Sep 29 2011

Race and Perception

A new study finds that the way people dress affect our perception of their race. At least, that is what the authors and the press release claim.

The basic concept makes sense in light of what we have learned about perception. Human perception is a highly constructed phenomenon – we do not have a camera in our heads that passively records the outside world. At every level what we think we see is highly filtered, enhanced, and interpreted by our brains to construct a meaningful (but not necessarily accurate) image of the world.

These processes occur at the most basic level – enhancing edges and contrast, perceiving color and motion; and also at more sophisticated levels –  interpreting three-dimensionality and therefore size and distance. But also at the highest levels of perception – putting it all together as a meaningful object and scene, and imbuing not only meaning but emotion onto what we see.

Perception is also processed in terms of temporal synchronization. It takes different amounts of time for sound and light to reach us, and for our brains to process that information, and yet when someone claps their hands near us the sight and sound seem perfectly synchronized. As long as the two sensations are within 80ms of each other, they will be perceived as simultaneous – our brains tweak the time perception to make them match. But if the two sensation are separated by more than 80ms they will seem out of sync. The transition is also abrupt.

Further, the images we see are compared to our internal model of reality. We search for matches and meaning in what we see – is that a human face, if so, who? The visual stream is also compared to the other senses, and even modified to bring them all into line.

A dramatic example of this is the comparison of what we hear to how a person’s lips are moving to create the perception of what letters they are saying. This is known as the Mcgurk effect, and can be quite stunning.

Another example of this is gender perception, which is affected by a person’s voice. This effect is most dramatic the more ambiguous a person’s gender by sight alone. It therefore makes perfect sense that race perception will also be affected by multiple variables, and the effect of these other variables will be most dramatic with ambiguous stimuli. I was therefore prepared to believe the results of this study. But let’s see what the study actually found.

Subjects were asked to look at pictures of people and determine whether they were caucasian or African – a forced binary choice. The pictures were a morphed composite, and ranged from unambiguously caucasian at one end to unambiguously African at the other. In the middle the picture was therefore maximally ambiguous for race. The same pictures were either dressed in a suit or in coveralls (high status and low status attire). The authors conclude:

Together, the findings show how stereotypes interact with physical cues to shape person categorization, and suggest that social and contextual factors guide the perception of race.

But let takes a closer look a the data. Here is it plotted by the two factors, racial morph and attire:

On the left is the entire data set. As you can see, the effect of attire is barely perceptible. The authors zoom in on the data in the right plot to show the effect. For the most ambiguous racial morph the effect of attire was to raise the probability of a black categorization from 61-65% – a 4% increase. This result was barely statistically significant at P<0.05. What is remarkable is how tiny the effect is – if it is even real, given that it is barely significant.

The authors also did a follow up experiment in which they tracked mouse movements of the subjects, and they found that even in cases where the subject categorized a picture with low-status attire as white, the mouse temporarily moved toward the black choice. They conclude that we have different racial categorization schemes, one based upon facial characteristics but another based upon racial stereotypes, like social status. What they are seeing with the mouse movements are conflicts between these two systems, revealing the influence of the social stereotype.

These conclusions may be correct – they are plausible and as I said above they match what we are discovering about how our brains construct perception. But I was struck by the complete absence of any mention in the study or the press coverage about how tiny the effect size is. To me it not only calls into question the effect itself, but also (if real) its significance. I think this warranted some discussion. Even when the authors discuss the limitations of the study, there is no mention of the impact of this small effect size on their interpretation. In my opinion, they could have easily concluded that there is no significant influence of social status on race perception.

This reflects, in my opinion, a confusion of statistically significant with clinically significant. Just because a result ekes over the statistically significant line does not mean that the effect size itself is significant – both factors should be considered when interpreting the results.

So while the results are plausible, I am not impressed and will withhold judgment until the results are replicated.

