The Database Knows Your Cousin: Can Genetic Genealogy Do What CODIS Could Not?

The claim circulates in policy discussions and law enforcement advocacy alike: putting someone’s DNA in a database makes them less likely to reoffend. There is actual research behind it. The question is what that research says, what it does not, and whether any of it applies to a forensic tool that does not work the way CODIS does.

Investigative genetic genealogy has solved high-profile cold cases and identified serial offenders who evaded law enforcement for decades. What it has not done is generate a body of deterrence research. That absence is worth examining, because the deterrence argument is increasingly being made on its behalf, often by analogy, without much scrutiny of whether the analogy holds.

What the Research Actually Found

The primary empirical work on DNA databases and recidivism comes from economist Jennifer L. Doleac, now at Texas A&M University. Her 2017 study, published in the American Economic Journal: Applied Economics, used a natural experiment design. State legislatures expanded their CODIS databases at different times and for different offense categories. Offenders released just before an expansion date were not added to the database. Offenders released just after were. Because nothing else changed about those individuals at that moment, the subsequent difference in reconviction rates is attributable to the database itself.

Using criminal history data from seven states, Doleac found that violent felons added to a DNA database were 17% less likely to be reconvicted within five years than comparable offenders who were not added. Property felons showed a smaller effect, around 6%, and the result was less statistically robust. Avinash Bhati’s 2010 study using Florida data found 2 to 3% reductions for robbery and burglary specifically, with increases in recidivism risk for other categories, suggesting the deterrence effect is not uniform across offense types.

The Denmark follow-up, co-authored by Doleac with Anne Sofie Tegner Anker and Rasmus Landerso and published in the American Economic Journal in 2021, found substantially larger effects. Offenders added to Denmark’s expanded database were up to 42% less likely to reoffend in the following year. Those offenders were also more likely to find stable employment and maintain family connections in the period following registration, a finding consistent with the hypothesis that keeping someone out of the criminal legal system in the short term can change their trajectory.

There is an additional wrinkle worth tracking. Doleac herself acknowledged that part of the observed deterrence effect is likely attributable to what criminologists call the CSI effect: an inflated perception of forensic capability driven by media coverage. If offenders believe DNA evidence is more reliably collected and matched than it actually is in practice, that perception raises their subjective estimate of detection risk. The deterrence is real. Some of the mechanism driving it is perception, not just capacity.

The Structural Difference That Changes the Argument

CODIS works on a direct-match logic. An offender’s DNA profile is entered into the database under legal compulsion following conviction or, in many states, arrest. If they reoffend and leave biological material at a scene, their profile is already there to be compared. The offender is in the system. They can be located by their own record.

Investigative genetic genealogy works on an entirely different principle. The offender is not in any law enforcement database. Law enforcement uploads crime scene DNA to a consumer genealogy platform, such as GEDmatch or FamilyTreeDNA, the only two major consumer databases that currently permit law enforcement use. The platform returns partial matches to the offender’s genetic relatives: people who submitted their DNA voluntarily, typically to research their ancestry or find biological family members. Investigators then build a family tree backward from those relatives and work forward to identify the source of the crime scene sample.

As of late 2023, IGG had been credited with clearing more than 651 criminal cases and identifying 318 individual perpetrators in the United States. GEDmatch holds approximately 1.2 to 1.5 million profiles. Research has found that a database of that scale can identify a third cousin or closer for more than 90% of the population of European descent, which is the demographic most represented in consumer genealogy databases at this point. The identification reach is substantial. It is also, from the perspective of any individual offender, largely invisible and not within their control.

The Question No One Has Studied

No peer-reviewed study has measured whether awareness of investigative genetic genealogy changes offender behavior. That is not an oversight. It is a structural problem.

The deterrence research on CODIS works because there is a defined population to study: offenders who were added to the database, compared against offenders who were not. You can track their subsequent reconviction rates and attribute differences to the database. IGG does not create that kind of study population. No one is “added” to an IGG database. Potential suspects are identified retrospectively, through relatives who are already there. There is no enrollment event to use as a natural experiment cutoff.

What criminology does offer is a framework for thinking about when deterrence works. The literature is fairly consistent: deterrence is most effective when the perceived probability of getting caught is high, when that perception is accurate, and when the potential offender is engaging in rational cost-benefit calculation at the time of the offense. All three conditions are debatable for IGG. Offender awareness of IGG’s capability is uneven. The actual identification rate from consumer genealogy databases remains limited to cases where a biological sample was recovered and processed. And many violent offenses, particularly those committed impulsively or under the influence of substances or situational stress, do not involve the kind of deliberate calculation that deterrence theory assumes.

The CSI Effect, Reloaded

Here is where the analysis gets more interesting, and where the honest answer has to be: we do not know yet, but the mechanism for a very large effect does exist.

The CSI effect is well-documented in the criminology literature. The television franchise convinced large portions of the American public that forensic science could solve virtually any crime reliably and quickly. That perception was, and remains, significantly inflated relative to actual forensic laboratory capacity, processing timelines, and case clearance rates. But the perception changed behavior. Jurors expected DNA evidence in cases where it was irrelevant or unavailable. And, to the extent criminologists can measure it, offenders adjusted their behavior based on what they believed forensic science could do, not on what it actually could.

