TrueAllele solves uninterpretable DNA in mother and daughter double homicide

Back to Publications

Suspect-centric bias in DNA mixture interpretation

Perlin, M.W. Suspect-centric bias in DNA mixture interpretation. Forensic Magazine, 15(3):10-13, 2018.




Bias abounds in criminal justice. Predictive policing can bake bias into software, reflecting and reinforcing prior beliefs. Bail-risk computer programs may entrench pre-trial detention disparity. Human judgment pervades the process. Prosecutor and defender alike passionately argue their client’s case, drawing opposite conclusions from identical facts.

Science is above the fray. Objective data suggest forensic match between crime scene and suspect. Statistical data analysis yields incontrovertible numbers for the strength of match. Cold DNA facts are presented as confirmed theories in court.

But what if DNA analysts could pick and choose their data? Or adjust software parameters to suit their theories? Changing data and parameters will alter forensic match results. Quantitatively, subjective manipulation can artificially inflate match strength. Qualitatively, some DNA evidence that excludes a suspect may be statistically twisted to include him.

Suspect-centric bias has long plagued forensic science. The mythic infallibility of fingerprint analysis was shattered when the FBI misidentified Brandon Mayfield in the Madrid bombing case. Confirmation bias just puts a number to a foregone match conclusion. Suspect-centric thought twists forensic facts to suit prosecution theories.

DNA evidence is not immune to suspect-centric bias. Most DNA evidence is a mixture of two or more people. Popular mixture protocols have crime laboratories first decide whether a suspect’s DNA is in the evidence, and then pick and choose DNA data—all before calculating a match statistic. The FBI has abandoned their debunked Combined Probability of Inclusion (CPI) mixture statistic. But the suspect-centric practice of “decide first, calculate later” continues, with experts selecting DNA data.

Fortunately, new “probabilistic genotyping” methods have been developed. Such software can objectively unmix DNA mixture data into component genotypes to deliver reliable match statistics. But should software let users pick their data? Or dial in chosen parameters? What if a software program lacks the math to use all the data, forcing users to make suspect-centric choices? This is not the unbiased science we expect, nor does it provide the impartial justice we require.

Here is a cautionary tale of bias in criminal justice, of suspect-centric criminal investigation, prosecution and forensic science, and of how modern DNA software can be manipulated to falsely implicate an innocent man.