3. Diversity policies in the OECD and evidence on their effectiveness

This chapter assesses OECD countries’ existing policies for diversity in the labour market in light of the gradually growing body of literature evaluating the impact of such instruments. The effort to synthesise diversity instruments with the evidence in a comparative, cross-country manner must be preceded by a note of caution: the evidence currently at hand is still sparse and highly context-specific. The majority of studies evaluates diversity measures in a particular country, if not a particular firm, with limited transferability of results for a given policy instrument from one context to another.

It is, however, possible to provide initial pointers on how and under which circumstances existing diversity measures seem to promote anti-discrimination and diversity efforts. Taking an evidence-based view on what has been learned in several decades of “diversity and inclusion” on the political agenda of OECD countries, this overview seeks to derive lessons for future policymaking. A key result of this review is that diversity efforts must be targeted to the national context and the situation of a specific diverse group – otherwise, well-intended diversity measures can be ineffective and, at worst, counterproductive.

The chapter is based on a recent policy questionnaire answered by 30 OECD countries, as well as the findings of the OECD-Dauphine University HR survey conducted with around 2 400 HR professionals in eight OECD countries (see Box 3.1). Each section discusses a common policy instrument and its effectiveness.

In virtually all OECD countries, diverse populations are protected by non-discrimination legislation in the field of employment. In almost all countries, discrimination is prohibited on the grounds of sex/gender and gender identity, sexual orientation, age and race, ethnic origin or skin colour (see Annex A for an overview of protected grounds in OECD countries). Religion or belief is another commonly protected ground, while some provisions are specific to certain countries, e.g. for indigenous, native or aboriginal people in Canada, the United States and Australia. Australia, France, Finland, the Netherlands and Portugal notably go well beyond the commonly protected groups, including a variety of grounds such as political or trade union activity, criminal history, family or marital status, physical appearance, type of contract or place of domicile. Often, the above are non-exhaustive lists (including e.g. “any other grounds”) and eight countries explicitly recognise discrimination on multiple grounds (see Box 3.1).

Twelve out of the 30 OECD countries that answered the questionnaire include socio-economic disadvantage as a protected ground in their anti-discrimination legislation. The variety in definitions used, including social origin, social status or social class as well as wealth, income, financial or economic situation, illustrates the difficulty with operationalising socio-economic status (SES). As it targets a changing characteristic, discrimination based on SES is not straightforward to measure and a cut-off is difficult to determine (see Box 3.3).

The empirical literature on the effects of non-discrimination legislation remains scant, with most studies addressing the US context. Existing evidence suggests that the initial adoption of the landmark anti-discrimination law Title VII of the 1964 Civil Rights Act initially aided the economic status of black workers in terms of both employment and wages, but with diminishing returns over time (Donohue, 2005[5]; OECD, 2008[6]).

Other studies highlight possible unintended side-effects of legal protection. In their analysis of disabled workers’ employment following the American Disabilities Act (ADA), (Acemoglu and Angrist, 2001[7]) observe a decline in employment of disabled workers after the ADA went into effect. The authors suggest that with the requirements to accommodate the needs of disabled employees, as well as regulations regarding their hiring, firing and pay, the perceived costs of might have led employers to refrain from hiring workers with disabilities. It was not possible to identify the component of the law that caused the effect, which might have been the cost of accommodation of disabled workers (costly equipment etc.) as well as the higher cost of hiring due to fear of litigation. Other studies argue that the decrease in the employment of workers with disabilities during the 1990s was a result of changes that expanded Social Security Disability Insurance (SSDI) eligibility and benefits (Autour and Duggan, 2003).

Aside from effects on employers’ behaviour, the mere institutional decision to protect certain groups may impact the wider society’s perception of social norms and influence attitudes towards diverse groups. A recent study using experimental strategies investigated how the June 2015 US Supreme Court ruling in favour of same-sex marriage affected behaviour and attitudes towards minority groups (Tankard and Paluck, 2017[8]). The authors observe that attitudes in support of gay marriage were significantly more positive among participants who were told that a positive ruling was the most likely outcome. The subsequent five-wave longitudinal time-series indeed indicated that after the actual ruling, participants believed that social norms shifted in favour of same-sex marriage. These results are in line with previous research suggesting that changes in legal norms, if democratic in nature, are seen as a signal of where the public stands, which in turn can influence attitudes and behaviour (Cialdini and Goldstein, 2004[9]).

