The Limits/Entry 2.06/One technique, and what it can and cannot carry
Contextual bias
Giving examiners the same mixture and different background information changes the conclusions they reach.
Contextual Bias
When the story changes the science
The numbers on an electropherogram do not interpret themselves. A human examiner weighs them — and what that examiner has been told beforehand can shift the conclusion.
The effect has been demonstrated directly. In a study by Itiel Dror and Greg Hampikian published in Science & Justice in 2011, seventeen DNA examiners were given the same low-template mixture profile from an actual case and asked whether a reference sample could be excluded. Given no case context, most said exclusion was reasonable. The original case analysts, who had been given context suggesting the contributor was a suspect, had reached the opposite conclusion. Same data, different answers — driven by background information, not by anything in the peaks.
This is not a flaw unique to forensic science. It is a well-documented property of human judgment, studied under the term "contextual bias" or, more broadly, cognitive bias. The concern in forensic DNA is that it operates in a domain where practitioners and courts often treat the output as objective measurement. When the profile is clean and the template abundant, interpretation is largely constrained by the numbers. When the profile is a mixture or a low-template result, the examiner's discretion widens — and so does the window for bias to enter.
The structural solution is linear sequential unmasking: examiners work through the evidence in a defined order, encountering only the information they need at each stage, and case-narrative details arrive last or not at all. Several laboratories and standards bodies, including those working to SWGDAM guidelines, have discussed procedural safeguards along these lines, though implementation varies.
Probabilistic genotyping software narrows some of the interpretive space by fixing thresholds and models before the analyst sees the profile. But software parameters themselves reflect prior choices, and who sets them — and when, relative to case information — matters too. The bias problem does not disappear with automation; it relocates.
