Article body

I’m not going to spend my life being a color

Michael Jackson and Bill Bottrell, Black or White (1991)

This paper analyses the semiotic features and errors of logic at work in racial profiling and racial reckoning. “Race” has never been about the color of a person’s skin. It has always been about what the colour of a person’s skin represents, and is thus a semiotic subject (see Hill 1998 : 681). Anthropologists have long been interested in cultural distinctions, social hierarchies, and human biodiversity and they have analyzed and have critiqued practices of race, including those of racial profiling. And anthropologists have a long and continuing tradition of employing Peircean semiotic theory in their analyses of human cultural practices (Daniel 1987, 1996; Munn 1992; Parmentier 1994). More directly, anthropologists have investigated semiotic ideologies and social consequences; that is, how semiotic orders (indexicality or iconicity for example) produce, structure, and condition different pragmatic social effects (Keane 1997; Keane 2003; Silverstein 1976).[1] Yet there has been little research on the semiotic forms of race : that is to say, not just what race means, but how it means; and as linguistic anthropologists have long argued, how something means is often a part of what something means (Silverstein 1976, 2003). This article explores one such set of empirical practices : racial profiling, and the semiotic ideologies through which it works.

In 2001 in New Jersey, the State Senate Judiciary’s office conducted an investigation of State Troopers’ practice of racial profiling in the eastern United States along the New Jersey Turnpike (2001). It was a well-documented case where state police targeted Black and Hispanic drivers disproportionately.[2] According to Lamberth in his study (1998), “African Americans made up 13.5% of the Turnpike’s population and 15% of the speeders. But they represented 35% of those pulled over. In stark numbers, Blacks were 4.85 times as likely to be stopped as were others” (Lamberth 1998). From a similar study Lamberth had conducted in Maryland in 1996-1997, he concluded that “17.5% of the traffic violators on the I-95 north of Baltimore were African American, 28.8% of those stopped and 71.3% of those searched by the Maryland State Police were African American” (1998). In 1998 Governor of New Jersey Christine Whitman dismissed Superintendent of the State Police Carl Williams after he said :“Today with this drug problem, the drug problem is cocaine or marijuana.…It is most likely a minority group that’s involved with that” (Ramirez et al. 2006 : 82).

All of this was part of a federal drug interdiction program called “Operation Pipeline” that began in the 1980s (Kocieniewski 2000). Critics argued that this and similar programs explicitly presumed that Latino and Black people (and more specifically Afro-Caribbean/Jamaican) were to be suspected of trafficking drugs. Researchers have suggested that police arrest a disproportionate number of non-Whites because they target and stop more of them, not necessarily because more of them are involved in drug trafficking (Holbert and Rose 2004; Ramirez 2006), a “circularity” problem.[3] This and similar phenomena have been termed “Driving While Black” and critics have called this “law enforcement by hunch”, or more broadly, racial profiling (see Russell 2003). A useful definition of the practice might be the “targeting [of] individuals for police or security interdiction, detention or other disparate treatment based primarily on their race or ethnicity in the belief that certain minority groups are more likely to engage in unlawful behavior” (Laney 2006).

This paper is based primarily upon U.S. data and events, and while the cultural ideologies at work are not necessarily generalizable to other parts of the world, some of the internal logic at work and the semiotic practices at work might be (see Khemiri 2013). The broader significance of these issues is widely understood (e.g., Gross and Livingston 2002). Similar kinds of state policing practices occur all over the world and as fears and concerns about security become heightened and generalized, these practices are perhaps taking on certain common inferential and semiotic forms.[4] Some observers have defended the practice, arguing that the logic employed is justifiable and empirically supported (Kocieniewski 2002; MacDonald 2003). But the defenders and paradoxically even some of the critics (see Lamberth 1998) all employ a similar logic of racial reckoning. And it is this that I want to explore more deeply.