However, for non-discrimination legislation to have any effect, awareness of legal provisions is crucial, as are effective recourse mechanisms for individuals who feel discriminated against (Heath, Liebig and Simon, 2013[10]). In almost all OECD countries, specialised equality bodies and ombudsman offices assume this task alongside civil society initiatives, trade unions and NGOs.

The competences of equality bodies in OECD countries vary. Generally, such bodies provide information about non-discrimination laws and the possibility to take legal action. In some countries, including Italy, Mexico, Sweden, the United Kingdom, Belgium and Denmark, equality bodies have litigation powers. However, even when legal representation through an equality body is possible, it is not provided in a systematic way and only strategic cases that can provide clarifications on the law or create precedence are taken up. Another role of equality bodies is that of a third party in mediation and conciliation procedures, an approach that may be less damaging to the employer-employee relationship than litigation. In almost all countries, specialised bodies assume an awareness-raising role, conducting information campaigns and/or trainings related to anti-discrimination issues in their country (see Figure 3.1).

Figure 3.2 indicates that in all European OECD countries for which data is available, awareness of non-discrimination provisions has increased between 2007 and 2015. However, the numbers also show that there is still a long way to go in terms of awareness of rights. In the majority of countries, less than half of the population would know their rights if they were to become a victim of discrimination or harassment.

Further strengthening the awareness of workplace rights is critical. This must go together with strong recourse mechanisms ensuring that potential victims of discrimination can actually avail themselves of their rights, while being protected from retaliation (see Box 3.2).

Affirmative action (AA) in hiring is broadly defined as policies addressing the under-representation of disadvantaged groups in the labour market through active measures that go beyond simple non-discrimination. In other words, AA instruments seek to increase the numbers of women and minorities in the workforce, most commonly by setting numerical targets including compulsory quotas, tie-break rules, and aspirational goals, which carry different levels of sanctions in case of non-compliance. Versions of affirmative action are found as “positive action” measures in European countries, where such policies are more recent and less extensively analysed than in the US context.

Under affirmative action legislation in the United States, federal contractors were required to implement affirmative action plans in a “good faith effort” in reaching certain targets to ensure representativeness in racial minority and female employment. Regular compliance reviews were an important factor for the effective implementation of affirmative action. Leonard (1990[11]) finds that among the firms subject to affirmative action policies within labour markets of the same industry and region, reviewed contractors fared better than the non-reviewed. Moreover, federal affirmative action requirements typically led to assignment of responsibility within the organisation to assure compliance, which has been another crucial factor for the successful implementation of affirmative action (Kalev, Dobbin and Kelly, 2006[12]). Finally, overall employment growth is a strong contextual factor influencing the success of affirmative action – evidence shows that such policies have been far more successful at growing establishments that have job openings to accommodate federal pressure (Leonard, 1990).

Overall, several studies show statistically significant shifts in the redistribution of jobs from white males to minorities and women following the introduction of federal affirmative action legislation (see Table 3.1). These beneficial effects for women and minority groups are persistent, as (Miller, 2017[13]) finds when examining the long-term effects of federal affirmative action regulation. The study, relying on turnover in public procurement relationships to compare contractors before and after falling under regulation to those firms that have never been contractors, indicates that the share of black employees continued to grow for more than a decade after a firm was deregulated. Research in the US context further suggests that following affirmative action legislation, which applied less forcefully to smaller firms with few employees, female and minority employment moved to larger firms. This transition to larger companies, which generally offer better pay, had a positive impact on women and minorities’ wage growth (Carrington, McCue and Pierce, 2000[14]; Chay, 1998[15]).

