Ethnic profiling is often said to be both ineffective and inherently discriminatory. On closer scrutiny, however, this turns out not to be true. Profiling, including ethnic profiling, does have benefits if designed and applied with care.
A while ago Dutch mathematician and science communicator Ionica Smeets published an enlightening newspaper article(in Dutch) about the self-amplifying feedback loop, a pitfall inherent in ethnic profiling and largely responsible for its vicious, reprehensible character. In short, she showed that ethnic profiling does not cause the police to actually catch more criminals. It just makes them think they do, and makes them harass and demonise ethnic minorities to boot. In a nutshell, ethnic profiling, in her view, is a misleading I-told-you-so machine. But is this really the case? To find out, we need to take a closer look at her reasoning.
Raven and vultures
As a point of departure, Smeets introduces two equal sized ethnic minorities that we’ll call Raven and Vultures here. The police know that 49 percent of the Raven and 51 percent of Vultures illegally go around carrying a concealed weapon. They also know these percentages to be stable over time. Now the Bill decides to go seriously twenty-first century, opting for data driven investigation by means of ethnic profiling. From now on, 1.000 random persons will be stopped and searched for weapons each month, proportionally divided among Raven and Vultures according to the latest success rates. So in month one, PCs stop and search 490 Raven and 510 Vultures, catching 240 Raven (490 * .49 = 240.1) and 260 Vultures (510 * .51 = 260.1) red handed. “Ha, look at that!” chief constables around the country happily conclude: “we pulled 500 knife wielders and gunslingers off the streets, no less than 52 percent (260/500) of whom are Vultures!” So the next month the police target 520 Vultures as opposed to only 480 Raven. And lo-and-behold, this time 256 Vultures get nicked, just over 53 percent of arrests. Way to go, the police conclude and increase the number of Vultures to stop and search to 530. Month after month this percentage keeps climbing, until after seven years virtually all the people checked are Vultures: 970 to be exact, as opposed to a mere 30 Raven.
This is, Smeets explains, the accursed self-amplifying feedback loop that renders ethnic profiling so vile. For if your cap, hoody or skin colour brands you a Vulture, you will get stopped and searched ever more often — far too often in fact, since as before only 51 percent of your mates carry a weapon, hardly more than the 49 percent among those fucking Raven that the filth never check out anymore.
It seems like flawless reasoning, so airtight as to make NASA blush. But all is not what it seems.
Unrelated
To begin with, there is the ominous fact that all this data driven surveillance does not change the police’s success rate one bit, nor the level of crime. In Smeets’ model, after seven years the police are still confiscating 500 weapons a month from 1.000 detainees. Doing it this way, ethnic profiling has no effect on the number of offenders — the very aspect that is socially relevant. It just changes which miscreants get collared. Even the dumbest PC Plod would smell a rat here.
And Plod would be right, for in addition to failing to address the matter of net results, Smeets’ reasoning is devastatingly fallacious. She takes those 260 delinquent Vultures apprehended in the first month for 52 percent of the total number of 500 criminals caught. In fact, however, those two numbers are completely unrelated. The success rate is determined within the group of Vultures, regardless of the number of Raven that get stopped. Whether only two Raven are patted down a month or two million, is immaterial in this respect. All that matters is the number of Vultures detained (510; 520; 530; …) and the number of weapon carriers among them (260.1; 265.2; 270,3; …): exactly 51 percent. Poof, there goes the feedback loop, vanished into thin air!
What Smeets showed was not a fatal flaw in the concept of ethnic profiling. It was the disastrous consequences of misjudging and misapplying data. I take it that Smeets, a mathematician, was aware of this, although she does not let on in her article. However, less gifted arithmeticians at the nicks might very well systematically fall for it. It does not help, then, for Smeets to serve up this example of faulty thinking as proof again.
MOT
It is actually a good thing that Smeets is in the wrong, for if that feedback loop was indeed inherent to ethnic profiling, it would perforce be inherent to all other forms of profiling as well, rendering them useless as well. Replacing the morally and emotionally charged term “ethnic profiling” by the more neutral “screening” shows just how extensive the consequences would be in the public and semi-public domains, including all government agencies and services, all education and health care and the worlds of insurance and banking.
