In a recent essay Are We Biased in Our Attitude to Bias?, I sought to argue that the idea of bias is a subjective one and that an attempt to infer it is likely in itself to be subject to bias. Nonetheless, it continues to be regularly argued that evidence exists of bias in our thinking and that there is a moral imperative from the perspective of Social Justice that we should therefore seek to change our behaviour. One of the most common justifications used to support this view is “controlled” tests where subjects are given fictitious candidate CVs, some of which have foreign-sounding names. The observation that those with the foreign-sounding names are less likely to be picked is much-touted and furthermore cited as evidence of bias:
Name bias is the tendency to prefer certain names over others, usually Anglo-sounding names.
Name bias is most prevalent in recruitment. If a recruiter tends to offer interviews to candidates with Anglo-sounding names over equally qualified candidates with non-Anglo names, this bias is present.
Name bias can have a negative impact on diversity hiring and result in companies missing out on talented candidates.19 unconscious biases to overcome and help promote inclusivity
Frequently it is also suggested that this is a form of discrimination, evidence of xenophobia or racism even. What I would like to suggest is that there are a number of considerations here which need to be distinguished if we are to evaluate such claims properly.
First, let us consider whether test conclusions based on fictitious CVs constitute de facto evidence of bias. As I have already suggested, the inference of bias requires “normal” behaviour to be defined against which problematic behaviour can be contrasted, and this is inevitably a subjective matter. The assumed normal behaviour here is a vetting process whereby the objectively “best qualified” candidate is selected based on the information presented. But is there such a thing as an objective best candidate? If there were, an automated algorithm would be preferable to peer review of CVs. But of course, we are distrustful of algorithms and rightly so. Famously, candidates have gamed the system by writing in a transparent font the names of famous universities which computer algorithms read and potentially take into account, but not human beings. We recognise the need for human judgment to be applied. But how can we at the same time ensure “objective” standards are applied?
Well, bear in mind we are dealing here with fictitious candidates, so all we have is text-based claims about them not anchored in any objective reality, and the fact the “job” here is also fictitious compounds this problem. So, for example, a degree from a well-known UK university might be cited on the CV of a candidate with a foreign-sounding name. If that candidate is not then short-listed, that might be assumed to illustrate bias. But equally, it could be argued, not only the university but also the department is important to consider, since some departments have better track records than others. Also there is the question of whether the subject studied is relevant to the job applied for. To infer a candidate’s eligibility based on the name of the university attended could therefore be cited also as a type of bias. In the end the human brain makes inferences, consciously or unconsciously, based on heuristic algorithms constructed on the basis of prior experience and these may or may not work well in practice. We trust them because they are the only means we have to access the requisite experience. But they can only be improved on the basis of more experience/evidence, not, as is well-known, by telling people their heuristic reasoning (indeed everyone’s) is flawed.
Concerning the suggestion above that such “name bias” impacts negatively on diversity hiring, it surely must do, since the express purpose of diversity policies is to skew results in the opposite direction away from Anglo ethnicity, irrespective of judgments about talent! But the further suggestion that introducing a bias against Anglo ethnicity makes a process fairer and less “biased” overall is contentious: the introduction of such putatively counteracting bias may result in less non-Anglo talent being missed, but will surely also result in some Anglo talent being missed. Whether more or less talent is missed depends on the distribution of “talent” among candidates of different backgrounds in the candidate pool. It is not difficult to find evidence that organisations pursuing diversity targets struggle sometimes to find any candidates which tick their diversity boxes for certain roles and will indeed send recruiters to second tier universities where they often find an ethnic mix which better matches their target diversity profile. Big Four accountancy firm PwC announced recently that it was waiving its minimum requirement of a 2:1 degree to further its diversity goals. It is problematic to try and argue that eschewing the university graduates with the best degrees to secure candidates with Anglo ethnicity is missing out on talent, but doing the same thing to secure candidates with other ethnicities is not.
