5: Post Hoc Interventions and Swedish Discrimination Law

Anna Nilsson^[1]

Abstract. This chapter discusses the implications of Swedish discrimination law for the use of post hoc interventions during recruitment processes that involve the ranking of job candidates. It argues that such interventions may assist employers in preventing direct and indirect discrimination by alerting recruiters, and others responsible for hiring decisions, to the fact that biases may have influenced the recruitment process. In doing so, such interventions at the very least provide recruiters with a good reason to take a second look at their ranking choices and to reflect on whether the choices can be justified. The chapter also examines the circumstances in which employers that rely on incorrect recommendations from post hoc interventions can be held liable for discrimination.

1. Introduction

Imagine that you apply for a management position at a Swedish company. During the recruitment process, you are informed that the company uses a statistical tool called ‘GIIU’ to prevent bias and prejudice from influencing the recruiters’ decisions, including decisions about which candidates to interview and about the final ranking of candidates for the job. Initially, you find this approach professional and understandable. There is no shortage of studies that reveal discrimination in hiring decisions in Sweden. Studies have, for example, shown that Swedish employers tend to view people who are overweight as significantly less productive than people of average weight, and Arabs as less diligent than Swedes (Agerström and Rooth, 2007; Rooth, 2010; Agerström et al., 2012). Employers also tend to reject applicants over 55, in particular women over 60, and people with more than two children (Eriksson, Johansson, and Langenskiöld, 2012, pp. 13–17; Carlsson and Eriksson, 2017, pp. 12–14). Correspondence test studies^[2] have shown that homosexuals are less likely than heterosexuals with identical CVs to receive a positive response to a job application or to get invited for an interview (Ahmed, Andersson, and Mats Hammarstedt, 2013). For women, difficulties typically arise when seeking promotion or applying for managerial positions (Boschini, 2017, pp. 53–58). Studies from the United States have shown that women face a catch-22 situation when applying for managerial positions. When they present themselves as confident, competitive, and ambitious, they are viewed as highly competent, but they are nevertheless disliked, and therefore less likely to be hired (Rudman and Glick, 2001; Toneva, Heilman, and Pierre, 2020).

At the end of the recruitment process, you receive an email from the company informing you that you did not get the job. You start to wonder whether this negative outcome has anything to do with the GIIU tool. Did it really protect you against discrimination? Perhaps it saw biases that were not there and distorted the process. Wouldn’t that be discrimination?

This chapter discusses GIIU, the Generalized Informed Interval Scale Update, a prejudice-reducing intervention developed by Jönsson and colleagues in a series of articles (Jönsson and Sjödahl, 2017; Jönsson and Bergman, 2022; Jönsson, 2022). As the fictional example illustrates, interventions of this kind raise several legal questions. One set of questions relates to discrimination law. Do post hoc interventions such as GIIU facilitate better compliance with the Discrimination Act (2008:567)? If so, what specific legal wrongdoings do post hoc interventions address? And if GIIU makes a mistake, does the employer who bases decisions on that mistake engage in discrimination? This chapter discusses these questions. To facilitate the discussion, the next section provides a brief introduction to post hoc interventions. Sections three and four explore the possibility of using post hoc interventions to address direct and indirect discrimination, and section five examines the circumstances in which employers that rely on incorrect recommendations from post hoc interventions can be held liable for discrimination.

2. Post Hoc Interventions

As mentioned above, social science and psychology research has demonstrated that biased thinking and decision-making continue to be problems in the Swedish labour market. Post hoc interventions are new methods of preventing such malpractice. Behind them is the idea that we can identify biased rankings of job candidates through the statistical analysis of recruiters’ past rankings. Very briefly, this kind of intervention starts with an analysis of a specific recruiter’s past ranking with the aim of identifying patterns, such as, for example, a tendency to rank men higher than women, or people with Swedish-sounding names higher than people with Arabic names – or, indeed, vice versa.^[3] Such patterns are identified through the calculation and comparison of mean scores. First, we calculate the mean scores of members of the social group, or groups, that the recruiter may hold biases against (e.g. women or Arabs). Then, we compare these means with the mean scores that one would expect to find for these groups. If there is a statistically significant difference between the recruiter’s means and the expected means, the assumption is that the discrepancy is due to the recruiter’s ranking being influenced by prejudice or bias. The magnitude of the difference in mean scores is then used to propose a way of improving later rankings to better reflect the actual competences of the candidates (Jönsson and Bergman, 2022, pp. 5–7).