18 responses so far

18 thoughts on “Race and Perception”

  1. locutusbrg says:

    The story leads you to believe that there is a extremely strong relationship between clothes and race perception. The quotes from the researchers give the impression that they think so. I thought 6% was rather low. I was only looking at the information they provided not the study itself.
    The movement of the mouse is not a scientific evaluation. It is easily interpretative to confirmation, or experimenter bias. It does not give any support to the concept. It is a better example of how the experimenters may have a poorly controlled for bias and that is represented in the conclusions.
    Another example of bad science reporting.

  2. PharmD28 says:

    A few basic questions.

    Why was the blue attire one of “low status” – the attire looks just like every day attire – I mean a button up neat looking dress shirt?…if anything it is representative of plain old average clothing…it would not seem to drive any sort of perception of “low” status stereotype I would think….seems like if they had wanted to even try and address a stereotype they would dress half of the pictures up in some true racial sterotypical clothing…perhaps I am off base with that comment.

    Also, I assume that they looked at these pictures on a computer screen…my current computer screen is not very high quality and so depending on the angle at which I view the photos, the darkness/lightness changes of the pictures…not sure if this thought is applicable to the monitors they used…but when I look at the screen right now and move my head up and down, their shades significantly change…so could that fowl up the reliability of their results and introduce even more noise? I know, a crazy thought but there it is for whatever that is worth…

    I agree too that the magnitude of difference in this case screams “who cares, irrelevant”…if these numbers were talking about ischemic stroke we may take more notice, but I am inclined to completely agree and say the “clinical” relevance of these results are little to none.

  3. sonic says:

    Tiny effect size is one thing- tiny sample (experiment 2 was 21 people) of non-randomly selected people (all undergrads at a US university) is another problem.
    To make claims about how “we” see things based on that sample is kinda silly- no?

    “…a confusion of statistically significant with clinically significant.”
    A brilliant distinction. Is it quantifiable?

  4. PharmD28 says:

    “A brilliant distinction. Is it quantifiable?”

    Interesting question…I will be interested to hear what Dr. N and others say about such a question. This comes up with my students and residents and we talk about this distinction, but I am not sure we have given it due diligence and thought perhaps?

    I am guessing that much of the phrase “clinically significant” is judgement…but judgement based on looking at that study, its design, possible flaws, confounders, inherent noise in the data…and how a given % change lets say may matter in the context of clinical practice…and perhaps a subsequent consensus statement from larger bodies of experts.

    If we assume a certain outcome in a given study is not bunk, and is externally valid…but it is a very small yet statistically significant difference….the “clinical significance” of that can be also expressed in types of analaysis such as pharmacoeconomic analysis – things like QOLY’s (quality adjusted life years), cost impact on society more broadly…but even such analaysis I have found within the ranks of clinical pharmacy when it comes to for example formulary decision making…many times those forms of quantifying these things are sometimes are viewed with some subjectivity because of concerns for the external validity or possible flaws of the original data…

    Will be interested how others think about this distinction.

  5. Bronze Dog says:

    I’m skeptical of the results. I would think such a connection is plausible, and worth testing, but I wouldn’t buy it based off this particular study. The sample’s just not good enough.

    If the effect does exist, I wonder just how pronounced it could end up with more variation in clothes, like if one line of figures was dressed up in stereotypical “gangsta” fashion, to use one extreme example. Such extremes might change the results by making the subjects more prone to second-guessing themselves.

  6. SARA says:

    When I ran my eyes over the two lines horizontally, I made the distinction of race at the same place.
    But what I found more interesting was the fact that I see a distinction in the same face when comparing them vertically.

    Compare the exact two faces, one above the other. They appear similar but not the same face. I had to force myself to study each feature to recognize that they are in fact the same.

    I think that is far more conclusive that we make a distinction to our perception based on the clothing.

    The question is – what distinction. I’m not sure it was race. I noticed a wider face on the lower. I also notice darker skin – even on the first entirely caucasian man. So perhaps it is race distinction, or perhaps it is just the way our perception creates distinctions associated with color. In other words – the dark suit makes the subject appear to be lighter. Artists use very dark to make an adjacent area appear lighter than it actually is. The eye adjusts for those things.

    However, like Dr. N, I don’t rule out the race, social status issue. There are so many interesting studies that show such distinction associated with our socialized understanding of people.