What “Race” Is and Is Not

It should be stated clearly at the outset that “race” is not an accurate way of describing past or contemporary human biodiversity. Anthropologists have for decades argued through empirical data that human biodiversity is better understood and described through frequency distributions of specific traits (see Livingstone 1993). Even if we were to look at a trait generally characterized as a “racial trait” such as skin colour, we know that it is clinal in variation : there are no “clean breaks” in skin colour variation across populations. There is no one gene or allele that everyone in one of our conventional “racial groups” has and that no one in any of the other racial groups has. There are no non-tautological biologically homogeneous populations.[5] There has been gene flow across human populations since our species’ initial radiation from Africa over the past 100,000 years (and possibly prior to that), maintaining and reproducing our species continuity. Thus we can say definitively and unequivocally that there are no discrete biological human races (Goodman and Moses 2012; Hirschman 2004; Jorde and Wooding 2004).[6]

And yet we know that race is a very real social, cultural, political, and economic concept. It has real pragmatic effects on people in the world (Goodman and Moses 2012). Further, race often rests and rides upon human cultural arbitrations of physiological traits such as certain phenotypic features (skin colour would be a primary one : see Jablonksi 2004). So while human biodiversity does not naturally sort itself into discrete racial categories, certain biological characteristics that manifest as phenotypes are used by people to categorize humans into racial groups.

The operational concept of race that I am working with here is one that understands that race and racial identities are socially and culturally constructed, not biologically given. Having said that, the concept of race I am using is one that does involve biological phenomena, specifically, certain phenotypic aspects of humans. These phenotypic manifestations are used in calculating, arbitrating, and ascribing racial social position, and used also in the practices that reproduce social strata and hierarchies. Race, then, is a cultural taxonomy of biophysical symbols (about heritability, about phenotype) representing relative social power.

In what follows, I present first a discussion of the forms of inferential logic as described by Charles Sanders Peirce, and then I discuss how this corresponds with conditional probabilities. Following that, I discuss iconicity and symbolism and their relation to inferences about race and racial categories.

Inferential Reasoning

“See the value of imagination”, said Holmes. “It is the one quality which Gregory lacks. We imagined what might have happened, acted upon the supposition, and find ourselves justified. Let us proceed”.

Sherlock Holmes, in Silver Blaze [Doyle 2004 : 326][7]

In his famous essay “The Fixation of Belief” Peirce wrote :

The object of reasoning is to find out, from the consideration of what we already know, something else which we do not know. Consequently, reasoning is good if it be such as to give a true conclusion from true premises, and not otherwise. Thus, the question of its validity is purely one of fact and not of thinking. “A” being the premises and “B” the conclusion, the question is, whether these facts are really so related that if “A” is “B” is. If so, the inference is valid; if not, not

CP 5.365

In essence, to understand, reveal, uncover, re-present something we do not know, we extend what we know into that which we do not know – and for Peirce, logic was deeply intertwined with semiotic : “Logic”, he wrote, “in its general sense, is, as I believe I have shown, only another name for semiotic (sémeiötiké), the quasi-necessary, or formal, doctrine of signs” (CP 2.227). Semiotic, and the different grounds of representation between representamen and object, extends an aspect of the object for the interpretant. That ground is where we extend what we know into what we do not know.

Peirce described three forms of logical inference (CP 2.623; CP 8.209; Parker 1998) : Deduction, Induction, and Retroduction[8], and while some scholars argue that the third form is really just a subset or variant of induction, Peirce insisted otherwise. Let me give Peirce’s description of all three, using one of his most famous examples (only slightly modified), the one involving beans (CP 2.623).

Deduction can be thought of as a “Rule – Case – Result” sequence of inference :

Suppose you have a bag of beans, and are given the rule that all the beans in the bag are red. And then you take a handful of beans (the Case) from the bag. You can deduce, axiomatically, that the beans in your hand must be red (given that the rule is true). That is deduction : it is quick and efficient, but provides the least amount of information.

Induction can be thought of as the “Case – Result – Rule” sequence of inference :

You are given a bag of beans and begin pulling some out one at a time; each of the beans you pull out is red. After a while you inductively arrive at the rule (the rule is generated) that all the beans in the bag are probably red. That is induction, from many individual cases or instances to a general rule that describes and might predict for all the beans in the bag. The rule could be disproven with more data. And keep in mind it is the class “beans in the bag” for which the rule is generated (and not for the class of “red beans”).