In the context of positive action in Europe, countries are increasingly imposing concrete numerical goals on firms. Here, it is important to distinguish between targets, which are indicative, and quotas, which are compulsory (Heath, Liebig and Simon, 2013[10]). In the past 10 years, mandatory quotas for women in management boards and persons with disabilities have proliferated in the European context (see Table 3.2).

Some countries, in contrast, set voluntary targets. The Swedish government, for example, suggests that 40% of management board members should be women; a similar target is set in the Netherlands. The Swedish government has also set the goal that at least 25% of newly recruited employees in the heavily male-dominated construction industry should be female by 2030. Furthermore, firms voluntarily self-regulate in many European OECD countries, stipulating non-binding targets in Corporate Governance Codes. However, only around 10% of respondents in our HR survey indicated that their firm had set targets for interviewing a certain minimum number of applicants from an under-represented group. The main groups for which these firms set interview target were women (76%), disabled people (39%), ethnic minorities (30%) and migrants (19%). Of Australian and Canadian firms with interview targets in place, 51% targeted indigenous populations.

Goals set in the context of affirmative action had a significant correlation with firms’ improvement in employment of women and minorities in the United States (Leonard, 1985[16]). The impact in Europe is less researched, but voluntary targets adopted under Dutch affirmative action law Wet SAMEN had no significant effect on firms’ end-of-year workforce composition (Groeneveld and Verbeek, 2012[17]). Unsurprisingly, non-binding targets are not being fulfilled with the same rigidity as are mandatory quotas. By way of example, the Norwegian gender quota requiring 40% females in company boards shows that under threat of dissolution of the board, which is foreseen in legislation in case of non-compliance, all companies ultimately complied. Female representation in Norwegian boards raised from less than 8% in 2002 to the required 40% after the deadline imposed by the law (Storvik and Teigen, 2010[18]). Similar results are found in the European comparison, where Germany, Belgium, France, and Italy have successfully increased the presence of female directors on boards through quota regulations (Horst and Wrohlich, 2018[19]).

The effectiveness of quotas in the European context is likely attributable to the strict sanctions imposed. A common sanction for firms that fail to comply with gender quotas in company boards is that any new appointment not in compliance with the respective law is automatically nullified. This so-called “empty chair rule” is applied in Austria, Germany and France. In Belgium, the law foresees the suspension of any financial or other benefits of all board members until the quota law has been complied with. In Italy, sanctions are applied progressively, starting from warnings, moving to fines and finally to forfeiture of the offices of all members of the board.

Firms that fail to comply with disability quotas generally incur financial penalties. Most commonly, countries have so-called “quota levy systems” in place – an equalisation mechanism that redistributes the funds raised from non-compliant firms to cover the costs of those firms that actually do employ disabled people. Some of the funds raised are also used for other purposes that further the labour market integration of disabled people. In some countries, firms that fail to hire the required number of disabled people can still be in compliance with quota. In France, firms can discharge their obligations to fulfil the disability quota by contributing into a fund for the promotion of employment of persons with disabilities. In Slovenia, firms can also partially outsource their activities to a company with more than 40% employees with disabilities.

The public discussion, however, is less centred around questions of effectiveness and focuses more on questions of fairness. This is particularly the case with regard to mandatory quotas, as the impossibility of ascertaining disadvantaged groups’ counterfactual representation were there no discrimination inherently leads to suspicion that quotas excessively put minority groups at an advantage (Valfort, 2018[20]). This can trigger stigmatisation and backlash, as women and minorities may be perceived as receiving unfair preferential treatment or as being underqualified.

Furthermore, the overall positive effect of affirmative action must be qualified. While affirmative action leads to increases in employment for all groups, the extent of the benefit seems to differ, with some gaining significantly less than others. Evidence from the United States shows that the positive effect is strongest for White females, who account for more of the difference in employment rates following the introduction of affirmative action than do Black males, Black females or Hispanics (Leonard, 1985[16]; Holzer and Neumark, 2000[21]; Kalev, Dobbin and Kelly, 2006[12]). Why this is the case is more difficult to determine. White women, as the largest group of workers after White men, who are on average more highly educated and maybe more qualified than minorities, might be easier to recruit than other disadvantaged groups (Holzer and Neumark, 2000[21]).

