Illusory perception pervades every aspect of life, including, unfortunately, courts of law, where decisions are meant to be blind and unbiased. Studies of bias in legal decisions have begun to reveal the alarming inability of jurists to judge fairly when influenced by certain uncontrolled, extraneous conditions. For example, a study in 2011 showed that parole boards are less likely to grant clemency when making their judgments on an empty stomach.

One way to better understand legal bias would be to analyze how convicts’ appearance influences their sentencing, and a new study published in Psychological Science, by John Paul Wilson and Nicholas O. Rule of the University of Toronto, does just that. The researchers examined the sentencing of convicted murderers as a function of the trustworthiness of their faces. Specifically, they looked at whether or not the trustworthiness of each convict’s face could be used to predict whether the person was sentenced to either death or life-imprisonment.

It’s important to note that the facial appearance of trustworthiness is not a measure of actual trustworthiness, but scientists have shown that people tend to agree with one another in their assessment of individual faces. Therefore trustworthiness is a reliable and agreed upon feature of human facial processing, even if it’s not necessarily indicative of actual trustworthiness, and it’s a fair and important question as to whether criminal sentencing is affected by how defendants look.

For the study, the authors first analyzed the faces of 371 death row inmates (226 white, 145 black) in Florida. Though there were a total of 394 available at the time of the study, they excluded 23 to limit the study to only those that were rated unambiguously as either white or black males. As a control group the authors found an additional cohort of non-death-row inmates (also 226 white and 145 black) serving life-imprisonment terms for first-degree murder. Because there were equal numbers of each race in each cohort, any inherent racial bias among raters would thus cancel out. Had there been, say, more blacks in life-imprisonment cohort and more whites in the death-sentence cohort, any differences found between the cohorts could have potentially been due to rater racism.

Raters (recruited on Amazon’s Mechanical Turk service) characterized the faces for trustworthiness, Afrocentricity, attractiveness, and aggressiveness (based upon facial height-to-width ratio, a measure shown previously to be related to both actual and perceived aggression). Lab assistants also determined in each photo whether the criminals wore spectacles or had facial and/or neck tattoos, as these too might play a role in perceived trustworthiness.

The results were interesting. First, convicts with untrustworthy faces were indeed more likely to receive a death sentence than life in prison. That’s bad. The authors showed that this was true within both black and white groups, so the trustworthiness of the face was a factor irrespective of race. The scientists found that more aggressive faces among convicts, as measured by their height-to-width ratio, also led to higher sentencing of death over life-imprisonment. However, there was no statistical relationship between trustworthiness and aggression—meaning that untrustworthy faces were not systematically particularly aggressive faces—which indicates that both facial trustworthy and aggressive appearances play independent roles in criminal sentencing.

So courts clearly tend to put one to death more often when one’s face is either untrustworthy or aggressive. Since the photos that study participants used to evaluate convicts’ faces were taken after sentencing, perhaps the problem is that those sentenced to death look different than those sentenced to life because their time on death row altered their appearance. The study’s authors argue this affected the outcome of their research since the measure of aggression is based facial width-to-height ratio, which is determined during puberty. I find this logic weak. Because the appearance of aggression and trustworthiness are unrelated, it could very well be that trustworthiness changes in jail independently of other factors like aggression. Having said that, there is no reason why being sentenced to death rather than life-imprisonment would make one appear less trustworthy—if that’s the case, perhaps the post-sentencing photos used for facial evaluation didn’t affect the study’s outcome, but it remains an open question in mind, for now.

The larger gorilla in the room is that maybe the trustworthiness of the convicts’ faces is a true representation of their guilt and should be used in sentencing. This possibility is problematic for a number of reasons, but in order to address this concern, the authors performed a second analysis of criminal faces. This time, they examined the Innocence Project Web, a listing of convicted criminals who were later exonerated, usually due to DNA evidence that proved their innocence. From these records, the authors collected photos of 37 inmates (20 black and 17 white/Hispanic, 20 of whom had been sentenced to death and 17 to life-imprisonment), all of who had been eligible for the death sentence in the state in which they were convicted. They then had volunteers evaluate the faces as in the Florida study.

The researchers found that even among convicts who were later shown to be innocent, those who appeared less trustworthy were more likely to be sentenced to death. This shows that facial trustworthiness is not a fair measure of guilt, because these guys were all innocent and trustworthiness was therefore irrelevant. It follows that the courts are unfairly sentencing convicts to death when they have untrustworthy faces.

There is no governmental institution more dedicated to rational thought and fairness than the justice system, but this study shows that illusory facial perception is clouding the minds of our judges and juries, who can potentially decide to kill people based on their appearance. We must work to purge our system of these biases, and ensure that both convictions and sentencing are based on blind and fair measures: the stakes could not possibly be higher.