Using DNA typing in legal suits is controversial in two ways. First, establishing matches between samples collected at a crime scene and those taken from a suspect is problematic. Until recently, there has been little standardization of laboratory criteria for declaring a match (Lander 1991). Similarly, from the perspective of the contestants in a given suit, human error made during the typing process can result in false matches, or in undetected matches. Second, a match is not a conclusive indicator of guilt. Because the molecular markers used in typing are shared among individuals, matches are useful only when the likelihood of being shared is very low. Moreover, this likelihood depends on allele frequencies in a reference population. Because targeted allele frequencies may vary among human populations, controversy surrounds selection of the appropriate reference population. Together, these points of controversy suggest that DNA evidence is not absolutely sufficient, alone, to indicate guilt. This seems trivial; one of the most exacting standards of proof in American jurisprudence is employed in criminal trials, in which proof of guilt is predicted on "reasonable doubt", and which rarely are conducted solely with genetic evidence.
DNA matches are only useful when they are certain, but improving the match determining power of laboratory techniques seems to involve mainly technical issues of standardizing and optimizing them to maximize their reproducibility and minimize opportunities for error. To use those results, however, one must address the chance that an error has occurred yielding either a false match, or failing to detect a true match. False matches seem to elicit more concern than failing to detect a true match. This apparent asymmetry in concern is reflected in our corresponding biased tolerance of the consequences of an error: in the case of a false match, for the accused; or for an undetected true match, for our system of justice. Despite this asymmetry, we do not employ an ultimate threshold for assigning guilt. Instead, our system allows a malleable standard of innocence: guilt beyond a reasonable doubt. The current challenge of DNA evidence is to derive from it an understanding of doubt, not to use it to assign guilt.
Although we may doubt the surety of our methods for detecting all DNA matches, our willingness to tolerate type II errors means that DNA evidence is very useful for establishing innocence. If the DNA of a suspect does not match that collected at a crime scene, we take it as strong support for the suspect's innocence. When the samples do match, we must calculate the likelihood that the match was fortuitous. Essentially, this requires us to evaluate competing hypotheses that either accuse or absolve a suspect, given that the suspect's DNA matches that found at the crime scene. This is done using a likelihood ratio, which employs a Bayesian perspective to modify traditional estimates of error probability, by prior belief about the veracity of the two hypotheses. Traditional estimates of error probability depend on allele frequencies in the reference population and linkage disequilibrium among the several loci that may be used in the analysis. Inbreeding within subgroups of structured population may alter both allele frequencies and linkage disequilibrium. Human populations were until recently probably strongly structured, and the prospect that genetic traces of that structure may persist makes selection of the appropriate reference population theoretically crucial to proper estimating of traditional error probabilities. Interestingly, the sensitivity of these statistical techniques has been shown to be low (as cited in Weir and Evett 1993), by deliberately employing inappropriate reference populations with little effect. If such deliberate misapplications of the methods do not change the outcome of likelihood ratios and ultimately show that doubt in the methods of DNA typing falls within the range of "reasonable", then it seems likely that DNA evidence may be used increasingly in future trials. It may be employed increasingly to assign guilt.
Literature Cited
Lander, E. S. 1991. Lander reply. American Journal of Human Genetics 49:899-903.
Weir, B. S., and I. W. Evett. 1993. Reply to Lewontin. American Journal of Human Genetics 52:206.