Beyond the STRmix Number: What Makes STRmix Testimony Reliable?
Written by Virginia Barron | Criminal Defense Attorney & DNA Forensic Litigator
Lessons from a forensic science workshop on validation, likelihood ratios, limitations, and testimony in the courtroom
Good forensic testimony is not testimony that cannot be challenged. It is testimony that remains accurate when it is.
I recently had the opportunity to co-present a workshop on STRmix testimony at the Midwestern Association of Forensic Scientists. I presented alongside Anne Ciecko of the Midwest Regional Forensic Laboratory, Christine Clouser, PhD, of the Hennepin County Sheriff’s Office Forensic Science Laboratory, Amber Rasmussen of the Wisconsin State Crime Laboratory, and Kelly Ronquist of the Eau Claire County District Attorney’s Office. The audience was largely forensic laboratory analysts.
What made the workshop valuable was not that everyone approached STRmix from the same perspective. We did not. Laboratory scientists, a prosecutor, and a defense attorney necessarily encounter the evidence in different roles. But the workshop kept returning to the same basic question: what does it take for complicated DNA testimony to be scientifically sound, accurately communicated, and able to withstand meaningful scrutiny?
The answer begins long before an analyst takes the witness stand. It begins with validation, training, documentation, transparency, thoughtful laboratory policy, and an honest understanding of what the DNA result can—and cannot—say.
For readers who want a basic explanation of the software first, see my article What Is STRmix and How Is It Used in Minnesota Courts? This article asks a different question: once STRmix is being used, what makes the result and the testimony about it worthy of confidence?
Reliable STRmix Testimony Starts With the Science Before Court
One of the central themes of our workshop was simple: successful testimony starts with good science. In the presentation, that meant four things working together: validation, continuous feedback and improvement, training, and documentation and transparency.
Those are not merely administrative requirements. They determine how well a laboratory understands the method it is using, whether its policies reflect what its own validation actually showed, whether analysts recognize unusual or difficult results, and whether another scientist—or an attorney—can reconstruct how a case interpretation was reached.
The 2024 NIST report on human factors in forensic DNA interpretation makes a similar point. It uses a systems approach to examine how decisions, procedures, training, documentation, communication, and organizational practices can affect DNA interpretation and testimony. NIST’s report includes recommendations aimed at improving the production, interpretation, evaluation, documentation, and communication of DNA results.
The best way to think about this is that validation should not merely answer, “Can our laboratory run this software?” It should help the laboratory understand where the method performs well, where it becomes more difficult, what choices matter, and what warning signs require additional review.
Validation Should Build Understanding, Not Just Check a Box
During the workshop, we distinguished between validating for compliance and validating for understanding. A laboratory obviously must satisfy the applicable standards. But a technically compliant validation can still leave important questions unanswered if the laboratory has not pushed the method into the difficult situations it will encounter in real casework.
For example:
How accurately can analysts determine the number of contributors in difficult mixtures?
What happens when the assumed number of contributors is wrong?
What did the validation show about low-level or degraded samples?
What diagnostics matter when the STRmix output does not fit what the analyst can see in the electropherogram?
How are relatives or substantial allele sharing handled?
What happens when a laboratory changes a threshold, reworks a locus, or adopts a policy that differs from another laboratory?
What did court challenges, published research, proficiency experience, or casework reveal after the original validation was completed?
This is why validation should not be treated as something that ends on the day the laboratory director signs the approval page. Real casework creates new questions. Literature changes. Analysts encounter unusual mixtures. Attorneys expose ambiguities that may not have been obvious in the laboratory. A good quality system learns from those events rather than treating validation as a one-time hurdle.
That idea also appears in the current SWGDAM framework. SWGDAM now publishes specific guidance for probabilistic genotyping and likelihood-ratio reporting, in addition to broader STR interpretation and validation guidance.
STRmix Does Not Remove Human Decision-Making
STRmix is sophisticated software. It uses biological modeling and mathematical processes, including Markov Chain Monte Carlo (MCMC), to explore possible genotype combinations that can explain an observed DNA mixture. But the software does not receive a case file and independently decide every question from beginning to end.
Human decisions still matter. Before and around the software analysis, analysts make decisions about the profile, including the number of contributors, whether the sample is suitable for interpretation, conditioning, the comparison propositions, and whether the resulting interpretation and diagnostics are acceptable under the laboratory’s validated procedures.
The fact that STRmix uses an MCMC process also means repeated runs can produce slightly different point estimates. That is not, by itself, evidence that the software is malfunctioning. It is a known feature of Monte Carlo sampling, and STRmix has published technical material addressing Monte Carlo variation.
