In Consideration of False Negatives
By Boyd Baumgartner, Latent Print Examiner

There is little room for the perception of fallibility in the arena of friction ridge identification. With a public fascinated by forensics and the identification sciences under constant legal and cultural scrutiny there is always someone looking for a chink in the armor of credibility that has been afforded to the discipline of friction ridge identification. Such credibility is the result of several factors including a history of accurate application, a supporting body of scientific research underlying the practice, the inability of critics to dissuade public opinion, and the lack of an avenue through which a legal challenge could be pursued.

Along with the modern age came the accessibility of information, relentless media coverage of high profile cases involving friction ridge evidence, and evolving case law. While none of these factors have been able to undermine the scientific validity of friction ridge identification, they have been able to color the public consciousness with various degrees of truthfulness and have forced the friction ridge identification industry to articulate and justify their philosophies and methods.

Fundamental to those justifications and the ethical underpinnings that are inherent in opinions of identity, are admissions of error when they have been made. Due to the ternary nature of the opinions available to friction ridge identification practitioners and the qualitative subjectivity of the opinion, past definitions of error seem to have been limited to those opinions that are falsely positive, that is when an area of friction ridge skin has been individualized incorrectly.

The definition of error though, is more complex than just being wrong. It is a nuanced occurrence, conditional on cognitive, philosophical and experiential factors. More specifically it is dependant on how the identification methodology is viewed, the knowledge of the practitioner, the threshold of sufficiency per opinion, and the logic that leads to the opinion of identity.

ACE-V, the methodology of friction ridge identification can be thought of in one of two ways, as hypothesis testing or as ternary predicate logic. If one views ACE-V as hypothesis testing, the alternate hypothesis and null hypotheses are first formed and then tested. Based upon this view of ACE-V, there can be only two conclusions: accept or reject the null hypothesis.

Errors under this line of thinking consist of Type I errors and Type II errors which correspond to incorrectly accepting the null hypothesis when it is false or incorrectly rejecting the null hypothesis when it is true. These are called false positives and false negatives respectively. False positives correspond to incorrectly associating the identity of an individual to friction ridge detail thus linking them to a crime. False negatives correspond to failing to correctly associate the identity of an individual to friction ridge detail and thus failing to link them to a crime.

There are two problems with viewing ACE-V as hypothesis testing; the qualitative aspect inherent in the friction ridge identification methodology and the meanings of conclusions reached in hypothesis testing. While hypothesis are formed and tested under the ACE-V process, hypothesis testing is better adapted to meet the needs of other disciplines.

Classical hypothesis testing is performed on distinct measurable quantities and is concerned with probabilities, significance levels, and distributions. This means that quantity dictates qualitative interpretations. Friction ridge identification does not have the luxury of dealing with measurable quantities in that way. The existence of what is being observed can be called into question in friction ridge identification. There is no test that can confirm or deny the existence of observable characteristics as they are the result of perception and interpretation and as such are influenced by variable distortions and physiological systems. Therefore, quantities in friction ridge identification have variable value. An increased quantity cannot be strongly correlated with increased qualitative value. The bottom line is that quality is successive of quantity in hypothesis testing whereas this is not the case for friction ridge identifications.

Further compounding the qualitative dilemma is the problem of multiple comparisons. “The multiple comparisons problem occurs when one subjects a number of independent observations to the same acceptance criterion that would be used when considering a single event.” [1] Considering that friction ridge identification concerns itself with a number of independent observations of various level 2 and level 3 characteristics, this becomes a prevalent risk when employing the philosophy of hypothesis testing. This known problem in hypothesis testing leads directly to an increase in Type I errors or false positives.

Additionally, the available conclusions in hypothesis testing are inadequate to fully represent what is happening in evaluating friction ridge skin. Conclusions in hypothesis testing are given solely in terms of the null hypothesis. By definition, the “null hypothesis” is the hypothesis of no difference. [2] Therefore, definitively speaking, you must always structure your test so that your null hypothesis is such that any two friction ridge impressions are no different. The null hypothesis is presumed to be true until evidence in the form of a hypothesis test (called the alternate hypothesis) shows otherwise.

The conclusions regarding the hypothesis test are one of two possibilities. Reject the null hypothesis or accept the null hypothesis. “To reject the null hypothesis is to conclude that it is false.” [3] The problem appears however, when you accept the null hypothesis, as accepting hypothesis in hypothesis testing only means that you do not have evidence to believe otherwise.

This creates deficiency in being able to articulate that the results of your hypothesis test are inconclusive, which is a real possibility. That is to say, that the information you have available to you does not lend itself to either individualization or to exclusion. The closest available argument you would have under the dictates of hypothesis testing is that you do not have enough evidence to reject the possibility that the impressions being compared are different. This equates to an acceptance of the alternate hypothesis where the null hypothesis is that there is no difference between the impressions being compared.

It is my opinion that a more accurate way is to view ACE-V as ternary predicate logic conditional on the meta-analysis of observable characteristics in friction ridge skin. Ternary predicate logic simply states that an out come of -1, 0, or +1 is possible based upon predicate variables which can be existentially quantified. The values of -1, 0, and +1 correspond to exclusion, inconclusive and individualization respectively and are conditional based upon the validity of each argument that a level 2 or 3 characteristic exists.

