Friday, November 02, 2007

Organizational Responses to Cognitive Bias

What can organizations do to counteract cognitive biases such as gender or racial stereotypes? If so much of our responses are automatic and activated by stereotypes and perceptions, what can organizations do to affect such automatic, unconscious behavior? What can/does the IAT measure? How can Organizations check implicit bias? What is the most efficacious methods organizations can deploy in order to combat discrimination in the workplace and increase diversity?

Bargh and Chartrand* posit that much of human cognition is automatic, and that it takes signigicant effort to think consciusly and counteract stereotypes. It is much easier (and more common) to go on "auto-pilot," and gowith the automatic cognitive processes that are often influenced by stereotypes:

•Dual Process Models: Contemporary psychology recognizes that the locus of control of psychological phenomena are determined jointly by processes set in motion directly by one’s environment and by processes instigated by acts of conscious choice and will. (463)

•Perception-Behavior Link: Perceiving the aggressiveness of a movie actor increased the later aggressive behavior of the perceiver.

•Stereotype Activation: Perceiving the faces of African Americans causes the perceivers to later act more aggressively in games.

•Unconscious affects the Conscious: Goals and motives can become automatically activated by situations.

There's a lot of literature on this now, especially from those folks at Harvard spear-heading use of the Implicit Association Test (IAT) to argue for greater response by law and policy to counteract people's unconscious bias (and EVERYONE is biased).

From the IAT Project:

How does the IAT measure implicit attitudes? Answer: The IAT asks you to pair two concepts (e.g., young and good, or elderly and good). The more closely associated the two concepts are, the easier it is to respond to them as a single unit. So, if young and good are strongly associated, it should be easier to respond faster when you are asked to give the same response (i.e. the 'E' or 'I' key) to these two. If elderly and good are not so strongly associated, it should be harder to respond fast when they are paired. This gives a measure of how strongly associated the two types of concepts are. The more associated, the more rapidly you should be able to respond. The IAT is one method for measuring implicit or automatic attitudes and is featured on this website. There are other methods, using different procedures, that have been investigated in laboratory studies.•What is an 'implicit' attitude? Answer: An attitude is a positive or negative evaluation of some object. An implicit attitude is an attitude that can rub off on associated objects. Example: The company for which your spouse works is attacked in a legal suit. An inclination to believe that the company is guiltless could be a reflection of your positive attitude toward your spouse -- your positive attitude toward the company provides an indirect (implicit) indicator of the positive attitude toward your spouse. (If you believe the company guilty, the marriage may be in difficulty!) The word 'implicit' is used because these powerful attitudes are sometimes hidden from public view, and even from conscious awareness.

•What is an 'implicit' stereotype? Answer: A stereotype is a belief that members of a group generally possess some characteristic (for example, the belief that women are typically nurturing). An implicit stereotype is a stereotype that is powerful enough to operate without conscious control.

•What can I do about an automatic preference that I would rather not have? Answer: First, bear in mind that these website IAT tests are not perfectly accurate. You may want to repeat the test before drawing even a tentative conclusion of this sort. On the other hand, it is very possible to possess an automatic preference that you would rather not have (the researchers who developed this test fall into this category). One solution is to seek experiences that could undo or reverse the patterns of experience that could have created the unwanted preference. This could mean reading and seeing material that opposes the implicit preference. It could mean interacting with people that provide experiences that can counter your preference. A more practical alternative may be to remain alert to the existence of the undesired preference, recognizing that it may intrude in unwanted fashion into your judgments and actions. Additionally, you may decide to embark on consciously planned actions that can compensate for known unconscious preferences and beliefs. This may involve acts in ways that you may not naturally act – for example, smiling at people who are elderly if you know you have a implicit preference for the young. Identifying effective mechanisms for managing and changing unwanted automatic preferences is an active research question in psychological science. The good news is that automatic preferences, automatic as they are, are also malleable.

Before I go on further, let me make it clear that I'm quite liberal, self-identify as a feminist, write in the area of employment discrimination law, have a J.D. concentration in Critical Race Theory (even if I largely ignore that now), and have always been committed to the legal and scholarly project of anti-subordination.

That said, I'm rather skeptical of the IAT. While I wouldn't disagree with the premise that people have unconscious bias and operate on auto-cognition guided by stereotypes, simple heuristics and categorical biases, I'm not entirely convinced that the IAT measures this. I took the IAT several times (tons of IRB issues there, by the way), at different point of the day--late at night, after coffee, at 6 am---each time my responses were different, and I would vary from mild to moderate to strong in my biases. Oh yes, despite all my aforementioned work and commitment, I am apparently a misogynist and racist. I really wonder whether the test measures cognitive response more than it does implicit bias. I suck at Tetris, and am not dextrous in any way--it was hard enough for me to type "i" and "e" fast enough to match up phenotypes to racial categories, much less switch them in the next round, much much less correlate positive/negative words to racial categories.

But don't just take my anecdote at face value. I'm not Lisa Belkin!

There are other critiques of the IAT:

Fiedler, et. al**, posit the following critique:

1.The asymmetry of causal and diagnostic inferences (Bayesian problem): a. IAT+ scores may be more prevalent than the attitude to be diagnosed, and IAT effects may reflect other causes than genuine attitudes. Diagnostic probability is lower than causal probability.