To conduct such an analysis, we need data about the distribution of job-relevant competences across relevant social groups. In the absence of such information, we must rely on assumptions about such distributions. Suppose that, in the fictional example in the introduction, the recruiter has a history of recruitments involving about 100 candidates and that the mean score for male candidates is significantly higher than that of male candidates. Such a difference would, of course, be less worrying if we knew that men were, on average, more qualified than women in the particular field at issue in this case. If, on the other hand, we knew or had reason to believe that male and female candidates were, on average, equally qualified for such work, then we would have reason to suspect that the ranking was influenced by prejudice and to take precautionary measures to prevent biased rankings in the future. ^[4] As Jönsson and colleagues have shown, post hoc interventions can under certain conditions mitigate the influence of biases during recruitment.^[5] Such mitigation may not only increase the chance that the best qualified candidate gets the job, but also assist employers in preventing discrimination. The next section discusses the specific forms of discrimination that post hoc interventions might prevent.

3. The Prohibition of Discrimination

3.1 Direct Discrimination

The Discrimination Act prohibits six types of discrimination, including direct and indirect discrimination (Discrimination Act, ch. 1 §4). The act classifies some other acts as discrimination, including harassment and instructions to discriminate, but none of these acts seems relevant to the problem that post hoc interventions aim to address, namely biased rankings of job candidates. Direct discrimination in the recruitment context occurs when an employer treats a candidate less favourably than another candidate in a comparable situation for reasons associated with sex, transgender identity or expression, ethnicity, religion or other belief, disability, sexual orientation, or age (ibid., ch. 1, §4(1)). Candidates who are roughly similarly qualified are considered to be in a comparable situation (Government bill 2007/08:95, p. 487). To determine whether two applicants have equivalent qualifications, the Labour Court looks at the criteria set by the employer; what kind of knowledge, skills, and personal qualities the employer is looking for; and how well the candidates meet these criteria. To constitute direct discrimination, the employer’s behaviour must, of course, also be related in a certain way to one or more of the discrimination grounds listed above. The preparatory works speak about a “causal link” between the employer’s behaviour and the job applicant’s sex, ethnicity, disability, etc. (ibid., p. 488). The discrimination ground need not, however, be the sole or decisive reason behind an employer’s action. It is sufficient that the candidate’s sex, ethnicity, disability, etc. contributed to a negative recruitment decision (ibid, p. 489). Such a link is obviously present in a situation in which a recruiter chooses to rank, for example, Arab candidates lower than Swedish ones because the recruiter dislikes Arabs or holds negative stereotypes about them. It is also present if the recruiter puts Arabs in a disadvantageous position because he or she or prefers to work with people from his or her own culture (ibid., p. 488). Social science research has shown that in-group favouritism – that is, people being more loyal and more benevolent towards people they consider to be like themselves (their in-group) than towards people they do not identify themselves with (the out-group) – may prompt such behaviour (Tajfel and Turner, 1979; Brewer, 1999; Wolgast and Wolgast, 2021, pp. 28–29).