Retroduction can be thought of as inference from “Rule – Result – Case” :

You are given a bag of beans and know that all of the beans in the bag are red. You see some beans lying on a table nearby. Those beans on the table are red. You guess that those beans on the table are from the bag. That is retroduction and at its heart, it is a guess, an inference backwards (CP 8.238).[9] It is a guess that the beans on the table are a result of the general rule of “being from the bag”. Peirce argued that this is how hypotheses are formed (CP 6.469; CP 8.238; CP 8.228) and he is clear that Retroduction is potential, but incomplete :

As for the validity of the hypothesis, the retroduction, there seems at first to be no room at all for the question of what supports it, since from an actual fact it only infers a may-be (may-be and may-be not)

CP 8.238

We must keep in mind that it is the characteristic of “belonging to the bag” that is being guessed at – but here the redness of the beans becomes the leading interpretive principle, and the leading sign : that is, the redness of the beans are hypothetically meaningful because they stand for something other than redness – they hypothetically point to, or stand for, “being from that bag”. In other words, the retroductive guess is that because the beans are alike in one respect (red), perhaps they are alike in another (from the same bag).

Of course, they might not be from the bag : one could test them possibly to see if they are, but the testing of them is itself a practice that might constrain other possibilities for those beans. Hypotheses are never “neutral” and when put into practice, they most certainly are not neutral, especially when put into practice by human social subjects, whether by scientists or by police officers. Equally, hypotheses are always context-laden not just context-conditioned.

In one of his 1903 Harvard Lectures, Peirce described the relationship between the different forms of logical inference, his phenomenological categories of Firstness, Secondness, and Thirdness, and the different sign-types :

Abduction, or the suggestion of an explanatory theory, is inference through an Icon, and is thus connected with Firstness; Induction, or trying how things will act, is inference through an Index, and is thus connected with Secondness; Deduction, or recognition of the relations of general ideas, is inference through a Symbol, and is thus connected with Thirdness

in Turrisi 1997 : 276-277

This chain of inferential logic corresponds with what Peirce said were the inferential semiotic grounds of each type of inference. For retroduction, it is formal similarity that is represented – that is to say, iconic similarity serves as the ground for retroductive inference. This is the weakest of the bases for inference, and retroduction is the weakest form of inference. For induction, Peirce says it is indexes that are representable – that is, it is fact or instance, the index, that serves as the ground for inductive inference. For deduction, Peirce says it is a symbol that is the ground for this form of inference – that is, some rule or convention, is represented.


Instead of beans in a bag we might substitute drivers along the New Jersey turnpike. Is this what the state police did? In essence, yes. They identified what they perceived to be a formal similarity between certain drivers and suspected drug traffickers – or in this case, a formal similarity between certain drivers and some indicted drug traffickers – and they hypothesized that if they are alike in one respect (skin colour), perhaps they are alike in others (trafficking drugs). In other words, officers treat similarity of skin colour as a result, as though it were representative of a general rule, of certain kinds of drivers being more likely to be trafficking illicit drugs. And it is based upon the officers’ perception : that is, the officers are the interpretants of the iconic symbols of race. According to Ramirez, “troopers do not ask drivers to identify their race or ethnicity. Instead, officers rely on their perceptions to provide the racial/ethnic data” (2006 : 83; and see Kocieniewski 2002).

The analogy between Peirce’s example with coloured beans in a bag and drivers on the New Jersey Turnpike is only partly apt : whereas in Peirce’s example the given rule is that “all of the beans in the bag are red”, in the empirical case not all of the of drug traffickers on the Turnpike are Afro-Caribbean, only a proportion of them are, albeit according to the officers, an “over-represented proportion”. Peirce’s example is apt in that what is being guessed at is whether the quality of “redness” represents something else, in Peirce’s case “being from the bag”. In the Turnpike example, what is being guessed at is whether being Afro-Caribbean represents something else, that is “trafficking drugs”.

Conditional Probabilities and the Phantom Denominator

Probabilities are non-arbitrary : that is to say, they are indexes of empirical facts, though what is obscured in this is that the facts are events organized and arbitrated by historically and culturally – specific human beings operating within equally contingent conditions. But the probabilities themselves are not arbitrary. Let me give a quick example.

Suppose you have 100 students in a particular primary school class. Let us say that there are 30 girls and 70 boys in this class. And let us say that the category of ‘‘top students’ comprises those top 10 students in marks/grades. That category is arbitrary, one could say top 5 or top 11, etc. Therefore, only 10% of the students are in this top category. And let us say that 7 of the top students are girls and 3 are boys. Therefore, girls are ‘overrepresented’ among top students, relative to their overall population in the class (and I should note that the math, whether it be 33,167 or 100 is the same – the results are not simply an artefact of my example).

