Statistics means never having to say you’re certain.
When something is measured in a scientific test, there is uncertainty. Measurement uncertainty is a fundamental scientific concept. But how far off of “exact” might any given test result be? Examiners use statistics and probability to define the amount of uncertainty in tests.
The current thinking in forensic science is that measurement uncertainty – where measurable – should be presented along with the test results. Doing so provides fact-finders with a basis for judging the value of the results in the case. A Path Forward emphasizes the need for determining and stating uncertainty along with results, as does ISO 17025. Many of the Scientific Working Groups (SWGs) and Technical Working Groups (TWGs) are also addressing the issue.
Attorneys will need a basic understanding of what the measurement uncertainty data means, and how it was obtained.
It’s impossible to get very far into a discussion on measurement uncertainty without using words such as statistics and probability, two disciplines that are widely misunderstood by the public. The problem is twofold:
This situation leaves open the possibility for a lot of misleading questions and testimony on the part of lawyers and experts. An attorney does not need to become fluent in statistics and probability, but it is important to grasp a few key concepts, and to become aware of some fallacies that can arise in statistics-based testimony. Among those key concepts are Random Match Probability, Outliers, Frequency and Standard Deviation, which are described in further detail int eh videos at right.
Not every discipline under the forensic science umbrella is amenable to statistical analysis, as it relates to measurement uncertainty. Think of pattern evidence, for example. What statistics or uncertainty measures could be applied to examinations that rely heavily on observation and subjective interpretation (e.g. latent print comparisons, bullet comparisons, bite mark comparisons, or shoe impressions)?
However in quantitative tests, these measures can be extremely important. For instance, if a state has a blood alcohol cut-off of 0.08 for driving under the influence of alcohol, and a defendant’s blood measured 0.078 with a 99% confidence interval of 0.075 to 0.081, there is room to argue that the defendant was not at 0.08.