From the above we can conclude that there is a significant overlap between the kind of biased rankings that GIIU seeks to address and the behaviour outlawed by the prohibition of direct discrimination. This suggests that GIIU could indeed help employers to prevent this kind of discrimination and hence facilitate better compliance with the Discrimination Act. The overlap between the biased rankings identified by GIIU and the legal prohibition of direct discrimination is, however, not total. The prohibition of direct discrimination covers many more acts than just those related to hiring decisions, and unlike GIIU the prohibition of discrimination is concerned only with biased behaviour connected to one or more of the discrimination grounds. These differences aside, the most difficult aspect to assess is how well the statistical analysis, which is a key part of GIIU, corresponds to the legal analysis of particular job applicants’ competences, which forms the heart of discrimination analysis. If these two approaches to identifying biased and discriminatory behaviour tend to generate different outcomes, that would speak against the usefulness of GIIU in preventing discriminatory hiring decisions.

As described above, a legal assessment of whether a job applicant has been discriminated against involves a comparison of his or her qualifications and the qualifications of other candidates who made it further in the recruitment process. In such assessments, no attention is paid to the mean scores awarded by recruiters or data about competence distribution across groups. In a case concerning the recruitment of a production artist, the plaintiff, represented by the Equality Ombudsman, presented data showing that people of Swedish ethnic origin were in a clear majority in the workplace in question. To be relevant to discrimination analysis, the Labour Court held, such data had to be combined with data concerning the proportion of people in Sweden who are of another ethnic origin than Swedish or, perhaps better, with information about the extent to which persons of an ethnic origin other than Swedish are represented within the specific branch under consideration in this particular case (Labour Court, 2009, no. 16, p. 26). It ought to be noted that this was not the main reason why the court rejected the Equality Ombudsman’s claim. Still, the court’s reasoning provides some pointers about what kind of statistical data the court might find relevant in future cases.

The fact that statistical data and analysis have played a limited role in individual cases concerning direct discrimination law does not necessarily mean that they should continue to do so. To be sure, even a clear pattern of a recruiter repeatedly giving lower scores to candidates from marginalised or subordinated social groups than to candidates from more privileged groups in the labour market does not provide conclusive evidence that these rankings are biased. Other explanations are possible. Even if we could establish that a particular recruiter’s past rankings were biased, that would not necessarily mean that the recruiter continued to let his or her biases influence future rankings. For that reason, a careful investigation of the particular ranking decision at issue in a case is indispensable. Still, a history of skewed rankings suggests either that candidates from the social group that benefits from the higher rankings are indeed better qualified, or that the recruitment process does not provide all candidates with equal opportunities. These are empirical matters, which cannot be settled by stipulation, and determining the most plausible explanation in a given context will depend on what we know about the distribution of relevant competences across groups within the relevant sphere, in combination with our knowledge of how bias and prejudice may influence recruitment processes.

3.2 Indirect Discrimination

I proceed now to indirect discrimination and the question of whether post hoc interventions can assist employers in preventing such misconduct. Indirect discrimination involves the application of a criterion or procedure that appears to be neutral but that puts people of a certain sex, transgender identity or expression, ethnicity, religion or other belief, disability, sexual orientation, or age at a particular disadvantage, unless the criterion or procedure has a legitimate purpose and the means that are used are appropriate and necessary to achieve that purpose (Discrimination Act, ch. 1, §4(2)). In recruitment processes, examples of such superficially neutral criteria are language requirements and dress codes that may be more difficult for ‘foreign’ job seekers to comply with. At first glance, post hoc interventions and the prohibition of indirect discrimination do not seem to target the same phenomenon. GIIU is not designed to identify, let alone question, job requirements per se. GIIU looks at rankings, and is designed to target prejudice and biases, attitudes that cannot be said to be neutral – at least not if they concern any group protected under the Discrimination Act. Nevertheless, what GIIU classifies as biases are repeated misrepresentations of job candidates’ competences associated with their sex, ethnic origin, or similar factors that cannot be explained by real or assumed differences in competence between men and women, Swedes and foreigners, etc. As noted above, this tool does not investigate the reasons behind these misrepresentations. It does not make an independent assessment of how well the ranked candidates’ competences match the job requirements for a specific position. Thus, although what GIIU identifies as a biased ranking may be the result of stereotypical thinking and/or explicit or implicit biases related to sex, ethnic origin, age, etc., it may also be a result of the application of a neutral criterion, such as a language criterion, that puts certain groups at a disadvantage. Unless such requirements correspond to real business needs, such as, for example, the need to communicate with customers in Swedish or some other language, they cannot be justified and are thus likely to violate the prohibition of indirect discrimination (Labour Court, 2002, no. 128, and 2005, no. 98).