The Golden State Killer case functions as an IGG equivalent of that cultural moment. Joseph DeAngelo committed a documented minimum of 13 murders and 45 rapes across California between 1974 and 1986. He evaded law enforcement for more than four decades. A DNA sample from a 1980 crime scene was uploaded to GEDmatch in 2017. Investigators built a family tree from partial matches to distant relatives and identified DeAngelo, then 72, as the source. He was arrested in April 2018 and pleaded guilty to multiple counts of murder and kidnapping.

That case received extensive and sustained media coverage. It established a public association between consumer ancestry testing and the identification of violent offenders who believed themselves permanently safe. Whether it changed offender behavior is not something that has been measured. But the mechanism by which it could is the same mechanism Doleac identified as partially driving the CODIS deterrence effect: perceived detection probability, shaped by what people believe forensic tools can do.

The Minority Report Problem

The film Minority Report, and the Philip K. Dick short story it adapted, made a specific argument about pre-crime logic: that a system designed to identify offenders before they act carries within it the seeds of its own institutional corruption. The PreCrime division did not just predict crime. It created a carceral apparatus built on prediction, which meant it required the prediction to be treated as equivalent to the act itself.

IGG is not a pre-crime system. It identifies suspects after a crime has occurred, using biological evidence left at the scene. But the deterrence argument, if it scales, starts to push toward that territory in a specific way. The argument is: people should change their behavior now, before any offense, because of what a forensic tool could do to them later. The tool’s power comes not from catching everyone but from making everyone potentially reachable. The reach extends through relationships the offender did not create and cannot control. A person who has never submitted their DNA to any database is nonetheless potentially identifiable through their sibling, their parent, their second cousin once removed, if any of those people chose to learn about their heritage online.

That is either a very efficient crime deterrent or a very expansive surveillance architecture depending on which end of the analysis you approach it from. Law enforcement frames it as the former. Civil liberties scholars, privacy researchers, and legal academics studying IGG’s constitutional implications have raised sustained concerns about the latter. Those concerns are not unfounded. In 2019, a Florida circuit court approved a warrant overriding GEDmatch’s own privacy settings to enable a law enforcement search, a development that demonstrated the platform’s governance model was not as user-controlled as its terms of service implied.

What Would Actually Change the Landscape

The CODIS deterrence research succeeded in producing measurable findings because it had a clean natural experiment: the database expansion date. People on one side of that line were enrolled. People on the other side were not. Everything else was held roughly constant. The subsequent behavioral difference was attributable to enrollment.

IGG does not offer that structure. But several things could change the analytical landscape over the next decade.

First, consumer database participation rates continue to expand. Estimates suggest that databases of the scale currently maintained by GEDmatch and FamilyTreeDNA can already identify a genetic relative for the large majority of the U.S. population of European ancestry. As participation expands across other demographic groups, the effective identification reach of IGG widens, and with it, the plausibility of the deterrence argument for a broader population of potential offenders.

Second, the policy environment is shifting. More than a dozen states have moved toward codifying IGG use by law enforcement, including requirements about when it can be used, under what evidentiary thresholds, and with what oversight. As those frameworks solidify, the public awareness of IGG as a routine law enforcement tool is likely to grow. Awareness is the precondition for any deterrence effect based on perceived detection risk.

Third, the research methodology question is solvable, though not easily. Survey-based studies could measure offender perception of IGG’s identification capability and track whether that perception correlates with self-reported behavior or officially measured reoffending. Researchers studying high-profile IGG cases could attempt to measure changes in crime patterns in the jurisdictions and time periods following those cases’ media coverage, using the Golden State Killer arrest as an event study. None of that work has been published as of this writing. It should be.

The Honest Accounting

The research on CODIS-style databases is real and the effect sizes are not trivial, particularly in the Denmark study. The mechanism, raising perceived detection probability, is theoretically sound and empirically supported in the specific context where it has been studied. Whether that mechanism transfers to investigative genetic genealogy, a tool that works through a fundamentally different architecture, is an open question.

The case for a large future deterrence effect from IGG is not frivolous. The CSI effect demonstrates that forensic tool awareness shapes offender behavior through perception. The Golden State Killer case created exactly the kind of high-salience, widely covered forensic event that has historically shifted those perceptions. And if the identification logic generalizes, the subjective risk it imposes on potential offenders is unlike anything CODIS created, because it reaches people through relationships they did not choose and cannot sever.

That is the Minority Report version of the argument: not that the system predicts crime, but that the system’s reach is effectively total, because the database that matters is not the one law enforcement maintains. It is the one that exists in the genetic connections between every person who shares DNA with someone who was curious about their ancestry.

What the research does not yet tell us is whether any of that actually changes behavior. The honest answer is that the question has not been studied. That is not a reason to dismiss the theoretical case. It is a reason to demand the empirical work before the theoretical case is treated as settled.