Holy grail of mixture analysis? or How to frame a suspect?

Notes for How to frame a suspect? slides.

A simple way to calculate all mixtures

Comment – The case in which I participated is used only as an example.
The talk isn’t about that case, not about what happened or whether the verdict was correct. (I don’t know).
Prolific Florida DNA consultant Dr. Kevin McElfresh employs as LR a number which he calculates in very simple way.
If also correct, the method would achieve the “simple yet valid” holy grail of mixture analysis.
It’s obviously, and dangerously, wrong.

Outline

The 2009 NRC report praised the practice of forensic DNA, especially in comparison with the other forensic science modalities.
Experts in fingerprint, hair, or even bite mark would testify that a similarity constituted very strong, even certain, evidence of identity.
Eventually it became clear that in effect the expert culture was to quantify strength of evidence intuitively based on personal experience.

Accumulation of examples and studies showed that expert intuition exaggerates, errs, and works by confirmation bias.

DNA escaped the same bad reputation because it quantifies. DNA measures strength-of-evidence with the likelihood ratio (LR), a number with a real, appropriate, and scientific meaning.

Reporting that number distinguishes DNA evidence from malarkey.

“Informative” loci

Mixture evidence from the case

McElfresh defines “informative” as “all alleles expected per prosecution hypothesis (i.e. suspect and victim) are present”
Perhaps the thinking is that they’re informative of what you think you already know. Confirmation bias.
“Informative” panels are analogous to witnesses favorable to the prosecution; tinted panels to alibi witnesses.

I disagree with his testimony that the tinted panel evidence, if calculated, would be further prosecution-favorable evidence.
If that were right, DNA could never exculpate anybody!

Close-up — total dropout

Another locus

Evidence changes your opinion

Fingernail DNA evidence

Suspect=Brenner

The method will find positive evidence – LR>1 – for virtually anyone.
The murder was 30 years ago. Where was I at the time?

I don’t recall

Conclusion

Honest: It’s tempting, as an experienced DNA hand looking at mixture data and comparing it with a suspect profile, to make the judgement that the suspect contributed.
Stop there and you’re like the old-time hair-matching etc. experts — honest but dangerously unaware of your limitations.

Not honest: Add a patina of “science” with a calculation that lacks logical basis, thus disguising the subjectivity as objective.

Active market: Among 1000+ cases of DNA assessment by eyeball guesswork adorned by a meaningless number, I expect multiple convictions of factually innocent suspects.