The important courtroom point is not that human judgment or model variation automatically makes a result unreliable. It is that the result should be explained for what it actually is: the output of a validated scientific process that combines laboratory data, modeling, defined propositions, and human decisions.
A Likelihood Ratio Is Not the Probability That Someone Is the Contributor
This was one of the most important testimony issues we worked through with the analysts.
A likelihood ratio is a comparison between two propositions. At the source level, a common STRmix comparison asks how probable the observed DNA evidence is if the person of interest contributed DNA, compared with how probable the same evidence is if the DNA came from one or more alternative contributors described in the competing proposition.
That is not the same thing as the probability that the person is the contributor. It is also not the probability that the person committed the crime. Turning an LR into either of those statements transposes the conditional and can dramatically overstate what the statistic means.
Incorrect shortcut: “It is X times more likely that the person is a contributor.”
Better formulation: “The observed DNA results are X times more probable if Proposition 1 is true than if Proposition 2 is true.”
The wording can sound technical, but the distinction is foundational. The LR describes the relative probability of the evidence under two propositions. It does not directly assign a probability to the propositions themselves.
The Snow Does Not Change. The Hypotheses Do.
One courtroom example I like to use is much simpler than DNA: snow on the ground.
Imagine the observed evidence is snow. One proposition is that you are in Minnesota. The competing proposition is that you are on a tropical beach in Hawaii. Snow would generally fit the Minnesota proposition much better. Now keep the exact same evidence—snow on the ground—but change the competing proposition from Hawaii to Alaska. Suddenly the same observation may do very little to distinguish between the two propositions.
The evidence did not change. The propositions did.
That is the important lesson for DNA. A likelihood ratio is not a permanent number attached to a DNA profile. It depends on the propositions being compared, the contributor assumptions, the population assumptions, conditioning decisions, and other parts of the model. If the propositions change, the LR can change as well.
This matters particularly when a realistic alternative contributor is a relative. My article DNA Mixtures and Relatives: How a Family Member Can Affect STRmix Results discusses why an “unknown unrelated person” proposition may answer a different question from the one actually presented by the facts of a case.
Source-Level DNA Evidence Does Not Answer the Activity-Level Question
Another point we emphasized to analysts was the difference between source-level and activity-level conclusions.
A source-level comparison may provide statistical support for a proposition about whose DNA may be represented in a profile. That does not tell the jury how the DNA arrived on the item, when it arrived, why it is there, whether the person touched the item, or whether the DNA was deposited during the charged offense.
That distinction has become increasingly important as DNA testing has become more sensitive. The ability to detect smaller quantities of DNA increases the need to distinguish identification questions from transfer, persistence, prevalence, and activity questions.
The 2024 NIST human-factors report addresses the risks that arise when DNA evidence is communicated beyond the limits supported by the science.
For a defense attorney, this is often where the courtroom meaning of DNA becomes most important. A mathematically strong source-level LR may still leave the central case question unanswered: what does the presence of that DNA actually prove about what happened?
Reliable Testimony Includes the Limits
Some of the most useful parts of the workshop were not mathematical at all. They were about how a scientist communicates honestly under pressure.
Answer the scientific question the same way regardless of which attorney asks it.
Be willing to speak with the defense attorney as well as the prosecutor.
Say “I do not know” when you do not know.
Say “I do not remember” when you once knew the answer but cannot recall it on the stand.
Admit what the testing can and cannot establish.
Know the case notes, electropherograms, interpretation records, and report before testifying.
If you make a mistake, correct it promptly and fully rather than defending the mistake.
None of those practices weakens a forensic witness. They strengthen the testimony because they make clear where the science ends and where uncertainty begins.
A witness who acknowledges a limitation before opposing counsel forces the admission does more to build confidence than a witness who treats every question as an attack on the laboratory.
What a Defense Attorney Should Be Looking For
When STRmix evidence is used in a Minnesota criminal case, the defense should not limit review to the final report and the final likelihood ratio. Depending on the case, meaningful review may require the underlying records that explain how the result was produced.
The electropherograms and underlying analytical data.
The analyst’s number-of-contributors determination and the basis for it.
Conditioning decisions and the exact propositions used in the LR.
STRmix deconvolution outputs, diagnostics, and any re-runs or reworks.
The laboratory’s internal validation and the portions relevant to the type of mixture at issue.
The SOPs and interpretation guidance in effect when the case was analyzed.
Any issues involving relatives, low DNA input, degradation, allele sharing, or unusual diagnostics.
Technical-review changes and the reasons for them.
