And in 2016, a government-appointed advisory group of leading scientists and engineers released a study that cited serious concerns about the underlying science of ballistic analysis. The study said the practice relies on circular logic and its “conclusions are subjective.”
“It’s basically a guess,” said Jon Loevy, an attorney representing Pursley’s case in civil court. “I’m sure there’s a lot of ballistics examiners that swear by the science, but that doesn’t make it scientific.”
Specifically, Loevy questioned the subjective nature of the analysis.
“Why do we believe science? It’s not just because it’s a guy who says ‘I’m a scientist’ giving it — it’s because you can replicate results with studies,” Loevy said. “Well, there’s no study that corroborates that this shell casing looks like that shell casing — it’s just some guy’s opinion.”
I know a fair amount about biometrics and it appears some of the same issues which trouble the use of fingerprints are also issues with forensic ballistic analysis. Verification is much better than identification. That is, the examiner can say that the odds that a given bullet (partial fingerprint) matches a given gun (person) with 99% certainty. But given a bullet (partial fingerprint) then finding the correct gun (person) out of the entire population of guns (people) the certainty might be only 0.1%.
Think of it this way: Suppose each gun stamped a number from 1 to 100 on each bullet fired. Given a random bullet and a random gun you could reject 99 guns out of 100 as not shooting the bullet. But if there are 100,000 guns, and the numbers are distributed evenly, 1000 of them could have shot the bullet.
The markings left on the bullets (and shell casings) don’t have sufficient persistent “resolution” to uniquely distinguish between all guns when the population of guns is very large.