Figure 1

-> See the list of figures

If we set up a 2 x 2 cross-tab, we would have boys and girls in the rows as independent variables, and not-top students and top students in the columns as dependent variables. If a teacher were asked to identify a top student and the teacher saw two students, a boy and a girl, walking down the hall toward her would she (the teacher) be more likely to correctly identify a top student if she selects the girl or the boy, knowing nothing else about them other than their gender? My example assumes that gender (and gender distinction) is an obligatory category for this school, which may not be a generalizable claim. In other words, while the math may be universal, the categories are not always.

If we do the math the probability that any individual girl is a top student is approximately 23% (7/30 – the number of top students who are girls relative to the number of girls overall), and the probability that a boy is top student is approximately 4.3% (3/70 – the number of top students who are boys relative to the number of boys overall). So the teacher would seem reasonable in guessing that the girl is more likely to be a top student, because girls are overrepresented among the top students relative to their numbers in the class overall. In other words, the probability that any individual female student in general at this school is a top student (knowing nothing else about her) is greater than the probability that any individual male student in general at this school is a top student (also knowing nothing else about him). We should note that the probability does not tell us if this particular girl is a top student (the student as index). When the teacher approaches her it is almost as though there is an ambient cloud of probability circulating around the student that accretes onto a specific aspect of her : her gender. Probabilities might not be “individual events” themselves, but they are relationships between events, and probabilities, just like relationships, are real. Before that particular girl is selected, we know that there is a 23% chance that she is a top student, but a 100% chance that she is a girl (possibly, with the above caveats about the variability and cultural specificity of gender performance and identity). Keep in mind that what is being guessed at is top student, not girl or boy.

Humans might make these kinds of retroductive guesses every day based upon what they believe is accurate, ambient “data”. And there is perhaps a diagrammatic similarity between abduction and conditional probabilities. We might not have truly internalized the probabilistic nature of the universe, but we seem to have done so with respect to social statistics (and its deep etymological ties to ‘statecraft’). This is particularly true in contemporary life, where statistical data has become a privileged form of symbol production about human beings (Hacking 1990)[10].

Now, let us substitute the example from the state police in New Jersey :

Figure 2

-> See the list of figures

Let us use as our 2 x 2 cross tab rows non-Afro-Caribbean descent and Afro-Caribbean descent, and in our columns non-drug trafficker and drug trafficker. There may be data from the US government indicating that people of Afro-Caribbean descent and other racial minorities are overrepresented among indicted drug traffickers, relative to their numbers in the US population overall. It is actually incidental as to whether or not this is the case. What is important is that this at least is what some in the state police in New Jersey believed (Holbert and Rose 2004 : 90-1; and see Williams statement above in Ramirez et al. 2006 : 82).

But the variables of non-Afro-Caribbean and Afro-Caribbean are not what are assessed by police on the Turnpike. The aspect of identity used is not national identity, but rather that of “race” : skin colour, the “fact of blackness”[11], so we should have instead as our independent variables “light melanin skin” and “dark melanin skin” (or simply light skin colour and dark skin colour), if we were to make the presumption that the category of Afro-Caribbean person is equivalent with dark melanin and the non-Afro-Caribbean person with light melanin (perhaps itself a problematic presumption). Or we could include other phenotypic traits (eye colour, hair colour, hair type, etc.) that are used in the arbitration of racial identity and racial position.

Figure 3

-> See the list of figures

Conditional probabilities are all about the denominator : if you do not have an accurate denominator for whatever group you are trying to determine the probability for – whether it is the total number of top students, the total number of students, the total number of girls or of boys, or of drivers on the highway or of Afro-Caribbean people – you cannot do a conditional probability, full stop. So here you would need to know the total numbers of light skinned people and of dark skinned people.

But nobody knows that number. And it is in fact unknowable : the categories of light and dark skin are, at their root, arbitrary and, of course, relative. This is the fractious line of race. There will be, as there has been for centuries, contestation over the arbitration of that boundary. It is an axis around which social, political, and economic power is stratified and dialectically deployed. It is, as W.E.B. Du Bois famously wrote, “the problem of the color-line” (Du Bois 1903).

