Error rate was measured in terms of the difference between numbers of excellent look-alikes observed in precise fingerprint populations compared to the number of look-alikes calculated  by the T-Model.  It was also measured in terms of the number of look-alikes calculated for the most notable erroneous identifications ever recorded and largest and best amounts of corresponding ridge formations ever seen in a non-match based on conservative upper bound fingerprint populations.

Based on results from the 4 reported experiments (see Validation Study), the number of look-alikes observed in each experiment was less than the number of look-alikes expected.   As a result, it may be stated that the error rate for an observed amount of corresponding ridge formations in a latent v. exemplar to be less than the expected number of look-alikes calculated by the T-Model with subsequent risk for erroneous individualization, is zero.

Based on results for numbers of look-alikes calculated by the T-Model for the most notable erroneous fingerprint identification ever made, error rate was measured in terms of the number of times the model incorrectly established the number of look-alikes likely to occur as less than 1, and therefore sufficient to individualize.  Based on conservative upper bound fingerprint populations, the model correctly calculated the number of look-alikes for these erroneous identifications as greater than 1, and therefore without valid basis for sufficiency to individualize.  As a result, the error rate for the T-Model to identify the most notable erroneous fingerprint identifications ever recorded as having amounts of corresponding ridge formations insufficient to individualize, is zero.  
 
Based on results for numbers of look-alikes calculated for the largest and best look-alikes ever published, presented during examiner training, or otherwise recorded, error rate was measured in terms of the number of times the model incorrectly established the number of look-alikes likely to occur as less than 1 and therefore sufficient to individualize.  Based on  results for the number of look-alikes calculated by the T-Model using conservative upper bound fingerprint populations, the error rate for the model to identify the largest and best amounts of corresponding ridge formations ever seen in a non-match as insufficient to individualize, is zero. 

Based on results for numbers of look-alikes calculated for the largest and best look-alikes found as a result of local, state and national AFIS searches, error rate was measured in terms of the number of times the model incorrectly established the number of look-alikes likely to occur as less than 1 and therefore sufficient to individualize.   Based on Santa Clara County AFIS [24], State of California AFIS21 [26], and national (FBI) IAFIS [21] database searches, the T-Model reliability established numbers of look-alikes as greater than 1 for the largest and best look-alikes found and therefore without valid basis to establish sufficiency to individualize.  The error rate for the T-Model to identify the largest and best look-alikes found as a result of local, state and/or national AFIS searches as insufficient to individualize, is zero.

Based on results from the 4 reported experiments (see Validation Study) the T Model calculates roughly a 15% larger number of look-alikes compared to what is the actual number observed.  For example, if the number of look-alikes for an amount of matching fingerprint ridge features in two impressions is calculated by the T-Model to have 1.12 look-alikes based on an AFIS fingerprint population of X, e.g., greater than 1 and therefore exculpatory in favor of a defendant, then the actual number of look-alikes observed, if all fingerprints were compared, is predicted to be roughly 0.952, e.g., less than 1 and therefore inculpatory in favor of the prosecution.  As a result, T-Model calculations slightly favor the defendant.  For purposes of performing criminal casework in a conservative manner, e.g., to provide additional quality assurance to help guard against the erroneous fingerprint identification of an innocent person, these results are extremely appealing.

NEXT PAGE >>> 

Continued Validation Studies

Latent fingerprints are searched in county, state and national AFIS databases on a daily basis.  As a result, trillions of fingerprint searches are performed on a daily basis [43].  Latent print examiners routinely record the largest and best look-likes found during the course of these searches for purposes of examiner training.  These AFIS searches represent continued validation studies to find the largest and best look-alikes possible.   

The IAI position concerning latent fingerprint identification error rate states “it is generally held by practitioners, scientists, and legal authorities that the error rate for fingerprint identifications is extremely small, statistically insignificant, and not due to the methodology but instead of the inherent risk of error in any human endeavor".  The IAI also states that errors, or erroneous individualizations, "are corrected by applying the ACE-V method correctly" [40].

Based on a .00033 error rate established by Langenburg for fingerprint examiners to effect erroneous conclusions [51], any ACE-V method that fails to include a quality assurance method that has a proven error rate of zero cannot be used to "correct errors".

As a result, the IAI position that "errors are corrected by applying ACE-V correctly", which means using only the non-statistical "professional judgment" approach again to define sufficiency to individualize, should not be considered valid unless ACE-V includes the mandate that valid basis for sufficiency to individualize includes either, and at the very least, that amounts of corresponding ridge formations present in two impressions clearly and beyond doubt exceed the largest and best amount ever recorded in a non-match, i.e. look-alike, or based on the T Model, or equivalent, establish the number of look-alikes likely to occur as less than or equal to 1 based on relevant fingerprint population.