Presentation on theme: "Database Searches Non-random samples of N individuals Typically individuals convicted of some crime Maryland, people arrested but not convicted."— Presentation transcript:
Database Searches Non-random samples of N individuals Typically individuals convicted of some crime Maryland, people arrested but not convicted
What does the scientific community think happens? Balding and Donnelly (1996):..a search is made through the database with the result that exactly one of the profiles in the database is found to match the crime scene profile. NRCII:..the suspect is initially identified by searching a database to find a DNA profile matching that left at a crime scene. Stockmarr (1999):..exactly one profile in D is found to match TP Evett and Weir (1998):..the profile from the vaginal swab was searched against the database and his profile was found to match.
So whats a match? Evett and Weir (1998): They suppose that there are two samples, a crime and suspect sample that are typed by DNA techniques. The samples match when, The two samples are found to be of the same type.
What really happens? Target profile and database, candidate, profiles may be compared by two different criteria: high stringency and moderate stringency High stringency has the usual meaning of a match, that every allele in the target must be seen in the candidate and there can be no extra alleles in the candidate profile
Moderate Stringency If the target is a mixture with three or more alleles then matching candidates would be any of the possible pairs of samples Target = 9,11,14 Moderate matches = 9/9, 9/11, 9/14, 11/11, 11/14,14/14 This is similar to computing included genotypes in mixtures
Moderate Stringency: one allele If either the target or the candidate profile has only one allele, then moderate stringency matches are all genotypes with at least one copy of the single allele Target profile = 12 Matching candidates = 12/X, where X is any other allele This criteria is more generous than the typical definitions for mixture inclusions
Implications Vague protocols for matches by labs creates substantial liability on what constitutes a match. Example: single source target profile 13/13, would normally not be called a match to a 12/13 candidate Since labs insist on invoking allelic drop out the possibility of this type of match must always be considered relevant
Statistical Implications The class of matching genotypes to a 13/13 profile is then 13/X, which is greater than the frequency of the 13 allele but less than twice that frequency Apparently in California the moderate criteria is turned on in all searches Even if the search identifies a candidate that matches at high stringency the statistical penalty for the moderate search must be paid (see Venegas)
Are Balding, Donnelly, Evett, Weir, NRCII, and Stockmarr idiots? NO! However, because almost all forensic labs do not allow access to their software or databases, the scientific community is unaware of the real match criteria If scientific evidence is not yet ready for both scientific scrutiny and public re- evaluation by others, it is not yet ready for court. (NRC I)
Typing Errors Evett and Weir assume that the chance of not finding a match if the true perpetrator is in the database is zero. The difference at THO1 is a typing error FBI Bahamian  15 /16 17 /20 23 /24 10 /12 30 / /16 11 /13 8 /11 10 /12 9 /12 11 /11 8 /9 9 /11  15 /16 17 /20 23 /24 10 /12 30 / /16 11 /13 8 /11 10 /12 9 /12 11 /11 8 /10 9 /11
Solutions Any lab following NRCII or likelihood ratios would have to compute random match probabilities adding up all potential matching profiles Avoid these headaches and follow the NRC I recommendations