In the end, though, arguments about the statistical probabilities of securing better candidates are something of a distraction. For one, the claims made are ultimately not verifiable since they always involve the investigation of counterfactuals and there is no objective definition either of “talent” or of what constitutes a good fit for a role, an issue we shall consider further below. Indeed, in implicit acknowledgement of such problematic issues, attempts are frequently made to put forward evidence that companies with diversity policies are more profitable or productive. For example the oft-cited 2020 McKinsey report Diversity Wins found that “in 2019, top-quartile companies outperformed those in the fourth one by 36 percent in profitability, slightly up from 33 percent in 2017 and 35 percent in 2014.” But whether this constitutes a compelling case that increasing diversity gives rise to better performance is another matter: one is immediately embroiled in the correlation implies causation fallacy, another cognitive bias one is well advised to eschew. Further, the desire to be able to infer causality is in itself prima facie evidence that confirmation bias is going on. Another fallacy that may be operating here is the halo effect whereby companies who broadcast their commitment to diversity may be perceived by applicants as more likely to prosper, so allowing them to attract the most talented individuals independently of actual “diversity” considerations. To emphasise the point, consider the following similar but less compelling argument that the success of Amazon, Apple, Facebook, Google, Microsoft and Netflix leads us to conclude that companies with names starting with letters earlier in the English alphabet are more profitable. Whether there is causality or not there is another matter.
More important to consider is, I would suggest, the assumption often smuggled into such discussions that evidence of name bias can be used to justify allegations of discrimination and xenophobia or racism; and further that any attempt to resist such imputation is itself evidence of xenophobia or racism. This is a pernicious argument and requires careful examination in the light of how human cognitive processes work to ensure that we distinguish genuinely problematic from normal human behaviour. In the first instance, let’s suppose the unwillingness to choose a candidate with a foreign-sounding name is founded in a dislike or distrust of foreigners (xenophobia). That some people do hold such views means that some candidates with foreign-sounding names will be unfairly rejected. This is clearly an undesirable outcome from any reasonable perspective. However the main argument made about the rejections in tests is that unconscious bias against foreigners is in operation which the subject is unaware of. Is this conclusion justified?
I would suggest the situation is more complex than the simple dichotomy of fair or biased allows; and that the decision to interpret data in terms of such is driven by a desire to reach a simple conclusion, namely that people are biased and so their behaviour needs to be controlled by diversity policies, which as we see are becoming increasingly widespread and intrusive as efforts are made to compensate for the ‘bias’. What then would be a more nuanced way of looking at this matter?
As I have suggested above, the whole point of a CV review process is to allow the judgment of a human being with relevant experience to bring that experience to bear in making a choice. But of course every single person’s life experience is different so there will not be and cannot be uniformity of decision making. The reason we are willing to live with such inconsistency is precisely because there is no clear unbiased view about which candidates are best qualified or why! The way in which any human being makes a decision is known from investigations in psychology and cognitive science to be based on a combination of conscious and unconscious judgments, mainly the latter with the former usually kicking in to justify a decision which was arrived at without any rigorous process of deliberation having been entered into.
Let us suppose that we turn the question around and consider instead what unconscious processes or heuristics might be at work in rejecting foreign-sounding names. A visceral dislike of foreigners may be one, but as I have already argued such a view is unlikely to be something the person themself is unconscious of. What other considerations might come into play? Given that, in the tests, the subjects are not chosen for their expertise in the relevant career area and the job candidates and selection process are all hypothetical for the purpose of the test (and known to be such), many of the bases for sound judgment about a candidate are necessarily absent, meaning that any decision has to be based on more generic criteria, i.e. heuristics. But heuristics are known to suffer from the limitation that they may work most of the time but not all the time, so may be suboptimal in certain circumstances. Human beings compensate for this by constantly modifying and improving their heuristics, consciously or unconsciously, as new evidence is obtained and may even actively seek out new evidence to support this, for example by reading up on a subject.