To sum up, post hoc interventions seem to be designed to prevent direct discrimination in the form of biased ranking decisions that lead to discriminatory hiring decisions. Such interventions may, however, also capture instances of indirect discrimination. Given that GIIU does not evaluate possible explanations behind seemingly skewed ranking histories, except for explanations connected to the distribution of competences across groups, we cannot conclude that what GIIU classifies as a biased ranking will always result in unlawful discrimination unless the recruiter follows GIIU’s recommendation and updates the ranking. It is possible that the prior rankings can be explained or justified by reasons that GIIU has not considered. The next section discusses the room for such justifications in discrimination law.

4. Justifications and the Burden of Proof

Cases concerning direct discrimination often revolve around questions of evidence. Has the plaintiff been treated less favourably than others in a similar situation? If so, is the negative treatment related to the plaintiff’s sex, ethnicity, age, or any of the other prohibited discrimination grounds? The plaintiff must demonstrate circumstances that give reason to presume that he or she has been discriminated against (Discrimination Act, ch. 6, §3). If he or she is successful in doing so, the employer must show that discrimination has not occurred, in other words that the plaintiff was not subjected to less favourable treatment or that such treatment was not related to his or her sex, ethnicity, disability, etc. In other words, employers do not have to show that they selected the best qualified candidate for the job, but they need to convince the court that prejudice or other illegitimate considerations related to one or more of the discrimination grounds did not contribute – at all – to any unfavourable treatment. It is not enough simply to point to some other factor that also contributed to the decision (Government bill 2007/08:95, pp. 488–489). Employers have, for example, been held liable for discrimination based on sex in situations in which a job candidate’s pregnancy was one of the reasons why an employer decided not to offer her the job, even though the decision was also based on other (legitimate) reasons concerning doubts about her skills and enthusiasm for the job (Labour Court, 2011, no. 23, p. 12).

Some victims of discrimination have access to evidence revealing an employer’s “real” or openly discriminatory intentions, such as a secretly recorded conversation or similar evidence. In many cases, however, such evidence is not available, which means that claims about discrimination often depend on inferences from facts about the plaintiff’s competence in comparison to the competence of other candidates who differ from the plaintiff only with respect to their sex, ethnicity, or some other discrimination ground. To establish a presumption of discrimination based on, let us say, sex, a female candidate typically tries to establish that she has better, or at least equal, formal qualifications compared with one or more male candidates who were offered the job and/or invited for an interview. In a case concerning discrimination based on sex and age, the Labour Court found that it was sufficient to establish a presumption of discrimination based on sex, and thereby shift the burden of proof to the employer, for the plaintiff, a 62-year-old woman, who was not invited for an interview, to show that she had a stronger CV than some men who were invited for an interview (Labour Court, 2010, no. 91, p. 14). In addition, the fact that no woman over 50 was invited for an interview was sufficient to establish a presumption that the plaintiff was also discriminated against on the basis of age (ibid.). To defend its decision, the employer pointed to the fact that more women than men were interviewed, that the interviewees were of various ages, including a man in his 60s, and that two women were eventually hired (ibid., p. 7). None of these circumstances was, however, sufficient to rebut the presumption of discrimination.