As was noted above, human skin color is clinal in variation, with no natural divisions of human populations into dark and light skin colour (Livingstone 1993). Furthermore, skin colour variation is labile in nature :

Similar skin colors have evolved independently in human populations inhabiting similar environments. Darkly or lightly pigmented skin, therefore, provides evidence only about the nature of the past environments in which people have lived, rendering skin pigmentation useless as a marker for membership in a unique group or ‘race’

Jablonski 2004 : 614-5

In other words, skin colour, the “most visible aspect of the human phenotype” (Jablonski 2004 : 585), does not represent in any meaningful way uniform biological descent or discrete biological evolutionary descent within any of our arbitrated groupings of race, whether “light skin”, “medium- light”, or “dark skin”. Two people might be assessed by a particular interpretant to be in the “medium-light-skin” group, but the phenotype of their skin colour may have developed through different evolutionary pathways at different points in time of their ancestral genealogy. That is to say, similar phenotype of skin colour alone cannot be used to group people into common ancestral populations in any non-arbitrary way. And the bottom line is the boundary between the categories will always, always be arbitrary. The denominator is a phantom.

I should note that in policing practices and state security practices more generally, there are more variables than a simple 2 x 2 cross-tab. But with racial profiling, race is foregrounded and trumps these other variables. What occurs then is a retroductive reasoning that is treated almost as a deductive rule or principle to be applied. In other words, an inversion of the retroduction : the drivers of a specific racial category are treated as more likely to be criminals. Specific drivers are interpreted as members of that racial category (the first inference), hence they are inferred as more likely to be criminals (the second inference). This then projects into the future a condition that becomes self-fulfilling. Many critics have noted this “circularity problem” in racial profiling and its close cousin, the “ratchet effect” (Harcourt 2007 : 3-4, 28-9; Ramirez et al. 2006). For the person subjected to racial profiling, the probabilities and possibilities collapse, almost like the collapsing of a quantum state, and harden into certainty, no longer mere hypothesis or possibility. The probability that they were stopped is now 100%. Profiling itself is a practice, not a “neutral sampling method”. When a person is stopped, there is real Secondness followed quickly by the Thirdness of the consequences of racial prejudice. In this sense, the Firstness of retroduction and the possibilities available in Firstness are winnowed away into a convention of racial prejudice. In arbitrating racial identity, the leading sign working retroductively is perceived formal similarity of skin colour. While other habits of drivers are used in policing practices, as noted above, in assessing racial identity phenotypic aspects of humans seem to predominate. My argument is that skin colour (and perhaps other phenotypic characteristics) predominates through iconicity.


Skin colour in and of itself is not a sociocultural behaviour. And yet from the discussion above, in racial discourses and their practices, traits such as skin colour produce effects as though they are sociocultural practices. Phenotypic traits are foregrounded in the error of logic that is racial profiling. Why is race so amenable to and why does it lend itself so readily to these kinds of logical errors? As noted above, Peirce proposed that abduction was inference through iconicity. In turn iconicity is the semiotic form central to racial practices. Iconic signs represent for the interpretant some formal similarity or likeness. Iconic signs also correspond with the Peircean phenomenological category of Firstness, that of “pure possibility”. And unlike other kinds of human social identities, perhaps, race operates through representations of formal somatic similarities, such as skin colour. Furthermore, on the spectrum of arbitrariness, iconic signs are the least arbitrary. The iconic ground and this less arbitrary aspect maps quite well and represents other symbolic and arbitrary sociocultural conventions about race. Race as a sociocultural concept also involves cultural ideas of genealogy, ancestry, heredity, and descent – cultural discourses about our natural and biological selves. Here we see a central epistemological site for anthropology (and for anthropology’s main object “human being”), that is, human cultural practices where nature and culture are dialectically tied together. Our cultural conventions about race operate through iconic symbols and are built upon ideologies about iconicity, but also upon ideas of natural kinds, genealogy, and biological descent.[12]

As retroduction is inference through an icon, when forming conditional probabilities such as the ones that occur in racial profiling practices, the denominator upon which the probability is contingent is one that makes an inference about some ostensibly shared quality, in this case skin colour similarity. There is a kind of circularity, even tautology, that occurs with race-thinking, however, and this circularity itself diagrams the mirror-like quality of iconicity. Peirce writes :

