Thread by @radleybalko

Based on the comments at the WP to my piece yesterday, I think an explanatory thread is in order.

The study I wrote about was designed to see if medical examiners' manner of death determination can be influenced by cognitive bias.
washingtonpost.com/opinions/2021/…

This is important because we tend to think of an ME's opinion as grounded in science than, much more so than, say, a tire tread or fiber expert. But determining "manner" of death is typically much more subjective "cause" of death. This, despite ...

... the fact that a manner of death determination is profoundly more consequential. If an ME determines a death was an accident, we have a tragedy. If an ME determines a death was a homicide, there's a very good chance someone is going to prison for a long time. As I point ...

... out in the article, just *being told* that an ME determined a death to be a homicide has persuaded innocent people to plead guilty. This has been especially true in Shaken Baby Syndrome cases.

wrongfulconvictionsblog.org/2015/07/24/why…

In this study, researchers gave 133 medal examiners a fact pattern in which the medical information about a child's death was identical for all participants. But half were given non-medical information commonly associated with child deaths that are homicides -- the child ...

... was black, and the caretaker at the time of death was the mother's boyfriend. The other group was given non-medical information less associated with guilt -- a white child, left in a grandmother's care. These facts were based on the sorts of cases often presented to MEs.

If the purpose of this study was to see if medical examiners are racist, the critics are correct: this would be a poorly designed study. There was no attempt to distinguish between the variables (race of child; age, gender of implied race of caretaker).

But that was *not* the purpose of the study. The purpose of the study was to see if MEs can be influenced by non-medical information commonly associated with abuse, but that should be irrelevant to a scientific analysis.

Race *is* one of those factors, and that matters. But the study wasn't about measuring racism. It was about measuring bias. In fact, the three real-world examples I gave in my thread tweeting out my piece were all white ...

https://twitter.com/radleybalko/status/1363169625215799303

... men accused of murdering their girlfriend's child after dubious forensic testimony.

The study's findings are important. First, it shows manner of death determinations are far more subjective than commonly thought. Often, once an ME says homicide, police and prosecutors ...

... assume guilt, and go about finding corroborating evidence against a caretaker.

Second, the findings also raise questions about data we have on child deaths that are accidental vs. homicides -- including race-based data. The study suggests some of that data may be suspect ...

... because they're based on findings possibly tainted by cognitive bias. And as the study's lead researcher points out in my piece, cognitive bias based on possibly misleading data can be self-reinforcing. If researchers are more likely to assume certain types of people...

... are guilty because that's what the data say, it can lead to misdiagnosed deaths, which then adds to the misleading data.

Many commenters made a point along the lines of, "Of *course* we should be more suspicious of a boyfriend than a grandmother." Perhaps. But that ...

... should not affect an *ME* determination about manner of death. If after analyzing a child's injuries an ME characterizes a manner of death as "undetermined," and based on statistics or a hunch, a prosecutor looks and finds other evidence of abuse, that's one thing. But ME ...

... determinations about manner of death are usually presented to juries as science. And if they are indeed based on science, MEs should largely come to the same conclusion when looking at the same set of medically relevant information. This study showed that introducing ...

... non-medical information resulted in way too much variability in those determinations.

Think of it this way: Let's say we give 100 crime lab analysts a blood sample and ask them to test it for heroin. For half, we tell them the sample came from a black man arrested outside...

... an abandoned building known to be inhabited by junkies. For the other half, we tell them the sample is from an elderly white woman arrested outside a sports bar. We would expect the drug test results for that particular drug to be *identical* ...

(within the margin for error of that particular test) for both sets of analysts. The race, age, gender, and circumstances under which the blood's owner was arrested should be irrelevant. If we *did* find a lot of variability in the results, we'd be pretty alarmed.

And in this hypto too, there are more variables than just race. But we'd certainly be worried that the study at least suggested racial bias in what's supposed to be an objective finding, and we'd want to see more research.

We'd probably also conclude that ...

... there's no good reason to tell an analyst conducting a drug test on a blood sample about the race, age, gender, or circumstances under which the person the blood was taken from was arrested. It's all irrelevant.

The important takeaway from this study is that ...

...manner of death determinations are often subjective and susceptible to cognitive bias, including but not limited to racial bias. To the extent possible, we should shield MEs from non-medical information, and instruct juries that these findings aren't always hard science.

/end