An illustration of Sherlock Holmes and Doctor Watson from "The Adventure of Silver Blaze" in The Strand Magazine in 1892. (Image: Courtesy of Wikimedia Commons)

“The absence of evidence is not the evidence of absence.”

It’s a phrase that gained popular currency in the 21st century, especially since the bungled justification for the U.S.-led invasion of Iraq in 2003. Not seeing something doesn’t mean it’s not there, according to this line of thought—since there are known unknowns, and unknown unknowns, as former U.S. Secretary of Defense Donald Rumsfeld once put it.

But there can be rational conclusions about something not existing if it isn’t there, two academics write in the latest issue of Forensic Science International.

Using some Bayesian statistical theory—and an anecdote from a Sherlock Holmes story—the two University of California–Irvine professors find that there are instances where the non-existence of evidence may be reasoned out from a crime scene.

“As a general matter, negative evidence has inferential value for supporting a particular hypothesis over an alternative hypothesis to the extent a negative result is more likely under the favored hypothesis than under the alternative hypothesis,” write the authors. “This means that forensic scientists will need to think about a probability of a negative finding under the relevant hypotheses in order to assess the value of negative evidence.”

Essentially, more statistics about the probability of finding evidence need to factor into the likelihoods at play in examining a crime scene, write William C. Thompson, professor emeritus of criminology, and Nicholas Scurich, an associate professor at the school.

For instance, a toxicologist looking at metabolites in the blood of a person would have an easy forensic determination if the probability of detection was binary—either 0 percent of 100 percent.

But if the tox detection possibilities drop to 50 percent, because of elapsed time or false negatives or other factors, then the conclusions cannot be considered simply black-or-white, the criminology professors add.

In that case, more research into understanding how metabolites can evade toxicology detection based on quantifiable factors would be key, they opine. (Bayesian statistical theory, at its simplest essence, is making informed judgments based on a changing set of information and circumstances, according to experts.)

The Sherlock Holmes example is from the story “The Adventure of Silver Blaze,” a tale of the theft of a champion race horse, and the apparent murder of its trainer—and the famous “curious incident of the dog in the night-time.”

The curious incident, as it turns out, is that there was no incident—and the dog did not bark at an apparent intruder. That indicated the dog probably knew this intruder after all—although the dog could have been drugged, or been asleep, or wandered off. 

The UC–Irvine authors conclude that though the fictional Holmes realized his theory of the dog knowing the interloper wasn’t 100 percent assured, it still fit the circumstances. Holmes further concludes that his Scotland Yard counterpart is an extremely competent investigator—but who simply lacks the imagination to solve the crime.

“To draw scientifically sound conclusions (forensic scientists) will need empirical data on the probability of detecting traces of various kinds under various circumstances,” add Thompson and Scurich.