The goal is not to find something to criticize simply because software was used. The goal is to determine whether the scientific process applied in the case supports the opinion being presented to the jury.
In Minnesota, expert testimony must satisfy Minnesota Rule of Evidence 702. The meaningful reliability question in a STRmix case may be less about whether probabilistic genotyping exists as an accepted method and more about whether the particular interpretation, assumptions, and testimony have a reliable foundation in the case at hand.
Challenge Should Build Confidence, Not Destroy It
One of the most useful things about presenting this workshop to laboratory analysts was that the discussion did not assume that defense scrutiny and good forensic science are opposing goals.
They are not.
A laboratory should want to know whether its validation actually addresses difficult casework. An analyst should want to recognize when a result falls outside the patterns normally observed. A prosecutor should want testimony that accurately communicates the evidence. And a defense attorney should be able to test the assumptions, methods, and limitations without the act of questioning being treated as hostility toward science.
Confidence in forensic evidence should come from understanding why the result is reliable—not from avoiding difficult questions.
Good forensic testimony is not testimony that cannot be challenged. It is testimony that remains accurate when it is.
What This Means for a Minnesota DNA Case
For someone accused of a crime, a STRmix number can look overwhelming. It may be reported in the millions, billions, trillions, or beyond. But the size of the number does not eliminate the need to understand how the number was generated and what question it actually answers.
A careful forensic review separates several different questions: Was the underlying DNA testing sound? Was the mixture interpreted within the laboratory’s validated limits? Were the contributor assumptions reasonable? Were the propositions appropriate to the facts? Was the LR communicated correctly? And did the testimony stop at the level the science actually supports?
Those questions do not diminish good science. They are part of what makes good science defensible in court.
If you or a loved one are facing a charge involving DNA, or you are a defense attorney seeking consultation on what the DNA evidence in your case can and can not establish, call Ginny at 507-822-5735 to discuss your case.
Frequently Asked Questions About STRmix Testimony
-
STRmix is probabilistic genotyping software used to interpret complex DNA mixtures and calculate likelihood ratios comparing specified propositions about the evidence. It uses biological modeling and mathematical sampling methods, including MCMC, rather than simply declaring that a person “matches” a mixture.
-
No. STRmix evaluates DNA evidence under competing propositions and can provide statistical support for one proposition over another. A likelihood ratio is not a direct probability that a particular person is the contributor, and it does not establish that the person committed the crime.
-
A likelihood ratio compares how probable the observed DNA evidence is under one proposition with how probable the same evidence is under a competing proposition. An LR above one supports the numerator proposition relative to the denominator proposition; an LR below one supports the denominator proposition. The LR must be read together with the propositions that produced it.
-
Yes. The LR is conditional on the propositions and assumptions used in the calculation. If the propositions change—for example, if a realistic relative is considered instead of an unrelated unknown—the resulting LR may also change.
-
No. A source-level STRmix result does not by itself establish how DNA reached an item, when it was deposited, whether the person directly touched the item, or whether the DNA was deposited during the alleged crime. Those are different scientific and factual questions.
-
STRmix is asked to model a mixture using an assumed number of contributor positions. If the number-of-contributors determination is wrong or uncertain, the model may be solving a different problem from the one actually represented by the sample. That is why the basis for the NoC decision is an important part of forensic review.
-
Yes. Because STRmix uses MCMC sampling, repeated runs can produce some Monte Carlo variation. The existence of variation does not automatically mean the result is unreliable, but the laboratory should understand and evaluate that variation within its validated procedures.
-
Meaningful review includes requesting the entire underlying DNA lab file as well as the raw data. It includes the validation studies and the SOPs that were relied on in the DNA testing process and the procedures used to interpret that data. It includes the electropherograms, analyst notes, contributor-number assessment, STRmix outputs and diagnostics, propositions, conditioning decisions, technical review records, and applicable SOPs.
-
Independent review may be useful when the case involves a complex or low-level mixture, uncertainty about contributor number, close relatives, unusual diagnostics, conditioning, changing interpretations, very large likelihood ratios, or testimony that appears to go beyond what the DNA result itself establishes.
Primary Sources and Further Reading
Barron Law Office – What Is STRmix and How Is It Used in Minnesota Courts?
Barron Law Office – DNA Mixtures and Relatives: How a Family Member Can Affect STRmix Results
This article provides general information about forensic DNA evidence and Minnesota criminal cases. It is not legal advice for any particular case. DNA interpretation is case-specific, and the significance of a STRmix result depends on the sample, laboratory procedures, propositions, assumptions, and the other evidence in the case.