  7. locutusbrg says:

    bronze dog
    I think you bring up a good point and raise another one. What effect does the experimenters bias towards their definition of a “Stereotype” mean to the results. What does “Contextual Factors” mean. Like your example of Gangsta it could mean many things. Many different ways to look at this from that standpoint. What is the definition of a stereotype. For simplicity lets say English definition” A conventional, formulaic, and oversimplified conception, opinion, or image.”. That encompasses a lot of ground. A lot of wiggle room for the experimenters to decide what makes up the morph and the impact of that image.

  8. tmac57 says:

    Sara also raised an interesting point: What part does the contrast of shades in the clothing play in how the facial color is perceived?

  9. Crystal says:

    That binary choice isn’t consistent with the way people perceive race, at least not in the South where I live. A third category – light skinned or mixed race or biracial – should have been included.

    It is surprising to me that they chose coveralls as the low-status (and therefore associated with black people) category. Perhaps it’s due to my own family’s “lower” status, but I think of blue-collar workers as both white and black.

  10. Dan Roy says:

    And to add to Sarah’s point, the “low status attire” are place significantly higher up on the shoulders of the person – making him seem more broad shouldered and with a thicker neck.

    Not sure what this means but it explains the “wider” face problem I think.

    Either case, as with the contrast problem discussed above, it alters the perception of the face itself in a way probably not intended by the study constructors.

  11. cjablonski says:

    “I was therefore prepared to believe the results of this study.”

    This, I believe, is where most people (myself included) stop whenever we read about new research. Confirmation bias? Something more complex? As someone trying to become a more critical thinker, questions such as this make me realize I have a long way to go.

  12. daedalus2u says:

    In the actual tests there were finer gradations of change. In the image are shown only 5 gradations, in the tests there were 13.

    But the very low number of subjects present another problem. There were only 34 subjects. The change, from 61.7% to 64.7% represents a change from 21/34 to 22/34.

    The example in the BBC article, 65 to 71 is 22/34 to 24/34 or 64.7 to 70.6.

    This “data” represents the observation perception of a single person or of 2 individuals. The subject group (34) was white (26), male (13), black (1), East Asian (3), South Asian (3), Biracial (1).

    I would take this data as evidence of very little racial bias based on attire. If only a single person exhibited only a single gradation of bias, how is that evidence for rampant bias?

  13. milotoast says:

    I’m going to agree with the above comments regarding the colour and contrast of the clothes affecting the perception of the faces’ skin colour.

    This optical illusion came to mind:

  14. DLC says:

    I have to agree with Bronze Dog. this study is almost meaningless, simply due to sample size. Then there is confirmation bias also. “The mouse pointer hovered near” is not a valid metric, but an excuse to plant “the conventional wisdom” into the results. Con Wis says people perceive non-whites to be of lower social status and therefore of lower standards of dress.
    I’ve seen studies like this before, and always wondered about the effect size and various biases.

  15. sonic says:

    I think one thing I would look at is this– does the study sample the population one wants to make conclusions about?
    If one wants to make conclusions about humans, then I’m not sure that a non-random sample of college undergrads in US universities is going to be representative. In fact, I’m sure it isn’t.
    Lots of these studies are done– I understand that they sample the people available. And such a study could well be used to see if further testing is warranted– but no conclusions about populations not sampled– please.

    This is why these ‘mega’ studies can be misleading too.
    What good is summing a bunch of lousy studies?

    I’m still not quantifying.

  16. PharmD28 says:

    @Sonic…yeah I think we agree on this largely…

  17. noisyeardrum says:

    The critical question for me is “where are the error bars?”.
    Are they so small that they are obscured by the data symbols themselves or are they just absent?
    In my experience psychophysical and behavioural data such as these are very noisy – particularly across subjects.

    I agree, Steve, that if one is unable to observe an effect ‘by eye’ one cannot help but be highly skeptical.
    That said, small effects can be very real, ie. statistically robust and replicable.
    I can only wonder why the experimenters didn’t just move on.


Leave a Reply