The impact of affirmative action does not only differ between groups – a similar imbalance in advantage gained seems to hold for individuals within respective groups. For one, evaluations of the gender quotas in Norway and Italy have shown that the positive impact was confined to the direct effect on the women appointed for board management and that such policies had no impact on the advancement of the general female workforce of a firm (Maida and Weber, 2019[22]). Furthermore, critics argue that by targeting minority status rather than socio-economic factors, affirmative action benefits the most privileged middle- or upper-income brackets within minority groups at the expense of lower income ethnic minorities and lower-income White people (Kahlenberg (2012[23]), see Box 3.3).

Given that the business case at the firm level is not always clear-cut, financial incentives to hire disadvantaged groups can be a means to strengthen the business case for employers under certain conditions. Similarly, public procurement regulations that promote supplier diversity can be an important policy tool, seeing that OECD countries spent around 12% of GDP on goods and services through public procurement in 2016 (OECD, 2017[34]).

Using public procurement as a strategic tool to strengthen diversity and equal treatment is a fairly widespread policy approach in OECD countries (see Figure 3.3), mostly with regard to incentivising the hiring of women and employees with disabilities. Some governments give preference to firms in cases of tie-break between applications for public funds, e.g. in Latvia, others grant extra points in point-based contract award systems, as is the case in Japan and Mexico. In addition, many EU OECD countries, where public procurement is regulated by both domestic and EU law, follow the clauses for “social considerations” stipulated in the EU public procurement Directive of 2014 (2014/24/EU). These allow Member States the option to consider participation of vulnerable and disadvantaged people in the award of contracts as well as the possibility to reserve procurement procedures to social enterprises working for the inclusion of disadvantaged people.

Some countries have stronger and more concrete obligations in this regard. Australia has set a target in procurement – the Commonwealth Indigenous Procurement Policy requires that by 2020, 3% of Federal Government contracts each year be sourced from indigenous suppliers. In Canada, all contractors must sign an Agreement to Implement Employment Equity within 30 days of contract award. As per the agreement, firms are required to make reasonable efforts to ensure progress is made towards having full representation of women, disabled people, and members of visible minorities and indigenous people within their workforce. Progress is reviewed through regular compliance assessments by the Labour Program. Failure to adhere to the requirements can result in the loss of the current contract and/or in the contractor losing the right to bid on future contracts.

Governments using public procurement to support diversity objectives should, however, also consider the risk of disrupting the efficiency of public procurement. The expenses and trade-offs, if they exist, must be evaluated, assessing whether procurement is the most cost-effective way to achieve specific diversity objectives (OECD, 2017[35]; 2013[36]).

Subsidising wages in the employment of under-represented groups is another means by which OECD governments seek to strengthen the business case for diversity with regard to workers with disabilities, as well as migrant and older workers. There are two types of wage subsidies: Hiring subsidies are granted temporarily, whereas employment subsidies are provided over an extended period of time or indefinitely. While the latter assumes a permanent productivity gap, temporary subsidy schemes that phase-out after time follow the rationale that on-the-job training of certain groups takes more time, but that the productivity gap can eventually be closed. Temporary subsidies can also reduce the initial insecurity with regard to the actual productivity, lowering the costs that a “trial phase” imposes on employers. Across OECD countries, subsidies vary in type (wages and/or social security contributions) and the amount of labour cost compensated.

Wage subsidies are a long-standing component of active labour market policies to strengthen the employment of low-wage workers, long-term unemployed and young job seekers in OECD countries. A rich literature on the impact of such subsidies on the employment of disadvantaged groups has developed, highlighting the difficulties of getting incentives ‘right’. Issues with wage subsidies include risks of stigmatisation, substitution and displacement effects as well as dead-weight losses, i.e. providing subsidies to firms that would have hired workers without these financial incentives.

The effectiveness of subsidies for older workers has been investigated through studies of such schemes in Germany, Finland and Belgium. The evidence suggest that for these groups, the deadweight effect is high, meaning that employers’ decisions are unchanged in comparison to a hypothetical situation in which the subsidy is not paid. It also appears that generally, subsidised work simply displaces unsubsidised work, with little net gain in terms of employment (Boockmann, 2015[37]).