Heuristics are embedded in an internal model of the world and other people that we maintain in our mind-brains. It is known that our default model of ‘other people’ is (unavoidably) that they are like us other than with respect to certain differences which we have either learned through experience or imputed though the application of learned heuristics. This gives rise to the well-known phenomenon of homophily, whereby we engage more with people who are like us than with people who are different. The more like us people are, the less effort we have to invest into understanding them and accommodating them in our life. Specifically if people are encultured differently from us, we will naturally feel less close to them, given the importance of shared culture at the centre of much human interaction; in fact as the product of it. This is true whether a culture is a majority or minority culture. Indeed it tends to be stronger in the latter as minorities come together to support each other.
So, if we see a foreign name, we are likely to infer a cultural difference and feel less affinity than if the name was suggestive of a shared culture. If we envisage that a candidate may become a member of the team in our workplace, it is appropriate that we consider what kind of rapport might result. In the absence of other personal information in a CV, our assessment of the likelihood of cultural affinity is a reasonable heuristic for the prediction of good rapport. Of course, a name may not be a good indicator of a person’s cultural affinity: this aspect of the heuristic is limited in its accuracy. But such limitations are intrinsic to the process of judging a person by a CV and it is unfair to suggest ‘bias’ is in play when someone makes potentially incorrect inferences from a CV. The place to start fine-tuning inferences is at the face-to-face interview stage when clarifying questions can be asked and the potential for rapport assessed more accurately.
Are there other legitimate concerns which interviewers might look to address using candidates’ names as a proxy? One such would be whether a candidate has native language skills, which are important in many workplaces, whether in dealing with clients/customers or in internal communications. Again, there will be people with foreign-sounding names who are native speakers (likewise for the converse proposition), but the correlation is strong enough to make the heuristic a useful one. Taking native language skills into consideration will result in candidates with Anglo culture being selected preferentially, so would have to be considered ‘bias’ no less than would taking the ethnicity of names into consideration: the inaccuracy of the heuristic is not the fundamental problem here, but merely compounds it.
Another concern at the early stages of the interview process might be whether a candidate has a name which is easy to remember and/or pronounce. Phonetics vary greatly among languages and English speakers are notoriously poor in their mastery of foreign languages and pronunciation. Such issues might be a source of awkwardness at the interview stage if one failed to remember or mispronounced a candidate’s name. This can be avoided by giving preference to candidates with more familiar-sounding names. Indeed it is not uncommon, particularly for Orientals, to adopt Western-sounding nicknames which they use for business purposes as it allows them to establish a rapport more easily.
A final concern which is seldom addressed because of its sensitivity is fear of litigation; specifically the possibility of disputes at some point in the future which might result in complaints being made by a candidate, once hired, about alleged mistreatment at the hands of a manager or a colleague. If it can be suggested that the mistreatment was motivated or exacerbated by their having a non-Anglo ethnicity, such allegations would tend to be treated with much greater seriousness than otherwise. Of course, it would never be acknowledged explicitly, but the fear of such an eventuality is not an unreasonable one and could furthermore be allayed by precedence being given to candidates with Anglo ethnicity, even by reviewers of non-Anglo ethnicity. Screening out candidates with names suggestive of non-Anglo ethnicity would reduce the risk of such problems and furnish protection potentially for both employees and the company itself.
Ultimately, we would suggest, there is no feasible way of eliminating the effects of ‘name bias’ from a CV selection process, since the motivations which are likely to underlie this observed phenomenon will in many cases be legitimate concerns which are furthermore rooted in immutable human nature. The only workable strategy to prevent the likelihood of selection of candidates labelled with foreign-looking names from being reduced would be to remove the labels from all CVs, as is indeed the established practice of many companies now. Railing against implicit bias on the basis of names is akin to cursing the darkness. And accusing those whose judgment is influenced by names on CVs of discriminatory behaviour is akin to faulting them for stumbling in the darkness which their accusers have failed to illuminate.