The Problems with the Probabilities of STRmix

By Rachael M. Sedlacek posted 01-12-2020 20:05


STRmix sounds like it belongs in a sci-fi epic. But it is increasingly important when the DNA evidence at issue could have multiple sources, such as in criminal sexual conduct cases. STRmix boasts many uses, although Mary Chartier and David Champine boiled it down for us this way in our Criminal Sexual Conduct Cases on-demand seminar:

STRmix provides a likelihood ratio of what is contained in a sample. In order to get to that likelihood ratio, trained analysts have to input hypotheses into the software. For example, one hypothesis could be that the mixture is attributed to Person A and an unrelated, unknown person. The second hypothesis could be that the mixture is attributed to two unrelated, unknown persons. STRmix then provides a likelihood ratio of which hypothesis is more accurate.

The likelihood ratio can look something like the following: it would be 42 quadrillion times more likely that the mixture originated from Person A and an unrelated, unknown person than if it originated from two unrelated, unknown persons.

STRmix essentially nets probabilities, so it’s a crapshoot as to how it will play out in a given case. It may be entirely uninformative, or the results might be quite persuasive. Imagine a juror hearing that the DNA mixture was 42 quadrillion times more likely to contain DNA from your client and an unrelated person than from two unrelated people.

What’s more, the Michigan Court of Appeals has given STRmix a blessing of sorts. In People v Muhammad, a 2018 published case, the court held that the STRmix is admissible under MRE 702 and the Daubert standard. (The Michigan Supreme Court later denied leave.) The court of appeals stuck to this view in a 2019 unpublished decision.

Despite these obstacles, Mary contends that STRmix results are ripe for challenge. She notes that while we think of DNA evidence as “concrete,” STRmix requires the analyst to make a number of interpretive choices, such as the following:

  • the number of contributors in the DNA sample
  • the analytical thresholds
  • whether something is drop in or drop out
  • whether the sample is stutter or saturation

These decisions all affect the results. To challenge the analyst’s methods, Mary recommends using FOIA to obtain the lab notes and reports and also to look at the communication between law enforcement and the lab. If the analyst was told that the defendant’s DNA was likely in the sample, it may bias the testing.

And Mary isn’t alone in her critiques of STRmix. A federal judge in the Western District of Michigan recently issued a 49-page opinion excluding STRmix results and critiquing its scientific reliability. See US v Gissantaner, No 1:17cr-130 (WD Mich Oct 16, 2019).

For more suggestions on how to challenge STRmix evidence, check out segment 4 of the CSC seminar. You may also find helpful information by searching STRmix at the National Institute of Standards and Technology. Mary noted during our annual Criminal Advisory Board meeting that there is a growing need for information in this area, so stay tuned for more resources!