The Labour Court has, however, accepted other arguments as refuting a presumption of discrimination. In the context of discrimination based on ethnic origin, the court accepted the employer’s argument that a highly competent candidate was overqualified for the job (Labour Court, 2009, no. 16). The case concerned the recruitment of a production artist. The candidate, a man of Bosnian origin, made it to the interview stage. The interviewers, however, got the impression that he had “moved on” to more qualified and creative work, and was therefore less interested in the rather standardised tasks performed by a production artist. This, in combination with their impression of the candidate as being an individualist rather than a team player, made him less suitable for the job than Swedish candidates with poorer formal qualifications but more fitting personal qualities (ibid., pp. 24–26). A recent study of professional recruiters shows that outgroup applicants may prompt recruiters to focus more on the applicant’s values and social skills and to subject these to closer scrutiny (Wolgast, Björklund and Bäckström, 2018). However, this risk was not discussed in the court case, which was decided in 2006.

Moreover, in situations in which candidates are roughly equally qualified, the Labour Court has accepted minor differences between the candidates’ qualifications as sufficient to rebut a presumption of discrimination. In a case concerning recruitment to a hospital unit responsible for moving patients from one ward to another, the employer defended the decision to select two Swedish applicants over a candidate of Kosovo Albanian origin with reference to the fact that one of the Swedish applicants had knowledge of the hospital’s underground corridor network, and that the other applicant had a friend who worked at the hospital and had put in a good word for him (Labour Court, 2006, no. 60, p. 13). Although we have little reason to doubt that knowledge of the corridor network was relevant to the position, it was not a competence specified in the job advertisement. This case and the case concerning the overqualified production artist illustrate that the prohibition of non-discrimination does not oblige employers to choose the candidate with the best qualifications; rather, it prohibits employers from rejecting candidates for reasons connected to their sex, ethnicity, disability, etc. Employers’ rather broad freedom to select employees dates back to an agreement from 1906 between the labour unions and employers, and has since been reaffirmed in the jurisprudence of the Labour Court (Labour Court, AD 1985:129, p. 797, and AD 1996:147, p. 1189).

The court’s lenient approach to the arguments and explanations put forward by employers has nevertheless been criticised by legal scholars and practitioners (Fransson & Norberg, 2017, pp. 105–106; Schömer, 2016). The low success rate of discrimination cases, in particular cases involving discrimination based on ethnic origin, even prompted an official inquiry into whether the rule governing the burden of proof ought to be amended to enable the Discrimination Act to better achieve its aim of combating discrimination and promoting equal rights and opportunities (SOU 2016:87, ch. 15). However, the inquiry concluded that the difficulty of proving discrimination was related not to the design of the burden of proof rule but rather to its application in individual cases (ibid., 463).^[6]

Even if Swedish law grants private employers considerable freedom in employment decisions, it is reasonable to assume that many employers would be interested in a tool that could assist them in ensuring that their decisions are based on rankings that accurately reflect the candidates’ actual qualifications.^[7] Post hoc interventions are one such tool. However, using this tool to adjust rankings is not without risk. As described in section two, the method relies on assumptions that may turn out to be incorrect in particular situations. The next section asks what happens if an employer relies on an incorrect recommendation provided by a post hoc intervention and, as a result, offers a job to a less competent candidate at the expense of a more qualified one.

5. Liability for Decisions Based on Bad Advice

For post hoc interventions such as GIIU to work properly and generate correct recommendations, a few conditions must hold. There is not enough space here for a detailed discussion of these conditions, but Jönsson and Bergman address this topic elsewhere (Jönsson and Bergman, 2022, and Jönsson, 2022). If one or more of these preconditions is not fulfilled in a situation in which GIIU has been applied, there is a risk that the tool will either fail to identify a set of biased rankings as biased, or suggest ways of correcting for bias that is not in fact present. For GIIU to generate appropriate recommendations, the statistical analysis involved in the intervention must among other things group the candidates into (roughly) the same social groups as did the recruiter whose level of bias is being tested (Jönsson and Bergman, 2022, p. 17). If, for example, the statistical analysis is focused on prejudice against women, but the recruiter does not hold biases against women in general but only against old women or very feminine women, there is a risk that the analysis will miss these biases, because biases against subgroups of women might have a small impact on the mean scores of the entire group of ranked women. If, on the other hand, statistically significant differences in mean scores are found, GIIU will suggest compensating for prejudice in cases where there is none; it will suggest that all women in the ranking are compensated, even though the recruiter’s biases affected only old women or those who come across as very feminine (Jönsson, 2022, section 3). A similar problem arises if a recruiter is biased against subgroups of men and women that are of roughly equal size.^[8] If they are of equal size, an analysis that focuses on differences between men and women per se will not find any statistically significant differences. As Jönsson notes, the method struggles with intersectional prejudice, both in terms of identifying such prejudice and in terms of making accurate recommendations about how to compensate for it (ibid., p. 20).