For instance, countries like Britain and the Netherlands have been offering nationwide breast cancer screening for decades now, applying both gender profiling (only women are eligible) and age profiling (roughly, only menopausal women are eligible). The reasons behind this are both intentionally discriminatory and eminently sensible: breast cancer among men is rare, but it’s a one-in-eight threat to women, especially older ones. Age discrimination is also at the heart of the MOT-test, which is required of all cars older than three years, simply because newer vehicles almost never shows deficiencies that are relevant. For completely different reasons, age is a heavily discriminating factor when it comes to the treatment of minors and adults by the judicial and correctional systems. Doctors will sooner look for signs of lactose intolerance and sickle cell anaemia in African patients than in Nordic blondes, which is true blue ethnic profiling.
Stereotypes
Creating an effective, useful profile, ethnic or otherwise, isn’t science. It is an artisanal process of identifying practical associations and spotting steady correlations on the basis of recorded experience and expert opinions. If certain undesirable characteristics appear to occur more frequently or severely in some groups than in others, there is good reason for extra attention and investment. And yes, stereotypes are part and parcel of such processes. Stereotypes don’t come out of the blue, there is usually a grain of truth to them. This need not be a problem, as long as the profiling process is properly constructed, conducted and evaluated.
First off, profiling can only be useful if it concerns important and sizable differences between groups, much larger than the 2 percentage points of Smeets’ example. Subtle differences like that get too blurred in the noisy and messy real world full of interfering factors and human error, rendering a reliable assessment of results impossible.
Apart from a sufficiently large difference, we need an effective procedure to determine the results of our profiling effort, as well as its side effects. While designing a profiling campaign is not science, monitoring it and judging its outcomes are, especially where notions as fraught with emotions and mistrust as ethnicity are concerned. Clarity and reliability — in Popper’s terms explicitness and falsifiability — are of the utmost importance. So in every case, the profiling agency must first put down on paper exactly and completely explicitly which groups are targeted and on what grounds, what criteria are to be applied and which methods used. And, last but not least, there must be a crystal clear conception of the expected results. Then, it must be checked at set intervals whether the project is actually helping to achieve these result. If not, it’s game over. Judging results is not a matter of professional hunches or artisan know-how. It is a matter of theory and statistic data processing. If you want to know about the actual effects of the breast cancer screening program on the health of women, you don’t want an oncologist. Instead, you consult the cool, disinterested databases of the Office for National Statistics (ONS).
Dissimulation
Most important of all, however, is total transparency and disclosure right from the start. Openness forces high quality design, meticulous execution and scrupulous evaluation. If you risk being caught with your pants down at any moment, you won’t be tempted all that much to cut corners, treat data sloppily or worse: fiddle them to fit your preferences.
Unfortunately, openness about such matters is extremely hard to achieve, due to the rampant hysteria about discrimination that has taken possession of modern western societies over the past decades. It has created a deeply rooted culture of dissimulation that destroys much more than we are aware of. The principle of human equivalence, which is the very foundation of modern democracy, has degraded to a mindless dictate of banal equality, combined with a perverse aversion to recording and using relevant data for fear of stigmatising people. This ill-advised head-in-the-sand arrogance has become standard policy across the board: if you don’t record it, you needn’t confront it. It does not exist. This wilful ignorance seriously interferes with the development of effective policies from which everyone would benefit. Instead, it dangerously feeds mistrust and unfounded suppositions.
Yes, people are equivalent and share the same basic human rights. But no, that does not make them equal, they are neither exactly alike nor interchangeable. Men will treat women differently from other men, and conversely. The world view of religious people differs deeply from that of non-believers, and people differ radically in how they rate and respect social norms and ethical values. Such is life, much as certain ideological dogmatizers would like it to be different. But even more than in their character and attitudes, people differ in the ways they act and behave, individually as well as in groups. It is only wise to honour this when fighting crime, abuses and other wrongs.
Profiling, including ethnic profiling, can be a useful tool here, provided it remains eminently clear at all times that it serves to influence what people do or suffer from, not to discriminate them for who they are. This is exactly what went wrong in Smeets’ virtual experiment. But credit where it’s due: She did show us the paramount importance of openness and explicitness. For if she had not written about her reasoning so clearly and in such detail, we would never have been able to spot her fallacy.