2. The viability of the Underlying Association Modela. Unit of analysis? What attitudes does the IAT measure? The purpose of the IAT is to assess the strength or degree of association between an attitude and the attitude object. Does attitude = association? Implicit responses to negative traits may be due more to negative implications for the respondent rather than the person holding the trait. One-dimensional Attitudes do not necessarily equal multi-dimensional associations.

3.The lack of a testable model underlying IAT-based inferencesa.The IAT test is a speed-classification task requiring the respondent to sort subsets of stimuli into two response keys denoting positive and negative attitude/evaluation. IAT is an indirect measurement procedure of associations, with no testable model as its basis. The IAT may be less person-scaling, more stimuli-scaling.

4.Difficulties of interpreting difference scoresa.The IAT Effect difference measure implies that being fast on compatible trials counts as much as being slow on incompatible trials. Being fast for responding X + positive and for Y + negative; and slow for X +negative and Y + positive receive the same weight. This strong assumption may be unsupported, and do not account for errors or extraneous influences, and the component contributions (e.g. devaluing Y and valuing X) can differ to a great extent.

5.Susceptibility of the IAT to deliberate faking and strategic processinga.To influence one’s IAT score intentionally, it is sufficient to think of admired or disliked individuals or engage in counter-stereotypical imagery.

Tetlock and Mitchell*** provide further critiques of the limitations of and potential problems with the IAT and the Unconscious Bias research paradigm:

•Claims of Unconscious Bias Researchers:
–Pervasiveness of Implicit Prejudice
–Insidiousness of Implicit Prejudice
–Uncontrollability of Implicit Prejudice
–Potency of Implicit Prejudice

•If true, this would have strong implications for transforming the workplace to implement strict accountability measures to minimize subjectivity and check uncontrollable, but unconscious bias, going beyond trying to control disposition to trying to control outcomes.

•Challenges to the Unconscious Prejudice Research Program
–“(1) Construct-validity debates: do we know enough about the causes and correlates of scores on implicit measures of prejudice to grant that these measures are tapping into the target construct of " unconscious prejudice" rather than into alternative constructs?;
–(2) Psychometric debates: assuming that implicit measures of prejudice do tap into their target construct to some degree, can the statistical links sustain inferences of the form “people who display differential reaction times of a given magnitude are some practically and statistically significant percentage more likely to discriminate against protected groups"?;
–(3) External-validity debates: do the claimed links between implicit measures of prejudice and criterion variables hold up in the face of the institutionalized lines of defense against discrimination that many organizations have created in the post-civil-rights-era?

--> For each question, our answer is "probably not."

So what can organizations do to check bias? It is one thing to have implicit, unconscious stereotypes, quite another to activate them and enact them. It is so hard to counteract the automatic cognitive processes and stereotypes that such "changing minds and hearts" is not only difficult, but least efficacious in affecting the actual effects of bias. If you want to correct the products of bias, such as gender imbalance in the workplace and lack of women/minorities in the management ranks, then the best thing to do is to actively target the statistical imbalance--not by quotas, which are unconstitutional--but by having outcome-oriented goals and accountability programs.

Kalev, Dobbin and Kelly**** examine the efficacy of different diversity programs:

•Seven Common Diversity Programs:
–Affirmative Action Programs
–Diversity Committees and Taskforces
–Diversity Managers
–Diversity Training
–Diversity Evaluations for Managers
–Networking Programs
–Mentoring Programs


–Structures establishing responsibility (affirmative action plans, diversity committees, diversity managers) result in significant increases in managerial diversity.
–Programs that target managerial stereotyping through education and feedback (diversity training) are not effective in increasing diversity.
–Programs that address social isolation among women and minorities (networking and mentoring programs) result in only modest changes.
–The key is establishing organizational responsibility and accountability measures, results-oriented programs rather than attitudinal-shifting programs.

•This should be considered in conjunction with Tetlock and Mitchell, who while not fully answering the question “what must organizations do to check implicit bias?”, posit that minimalist approaches to debiasing through weak manipulations such as accountability, outcome interdependence, and the salience of egalitarian norms “underscore the need to establish baseline metrics for accountability and other workplace checks on prejudice.”

In sum, it's very hard for organizations to try to counteract cognitive biases. But it may not behoove them to pursue that strategy to the exclusion of others, particularly results-oriented diversity programs. Attitudinal-shifting prorams are not only difficult to implement, but also less effective, particularly in large companies where cognitive responses are so varied and diffuse.

* Bargh, J. A., & Chartrand, T. L. (1999). The unbearable automaticity of being. American Psychologist, 54, 462-479.

**Fiedler, Klaus, Claude Messner, and Matthias Bluemke. (2006). Unresolved problems with the I, the A and the T: A logical and psychometric critique of the Implicit Association Test (IAT). European Review of Social Psychology 17: 74-147.

*** Tetlock, P. E. & Mitchell, P. G. (2008). Unconscious prejudice and accountability systems: What must organizations do to check implicit bias? In B. Staw and A. Brief (eds), Research in organizational behavior (volume 29). New York: Elsevier

****Kalev, A., & Kelly, E. (2006). Best practices or best guesses? Assessing the efficacy of corporate affirmative action and diversity policies. American Sociological Review, 71, 589-617.


Links to this post:

Create a Link

<< Home