points,
the odds of them selecting the same pattern of points is as
follows:
Selecting one identical point = 1 in
36,001
Selecting two identical points = 1 in
1.3 billion
Selecting three identical points = 1 in
47 trillion
Selecting four identical points = 1 in
1.6 quintillion
Selecting five identical points = 1 in
60.5 sextillion
The real-life patterns observed by
forensic scientists are even more complex and varied;
therefore, the subsequent statistical expressions of
uniqueness will likely be even more impressive, and possibly
more mathematically definitive, than DNA.
We just need the research to show it.
We can never know the exact
chance that two different persons will leave the same
fingerprint pattern, or that two guns will impart the same
pattern of striae on the surface of a bullet.
The randomness of the stochastic processes involved
simply will not allow it. Perhaps this is
one reason why statistical research in the pattern
identification disciplines was never funded or prioritized
with any enthusiasm.
But research can, in fact, quantify the
margins and establish useful thresholds to help
practitioners explain the significance of their
observations. This will be a huge
achievement.
Coincidently, Taylor and Francis, the
same organization that publishes the new journal Forensic
Science Policy and Management, also publishes
Stochastics – an International Journal of Probability and
Stochastic Processes.
Dr. Elart Von Collani at the University
of Wurzburg in Germany has become a leading expert in the
stochastic sciences. In 1995, he
published an article in the European Journal of
Engineering that seemed to provide a reasonable
explanation for one of the problems observed by the National
Academy of Sciences in its 2009 report, Strengthening
Forensic Science in the United States – A Path Forward.
The NAS report complained that “the
forensic science enterprise is hindered by its extreme
disaggregation—marked by multiple types of
practitioners with different levels of education and
training and different professional cultures and standards
for performance and a reliance on apprentice-type training
and a guild-like structure of disciplines, which work
against the goal of a single forensic science profession.”
But according to Von Collani, this type
of fragmentation is to be expected. “Each field
of traditional science creates its own stochastical branch,”
he explained, “thus preventing unification.”
In other words, fragmentation
exists in the forensic sciences precisely because each
segment of the profession is dealing with its very own
specialized stochastical branch. The details and
minutia observed by a latent print examiner is of no
relevance to a toolmark examiner.
Admittedly, Von Collani was not
writing about forensic science specifically. But there
are strong parallels and common ties that bind all of the
stochastic pattern disciplines. It will take some time
for forensic practitioners to become fully aware of them.
In the meantime, the right research conducted by the right
people can liberate many embattled forensic disciplines and
empower their practitioners to report conclusions with more
scientific certainty.
One way for this transformation to
begin is for forensic scientists and researchers in academia
to think differently about stochastic patterns.
A striation on the surface of a
bullet, for example, is not just a scratch. Nor is the
bifurcation of a friction ridge in a fingerprint pattern
simply a second-level detail.
Instead, both the bullet striation
and the friction ridge bifurcation are events.
And like all events, they have a certain probability of
occurring.
It will take some time for
forensic scientists to stop thinking of patterns as a
collection of characteristics. But sooner than
later, they must begin to evaluate these patterns as a
collection of events. Only then can they further
elevate their respective forensic disciplines to the new
heights of scientific credibility that are now expected.
The forensic pattern
identification disciplines are, and have been, reliable and
valid. It’s the expectations that have changed.
For this reason, research must be designed and executed to
help us meet this new challenge.
This new opportunity to enhance
what is done in the future must not be construed as an
indictment of what was done in the past.
Anyone with scientific sense
knows the difference between the two.
*****
Please
click here to submit a comment.
We welcome opposing view points.
|
