8127/frame/fm

a.
Individualization and Opinion — A Special Case
How do we reconcile an opinion of individualization with the hypothesis
testing model, in which we accept a hypothesis only by failing to disprove
it?* By definition, we are left with only a provisional explanation which, at
any time, may be disproved by further work. This presents an interesting
dilemma for any applied science, including forensic science. If no hypothesis
is ever proved, merely disproved, how can an analyst ever draw any firm
conclusions from the data? Scientists understand that no fact is ever proved,
yet the criminal justice system looks to science for help in establishing facts.
The lawyers roll their eyes at the inability of the scientist to make up her mind,
for the attorney’s life is made up of practical events that require practical
decisions. In its purest form, science is of no value to the judicial system.
However, as a hypothesis continues to be tested, repeated failure to dis-
prove it strengthens our conviction that it is correct and weakens our skep-
ticism that it is wrong. Another way to say this is that, with each piece of
additional information, our uncertainty regarding the truth of the hypothesis
decreases. Finally, the analyst judges that no other potential reference source
could possess the observed traits. Therefore, the chance of encountering a
different reference with these characteristics by chance alone is zero.
Using LRs, the analyst judges the denominator (the probability of the
evidence assuming a different source) to be zero. Dividing by zero drives the
LR to infinity and, consequently also, the posterior odds. Purists, of course,
contend that the denominator can never be zero. Robertson and Vignaux
(1995), of the Bayesian persuasion, recognize this dilemma and suggest an
interim solution.
In these cases it seems that the expert is giving evidence of identity when,
and only when, in his judgment the probability of getting the evidence
assuming the alternative hypothesis is so small that it does not matter what
the numerator or even the prior are. At what point this is reached seems to
be a matter of judgment and experience and there most writers on expert
evidence are content to let the matter rest. This may have had the unfortunate
effect of removing the incentive to carry out the basic research necessary to
build appropriate models. Intellectually, this is unsatisfactory and further
* This must be an intelligent failure to disprove the hypothesis using discriminating tests.
For instance, failure to prove that an individual is not the donor of a stain using secretor
status (80% of the population are secretors ) is virtually meaningless and would not alter
our certainty very much one way or the other. 
8127/frame/ch06 Page 148 Friday, July 21, 2000 11:47 AM

---

Classification, Identification, and Individualization
149
work is required in order to understand the processes involved in making
these decisions. In the meantime the proposal that all forms of scientific
evidence be given in the form of a likelihood ratio is a counsel of perfection.
As a practical matter, we suggest that:
At some 
subjective
point, 
most
qualified 
authorities
would agree that, for
practical applications
, the likelihood that the hypothesis is untrue is so 
small
that it can be ignored.
This concept is critical to moving forward in any applied science. Before we
continue discussing its implications, let’s clarify the meaning of several key
concepts.