Die, Dead, Deceased, Lifeless, and Suicide

Last week, I wrote about the Implicit-Association Test (IAT), and whether it has use as a real-world predictor of behavior.  I have been skeptical of the IAT, because I do not believe that a longer latency on a language-pairing task necessarily reveals a real-world social bias.  How, as I mentioned last week, the IAT has predicted differences in treatment between black and white cardiac patients by physicians who did not declare any explicit racial bias.

And while I was researching the IAT, I found another paper (on which Dr. Mahzarin Banaji was also senior author), by Nock et al., from 2010, ‘Measuring the Suicidal Mind: Implicit Cognition Predicts Suicidal Behavior’, in which the IAT was used to positively predict suicidality in an at-risk patient population.

The IAT, as I have explained, is a measure of the speed at which a patient can assign a word to one of four categories, arranged into two pairs.

IAT - Task 3

The assumption which underlies the IAT is that, when the paired categories are cognitively linked in the mind of the subject, he or she will perform the task more quickly than when the paired categories are cognitively dissonant.

The authors of this paper administered a version of the IAT to 157 patients in a psychiatric emergency room, none of whom showed cognitive impairment.  Some had a history of suicidal behavior; some did not.  Some had co-morbid psychiatric disorders; some did not.

The particular IAT administered to these patients measured the latencies between matching words which signified ‘life’ versus ‘death’ (i.e. ‘alive’, ‘live’, ‘thrive’, and ‘die’, ‘dead’, ‘deceased’, ‘lifeless’, ‘suicide’), and words which signified ‘me’ versus ‘not me’ (i.e. ‘I’, ‘myself’, ‘mine’ and ‘they’, ‘them’, ‘other’)

Nock et al. found that a strong association on the IAT between ‘death’ and ‘self’ significantly predicted a suicide attempt: “Specifically, the presence of an implicit association with death/suicide was associated with an approximately 6-fold increase in the odds of making a suicide attempt in the next 6 months.”  This means that the IAT was a better predictor of future suicidality than either the prediction of the hospital clinicians or of the patients themselves.

Importantly, the Nock study didn’t establish a causal relationship between cognitive affect bias and suicidality.  There is no way to tell, from their data, whether negative cognitive bias actually makes you more likely to to attempt suicide, or whether it is an effect of the same emotional, chemical, or cognitive disruption that will lead one psychiatric patient to suicide and not another.  Either way, the paper suggests another way in which we are governed by, or reveal ourselves through, cognitive processes of which we are not consciously aware.

Image taken from Wikipedia.

Racial Bias and Implicit Associations

One of the speakers I was excited to see at the Society for Neuroscience conference this year was Dr. Mahzarin Banaji.  I have been curious, and had strong feelings, about her work for a long time.

Banaji is one of the researchers who has developed the idea that memory retrieval can be affected by our unconscious bias, and she is part of the team that runs Project Implicit, which uses the Implicit-Association Test (IAT) to examine unconscious racial bias (among other things).

The IAT asks subjects to assign words to one of two categories: for example, is the name ‘Aaliyah’ the name of someone ‘Black’ or a ‘White’?  These categories are on either the left or right side of the screen; the subject signals choice by picking the side of the category.

IAT - Task 1

The test then introduces another category: for example, does the word ‘Suffering’ belong in the category ‘Pleasant’ or ‘Unpleasant’?

IAT - Task 2

Finally, the test layers the two category choices, and asks the subject to pick the side of the screen which contains the correct category, of the four.

IAT - Task 3

The testers hypothesized that subjects would perform this task more quickly when the two paired categories on either side of the screen were also cognitively paired for the subjects, i.e. ‘White’ + ‘Pleasant’, or ‘Black’ + ‘Unpleasant’, in the case of racial bias.  So far, there is a fair amount of data which supports their hypothesis.

However, the IAT is problematic.  First of all, it is very easy to manipulate the test – the mechanism is simple, and once you’ve grasped it, you can easily delay your responses along whatever axis for which you do not wish to express a bias.

Second, and more importantly, the IAT assumes that a slower sorting time in a cognitive retrieval task meaningfully reveals something as complicated as racial bias.  There are many reasons why this might not be the case: language sorting and recall are governed by different parts of the brain than those which govern social parsing.  The two processes might be unrelated.

Which is why studies showing that implicit biases as revealed by the IAT actually predict bias in behavior are valuable.  For example, Banaji and her team published a paper in 2007 (Green et al., 2007, ‘Implicit Bias among Physicians and its Prediction of Thrombolysis Decisions for Black and White Patients’) showing that physicians who scored on high on the IAT for implicit racial bias were less likely to prescribe thrombolysis for black patients than for white patients with identical case files, despite the fact that the physicians reported (and probably believed) that they had no explicit racial bias.

In fact, white patients are much more likely than black patients to receive thrombolysis treatment, and there are a number of well-documented racial disparities in the treatment of cardiac patients in the United States.  Papers like Green et al., are important not only because they support the usefulness of the IAT, but also because they help explain how subtle and pernicious racial bias can be, and because they propose a quantitative, though imperfect, mechanism for finding it out in ourselves.

Sample IAT images borrowed from Wikipedia.