Another precondition that might give rise to incorrect recommendations in particular cases is that GIIU presumes that a recruiter’s prejudice is fairly stable between rankings. If in a particular case this is not true, and the recruiter’s prejudice has increased compared to previous recruitments, GIIU will undercompensate. It will still make a recommendation that will mitigate the effect of prejudice on the ranking under review, but it will not fully compensate for the negative impact of that prejudice (Jönsson and Bergman, 2022, pp. 16–17). If, on the other hand, the recruiter’s prejudice has decreased, the method will overcompensate. Following GIIU’s recommendations will, in such cases, decrease the veracity of the ranking, making it less representative of the candidates’ actual competences.

From the perspective of discrimination law, both undercompensation and overcompensation are problematic, but for different reasons and to varying degrees. Undercompensation (failure to fully correct for prejudice) implies that the use of a post hoc intervention will not be sufficient to avoid responsibility under the Discrimination Act: other measures will have to be implemented to ensure that no candidate is subjected to unfavourable treatment for reasons associated with sex, transgender identity or expression, ethnicity, etc. Such measures may involve, for example, criteria-based decision-making or the anonymisation of job applications. By contrast, overcompensation (correction for bias that is not there) is problematic because it entails a risk that the very use of a post hoc intervention will lead to a discriminatory decision. Take the example of a recruiter whose prejudice has decreased significantly since he or she compiled the rankings that were used to estimate his or her level of prejudice – perhaps thanks to some diversity or de-bias training.^[9] In an attempt to minimise the impact of any prejudice or stereotypical beliefs related to, for example, sex, he or she now uses GIIU to modify a ranking. GIIU recommends that female candidates have their scores increased and, as a result, a male candidate is ranked below a female one, even though the male candidate is actually better qualified. As a result, the male candidate is not invited for an interview or offered the job. This course of events seems to match the criteria for direct discrimination on the basis of sex (Discrimination Act, ch. 1 §4(1)). The man was certainly treated less favourably than similarly qualified women, and this negative treatment was undeniably related to his sex. Had he been a woman, he would have benefited from the same score increase as the female candidates. To constitute direct discrimination, it is sufficient that the candidate’s sex contributed to a negative recruitment decision; it does not have to be the sole or decisive reason behind that decision (see section 4, above).

Legally speaking, if a recruiter relies on incorrect recommendations from a post hoc intervention, it does not matter that the recruiter had no intention of treating candidates differently on the basis of a protected characteristic, nor does it matter that the recruiter was unaware that GIIU’s recommendations were erroneous. As described in section three, the Discrimination Act does not attach much weight to the employer’s intentions. Employers with benevolent intentions can also be held liable for discriminatory behaviour (Government bill 2007/09:95, p. 488). In a report on the use of automated decision-making in different areas covered by the Discrimination Act, the Equality Ombudsman argued that employers remain responsible for their decisions throughout the recruitment process regardless of which digital tools they use to make such decisions and regardless of whether they fully understand how such tools work (Equality Ombudsman, 2019, p. 16). If inspected by the Equality Ombudsman, an employer using a digital recruitment tool must, furthermore, be prepared to explain how it works and how it has been applied in particular recruitment cases. Given that it is up to the employer to design their recruitment process and determine what tools to use, and in view of the impact that hiring decisions have on people’s career prospects and livelihoods, this rule seems reasonable.