Framing ACE-V in this manner puts friction ridge identification into a formal logic system. As such, the premises (or characteristics identified in the analysis) can possess the qualities of consistency, soundness, and completeness and conclusions can be true, valid and sound. This makes friction ridge conclusions logical, valid deductions when ACE-V is applied correctly.

Now that an understanding of the logical model has been reached, it follows that errors should be defined. Errors can and should be considered based upon the aspects which involve them. Areas involving error include empirical, scientific, logical, industrial and ethical aspects of friction ridge identification, as each has some bearing on the existence and degree of seriousness of an error.

In the strictest sense of the definition, an error is a variance between a measured value and its true value. In this sense, as it relates to friction ridge identification, an error can occur when a Latent Examiner’s experience level is low. In the same way we would expect a novice driver to make less skillful judgments than a race car driver, a novice Latent Print Examiner would make less skillful observations than someone with more experience, assuming all else is equal. Therefore there is an empirical consideration when understanding an error.

From a scientific perspective, errors are the result of observational variances and are labeled either random or systematic. As it relates to friction ridge identification a random error could be typified as an imprecise observation due to distortion. When present, this phenomenon will vary from comparison to comparison, as the effects of distortion do. Systematic error however, is consistent. An example of systematic error as it relates to friction ridge identification would be pattern blindness or a similar visual imperfection which affects the ability to observe or perceive in some consistent way.

A logical error is the outcome of incorporating empirical and observational errors into deductive reasoning. Logical errors do not necessitate invalid arguments, but stem from unsound arguments which predicate themselves on empirical and observational errors. It is in this way that errors are expressed and constitute the measure of accuracy and precision when a subsequently reviewed in the Verification phase of ACE-V.

Errors as a function of the standards in the industry of friction ridge identification are governed by different sufficiency thresholds. Thresholds for individualization are higher than the thresholds for exclusion. As such, and in consideration of the empirical, scientific and logical aspects of errors, ACE-V is more error tolerant of exclusions because of the lower threshold in determining exclusion is justified.

Lastly, errors have real world implications. As it relates to friction ridge identification errors have ethical implications. In the criminal justice system it is more egregious to find and innocent person guilty of a crime than it is to let a guilty person go free. This is confirmed by the fact that there have only been five known false negatives determinations within the last 40 years.[4] Even critics of the friction ridge identification industry recognize as such when they say “The rate and occurrence of false positives, however, is more controversial.” [5] While this should have no effect on the decision making processes involved in friction ridge identification, it is a factor in the consideration of what constitutes an error.

Therefore, it is understandable given the cumulative failings of the various aspects and the high thresholds involved in the decision making process that falsely positive friction ridge identifications would be given the connotation of mistake that ‘error’ implies. This is less so however, for erroneous exclusions or false negatives as they are also known.

While it is valid to arrive at an ‘inconclusive’ determination, this decision is arrived at based upon subjective inabilities. Experience level and the interpretation of the industry standards play into this determination. If an examiner does not have the experience to be able to perceive the sufficiency requirements for individualization, then they must rule either inconclusive or exclusive. Some definitions of erroneous exclusion include inconclusive determinations when a definitive conclusion can be reached. [6]

Furthermore given the lower industry threshold for exclusionary standards, a false negative can be reached when an examiner is unable to articulate the reasoning behind a discrepancy. This is a sufficient basis for exclusion according to the Scientific Working Group on Friction Ridge Analysis, Study and Technology (SWGFAST). Further consideration to SWGFAST’s standards for conclusions, sufficient quantity and quality could be artificially reduced due to inexperience, resulting in an exclusion being incorrectly attributed to a latent impression.

One last consideration regarding the inclusion of false negatives into the realm of ‘error’ is where in the process a false negative is discovered. There is a distinction to be made between an error that makes it into evidence where a conviction relies in part or solely on the testimony of a latent print examiner and one that does not.

In conclusion, it is my opinion that there are combinations of variables that can lead to false negative conclusions, very few of which show a causal relationship to the misapplication of fundamental principles governing friction ridge identification by an examiner. Generally speaking, individuals’ biases color how they view false negative results. False negatives can be viewed as part of the general iterative process that mirrors scientific methodologies to those who study science. They can be viewed as valid arguments with faulty premises to those who study philosophy. If you are a critic of fingerprints however, they are more likely to be characterized as failings contributing to the ‘unreliability’ of the case against a person for whom you are being paid to act as an expert. The bottom line is that as an industry the current accepted methodology is biased in favor of minimizing false positive errors and rightly so. This is by design, as the morals of our judicial system and society place value on evidence that discerns guilt beyond a reasonable doubt, which friction ridge identification certainly accomplishes.

References:

1. http://en.wikipedia.org/wiki/Multiple_comparisons

2. http://davidmlane.com/hyperstat/A73664.html © 1993-2007 David M. Lane

3. http://www.itl.nist.gov/div898/handbook/eda/section3/eda35.htm © 2006 James J. Filliben

4. http://www.fprints.nwlean.net/e.htm © 2002-2007, Michele Triplett.

5. Simon A. Cole, "More Than Zero: Accounting for Error in Latent Fingerprint Identification," Journal of Criminal Law & Criminology, Volume 95, Number 3 (Spring 2005), pp. 985-1078.

6. http://www.fprints.nwlean.net/t.htm © 2002-2007, Michele Triplett.