In 2015, a team of statisticians set out to help rehabilitate forensic science, a field with a reputation for flimsy methods and dubious conclusions. The Center for Statistics and Applications in Forensic Evidence (CSAFE), based at Iowa State University, has grown to include more than 60 researchers at six universities working on new ways to analyze fingerprints, shoeprints, bullet marks, and other crime scene evidence. Now, CSAFE’s primary funder, the National Institute of Standards and Technology, has committed $20 million in new funding to the effort over the next 5 years. The center’s director, Iowa State statistician Alicia Carriquiry, told Science Insider about her vision for its second phase. The interview has been edited for clarity and brevity.

Q: What’s changed since CSAFE launched?

A: In 2015, there was nothing of this sort. There was a community of practitioners in forensic science that was quite separate from the community of academic researchers, with a few exceptions … I think the community of practitioners has come to the understanding that, first of all, we ’ re not here to tell them they ’ re doing everything wrong. We ’ re here to see whether we can help them do their job better, and develop tools for them to use.

Q: What advances have made you proudest?

A: One of the big problems in the criminal justice system is the lack of transparency. When somebody says, ‘This fingerprint matches this fingerprint we found [in an automated fingerprint identification system],” what do you mean, it matches? What algorithm did you use? What did you measure? How precise is it? … None of that is available. And this happens everywhere—with mixture DNA software, with ballistics software. You don ’ t know what ’ s going on inside the black box. We ’ re trying to really, really hammer on the notion that that ’ s not acceptable. If you’re going to put somebody in jail for something, that person should have the right to inspect the methods you used. … One of the things that I’m very proud of is that we ’ ve insisted in the last 5 years on the need of transparency and open-source algorithms, and open source data.

We ’ ve built enormous databases that are carefully constructed, very well documented, and searchable—and we ’ ve put all these data in the public domain. For example, one of the databases we created has 250,000 images of the bottoms of shoes. And for each one, it ’ s well-documented: “This is Nike, model blah-blah-blah, size blah-blah-blah. The same sole comes in all these colors.” That ’ s something that didn ’ t exist.

We have some pretty good algorithms to look at the question of source—do these two pieces of evidence have a common source? Of course, until all of that stuff is tested in real, working labs—until we can pilot this at a bigger scale—this is all proof-of-concept. One of the things that CSFAE 2.0 is going to be doing is trying to start some serious piloting.

Q: Is there a kind of evidence that has proved harder than you thought it would be to address with statistics?

A: Getting the data you need to develop methods for analyzing fingerprints has been very challenging. Getting an [institutional review board] approval to collect the fingerprints of a ton of people is very, very, very, difficult. The existing databases, like the FBI’s, are not accessible. Hopefully if we manage to partner with [forensic] labs, we might be able to access some of the data they hold.

Q: What else do you expect CSAFE 2.0 to do?

A: We have a training and education program, and I would really like to do this more systematically—develop the training materials for … forensic science students at the undergraduate and the graduate level. In many programs, these students don ’ t get any quantitative training. In fact, many of the forensic science programs are at small 4-year colleges that might not even have a department of statistics or a statistician inside a department of mathematics. Developing good courses that can be delivered at a distance for these students so they come out with a better degree is something that I would really like to do.

Q: Do you think there be a point where forensic scientists are using statistics appropriately, and you can just leave it to them?

A: We can always dream! At the very least, you want forensic scientists to be cognizant of issues such as uncertainty, the need to measure variability and take [it] into account when you make a conclusion, and the notion of probability. These are concepts that are not really familiar to many forensic practitioners. If we get to the point where you may not understand how to do statistics, but you understand why statistics is important, I think that would be a good place to get to.