6. Concluding Remarks

This chapter has discussed the implications of discrimination law for the use of post hoc interventions during recruitment, and has argued that post hoc interventions such as GIIU may serve as a form of decision-making support that helps recruiters to select the most qualified candidate for the job and thereby avoid discriminatory hiring decisions. This argument is based on the view that GIIU simply corrects for biases and prejudice. It does not provide any candidates with preferential treatment but merely corrects biased rankings so that they better reflect the candidates’ actual competences. On this view, nothing in the Discrimination Act prevents an employer who has doubts about whether their recruitment procedures provide all candidates with equal opportunities from using a post hoc intervention as a form of decision-making support during recruitment.^[10] Post hoc interventions may very well form a part of the employer’s systematic work of preventing discrimination and promoting equal rights and opportunities during recruitment and promotion – work that Swedish employers are obliged to undertake (Discrimination Act, ch. 3 §§4 and 5(3)).

It is also possible, however, to view what GIIU does as a form of preferential treatment. Think back to the example in the introduction. Imagine that a candidate of a different ethnicity than yours is given a higher ranking because GIIU suggests that the recruiter is biased against people of that ethnicity. As a result, you lose your place as the top candidate, despite the fact that you and your competitor are equally qualified. Wouldn’t that be preferential treatment on the basis of ethnicity? If so, it would not be lawful under the Discrimination Act, because it results in unfavourable treatment on the basis of ethnic origin and violates the prohibition of direct discrimination. We could perhaps avoid this problem if the employer merely used GIIU to indicate that bias might have influenced the process, and this indication triggered a second, careful consideration of the candidates’ competences, which in turn led to an adjustment of the ranking. Still, there is a risk that a court would find that considerations of ethnicity contributed to the adjustment of the ranking.

Even if we view GIIU as engaging in some form of preferential treatment, it would still be lawful to use it to compensate for prejudice against persons with disabilities and persons with transgender identity or expression. This is simply because the Discrimination Act does not protect persons without disabilities against disability-based discrimination, and nor does it protect persons who are not transgender against differential treatment associated with this characteristic. Moreover, it would arguably be lawful to use GIIU as part of a systematic plan to achieve gender equality at a workplace in which one gender is underrepresented (Discrimination Act, ch. 2 §2(2)). It is, however, important that GIIU remains a form of decision-making support and that the employer makes an “objective assessment” of the candidates’ qualifications before the hiring decision is made (Hellmut Marschall v. Land Nordrhein Westfalen, C-409/95, §33). According to EU law, affirmative action on the basis of sex must not entail an automatic preference for the candidate of the underrepresented gender.

Moreover, interventions such as the one discussed in this chapter must always be implemented with care. If incorrectly applied, they may decrease the veracity of rankings, and even contribute to discriminatory hiring decisions (see section 5, above). Thus, it is important that those using the tool understand how it works and are able to assess whether the preconditions for its proper functioning obtain. In my view, these constraints ought not to discourage employers interested in the technique. The tool builds on established statistical methods and is transparent about the rules that govern the outcome. If applied correctly and in the right circumstances, GIIU will increase the veracity of ranking decisions and mitigate the influence of bias and prejudice. It may not always produce perfect outcomes, but there is reason to believe that its results will often be better than those based on a recruiter’s judgement alone (Jönsson and Bergman, 2022, pp. 22–26).

Acknowledgements

The author wishes to thank Senior Lecturer Per Norberg, Senior Lecturer and Associate Professor Leila Brännström, and Professor Jenny Julén Votinius for valuable comments on previous drafts of this text.

References

Agerström J. et al. (2012) Warm and competent Hassan = cold and incompetent Eric: A harsh equation of real-life hiring discrimination. Basic and Applied Social Psychology, 34(4), 359–366.

Agerström, J. and Rooth, D-O. (2007) Etnicitet och övervikt: implicita arbetsrelaterade fördomar i Sverige [Ethnicity and obesity: evidence of implicit work performance stereotypes in Sweden]. Uppsala: Institute for Evaluation of Labour Market and Education Policy.