Icons are so completely substituted for their objects as hardly to be distinguished from them. A diagram, indeed, so far as it has a general signification, is not a pure icon; but in the middle part of our reasonings we forget that abstractness in great measure, and the diagram is for us the very thing. So in contemplating a painting, there is a moment when we lose the consciousness that it is not the thing, the distinction of the real and the copy disappears… At that moment we are contemplating an icon

CP 3.362

The iconic relationship that stands between people of similar skin colour is itself made meaningful through human arbitration – melanin itself really does not have any interest in our fabricated racial categories.[13] Skin colour similarity itself is not meaningful, it is just potential. But in racial profiling and retroductive guesses about racial identity, skin colour becomes meaningful on an iconic basis (Lele 2006; Ransdell 1986[14]). That is to say, it is meaningful not upon the “fact” of similarity, but upon the possibilities the similarity might afford. It is this potential possibility that allows for a retroductive hypothesis to be made.

How do these possibilities become winnowed into the rule and practices of race and racial profiling? In addition to mistaken understandings about human biodiversity and the concept of “race” as discussed above, iconicity and retroduction are abetted through the complicity of complementary cultural discourses of race, specifically those about “natural” cultural orders, kinship and genealogy.

As nearly any anthropologist would argue, “Culture is our nature” (to paraphrase Edmund Burke[15]). Some cultural forms such as kinship are often proposed as “natural”, even though as anthropologists have demonstrated, they are culturally (and historically) variable in their form and in their pragmatic effects. The concept of “nature” obscures more than it reveals (Carsten 2004; Schneider 1980; Wade 2009; and see Herzfeld 2005 : 28 on “cultural resemblance and the ‘natural’” as iconic). This is particularly true in understanding the ways in which race articulates with kinship. Researchers have noted that racial groups are often understood as “fictive kin groups”, providing social support and networks of social relations that produce important pragmatic effects for people within a particular racial community (Sarkisian and Gerstel 2004). In addition, race continues to be conflated with early modern Linnean taxonomies and the “phylogenetic tree”, contributing to misperceptions and misunderstandings about human biodiversity (Goodman and Moses 2012; Hirschman 2004). As I have described elsewhere, naming practices are one of the primary kinship practices through which cultural ideologies about “natural social orders” are reproduced (Lele 2009). Citizenship laws of nation-states and the political assessments of birth are another, where the political discourses of citizenship organizes the practices of biological birth into a particular political status (Lele 2008). In the practice of racial profiling, the confluence of ambient probabilities, prejudice, and the possibilities inhering in iconic similarity accrete onto a specific person – racializing that person, making their skin meaningful (see Nash 1962 for an early critique of this).[16]


It could be that humans are cognitively predisposed to categorize in ways amenable to racial thinking, but these practices are always bound up with more complex sociocultural practices through which humans live their lives (Hirschfeld 1996; Lele 2010). It is in the deployment and articulation of the discourses of iconicity, heredity, kinship and genealogies, natural cultural orders, human phenotypes, biological communities and human biodiversity, and retroduction, that the possibilities of a guess, of Firstness, and of iconic signs are winnowed into Fact, and ultimately into habitual social structures.

Racial profiling involves a key error of logical inference, treating a retroductive guess as though it were a deductive rule to be applied. Why this happens is complex. My argument here is that ideological work of racial discourse is made possible by iconicity; and yet at the same time, iconic similarity is used for symbolic discursive purposes to prejudicially organize certain people into prior (probabilistic) categories. The cruel irony and circularity of race is that the possibilities seemingly afforded by iconicity are used against racialized persons who have to then account for their membership in a symbolically constituted social racial group, based ostensibly upon iconic similarity.[17]

The main semiotic forms at work in racial reckoning are iconic symbols. There is a tension in an iconic symbol : it suggests possibilities, as icons are incomplete, degenerate signs. Symbols, by contrast, are more settled and they are effective only to the extent that they can convene through (human) arbitration. On their own they can be taken up and put down again, attended to or ignored. But combined with iconicity, they bear a seemingly self-evident weight. Race, and racism, persist in part due to this complex semiotic tension.[18]