Ahmed, A. M., Andersson, L. and Hammarstedt, M. (2013) Are gay men and lesbians discriminated against in the hiring process?. Southern Economic Journal, 79(3), 565–585.

Boschini, A. (2017) Olika kön, olika lön – en ESO-rapport om diskriminering på arbetsmarknaden [Different sex, different salary – a study of discrimination at the labour market]. Stockholm: Wolter Kluwers.

Brewer, M. B. (1999) The psychology of prejudice: Ingroup love or outgroup hate?. Journal of Social Issues, 55(3), 429–444.

Carlsson, M. and Eriksson, S. (2017) Påverkar arbetssökandes ålder och kön chansen att få svar på en jobbansökan? Resultat från ett fältexperiment [Does a job applicant’s age and sex affect his or her chances to of getting a response on to a job application? Results from a field experiment]. Uppsala: Institute for Evaluation of Labour Market and Education Policy.

Equality Ombudsman (2019) Kunskapsöversikt om användningen och utvecklingen av automatiserad databehandling med algoritmer (artificiell intelligens) och stordata och diskriminering eller risker för diskriminering [Knowledge overview of the use and development of automated data processing with algorithms (artificial intelligence) and big data and discrimination or risks of discrimination]. Stockholm: Equality Ombudsman

Eriksson, S., Johansson, P. and Langenskiöld, S. (2012) Vad är rätt profil för att få ett jobb? En experimentell studie av rekryteringsprocessen [What is the right profile to get a job? An experimental study of the recruitment process]. Uppsala: Institute for Evaluation of Labour Market and Education Policy.

FitzGerald, C. et al., (2019) Interventions designed to reduce implicit prejudices and implicit stereotypes in real world contexts: a systematic review. BMC Psychology, 7(29).

Fransson, S. & Norberg, P., (2017) Att förstå lagstiftning om diskriminering och mänskliga rättigheter [Understanding legislation about discrimination and human rights] Falun: Premiss förlag.

Jönsson, M. and Bergman, J. (2022) Improving misrepresentations amid avid misrepresentors. Synthese, 200.

Jönsson, M. and Sjödahl, J. (2017) Increasing the veracity of implicitly biased rankings. Episteme, 14(4), 499–517.

Jönsson, M. (2022) On the prerequisites for improving prejudiced ranking(s) with individual and post hoc interventions. Erkenntnis.

Palluck, E. L. et al. (2021) Prejudice reduction: Progress and challenges. Annual Review of Psychology, 72, 533–560.

Rooth, D-O. (2010) Automatic associations and discrimination in hiring: Real world evidence. Labour Economics, 17(3), 523–534.

Rudman, L. A. and Glick, P. (2001) Prescriptive gender stereotypes and backlash toward agentic women. Journal of Social Issues, 57(4), 743–762.

Schömer, E. (2016) Sweden, a society of covert racism: Equal from the outside: Everyday racism and ethnic discrimination in Swedish society. Oñati Socio-legal Series, 6(3), 837–856.

Tajfel, H. and Turner, J. (1979) An integrative theory of intergroup conflict. In Austin, W. G. and Worchel, S. (ed.) The social psychology of intergroup relations. Monterey, California: Brooks/Cole, pp. 33–48.

Toneva, Y., Heilman, M. E. and Pierre, G. (2020) Choice or circumstance: When are women penalized for their success?. Journal of Applied Social Psychology, 50(11), 651–659.

Wolgast, M. and Wolgast, S. N. (2021) Vita privilegier och diskriminering - Processer som vidmakthåller rasifierade ojämlikheter på arbetsmarknaden [White privilege and discrimination – Processes perpetuating racial inequalities at the labor market]. Stockholm County Administrative Board.

Wolgast, S., Björklund, F. and Bäckström, M. (2018) Applicant ethnicity affects which questions are asked in a job interview. Journal of Personnel Psychology, 17(2), 66–74.

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.