Although rarely explicitly targeting migrants, wage subsidies have proven effective for this group. A meta-analysis by Butschek and Walter (2014[38]) on interventions in Nordic countries, Germany, the Netherlands and Switzerland finds that effect estimates for wage subsidy programmes are mostly positive, suggesting that such subsidies are indeed a promising measure to increase employment rates of immigrants. Migrants, however, tend to be underrepresented among beneficiaries of such schemes, possibly because they themselves or employers do not know about their eligibility (OECD, 2014[39]).

Other studies on subsidies for workers with disabilities show that some of the general risks around active labour market policy, e.g. regarding deadweight or substitution effects, is less relevant for this group, because of the permanent productivity loss of some groups of people with disabilities (OECD, 2003[40]). Especially for this group, schemes must consider that persons with disabilities are a very heterogeneous group. People have different types of disabilities and these can be more or less severe and can be acquired at birth or later in life. Persons with disabilities also vary in other demographic characteristics to which the subsidies should be well-targeted. For example, age seems to be an important factor determining if subsidies are successful or counterproductive and stigmatising (Deuchert and Kauer, 2017[41]). The level of subsidy should be dependent on the capacity of the worker as well as flexible, as work capacity of persons with disabilities may also change over time. On the other hand, subsidies might be needed for a long period and even a permanent subsidy might be justified in some cases (OECD, 2010[42]).

OECD countries also offer financial compensation and rewards to firms for taking action beyond hiring disadvantaged groups. One such instrument are accommodation subsidies. Slovakia, for example, provides cash allowances for firms that wish to establish sheltered workplaces, hire a personal assistant for a disabled employee, or facilitate their transport. Another type of incentives seeks to encourage disadvantaged groups’ career development. Japan, for instance, offers subsidy schemes providing a financial reward to employers that achieve a self-imposed numerical target for the advancement of female employees set in action plans. A majority of countries also provides non-financial incentives (see Box 3.4).

The importance of networks in finding employment is well established; social and professional networks may not only help for being referred and making personal connections, but also for obtaining information on job openings in the first place. Groups traditionally disadvantaged in the labour market may not have these connections. For example, people with a migrant background and ethnic minorities tend to have fewer contacts with people in higher social positions (Li, Savage and Warde (2008[43]) for the United Kingdom); get fewer job leads (McDonald, Lin and Ao (2009[44]) for the United States); and receive less help from their social network when applying for apprenticeships (Beicht and Granato (2010[45]) for Germany). Thus, an important component of diversity policies is to mitigate this initial disadvantage and increase the pool of qualified candidates through proactive outreach measures. In a number of OECD countries, such campaigns have been developed, often on a city or regional level, specifically targeting candidates with a migration background (see Box 3.5).

In addition, outreach to underrepresented groups can address barriers that are not only related to a lack of networks, but also to gender stereotypes about what are perceived as typically male and female jobs, leading to fewer candidates in STEM jobs for women and education and care for men. For instance, 5.5% of male workers in OECD countries are ICT specialists, compared to only 1.4% of female workers. At the same time, women remain strongly over-represented in the education and care sector; among primary school teachers, more than four in five are women (OECD, 2017[46]).

Difficulties in increasing the applicant pool also became apparent in the findings of the HR survey. When asked about the main difficulties faced while implementing diversity measures, the most common obstacle named was that too few qualified candidates from disadvantaged groups applied (47%).

The majority of HR representatives believe that outreach campaigns would help their company increase diversity (see Figure 3.4). Nevertheless, only 20% of all firms surveyed indicated that they have implemented outreach or recruitment campaigns targeting under-represented groups. Partly, this reflect that outreach campaigns are costly to implement and may not be feasible for smaller companies; of the firms with diversity measures, only 20% of firms with less than 250 employees had outreach measures in place, compared to 40% among firms with over 250 employees.

Many public administrations in OECD countries have implemented outreach campaigns and traineeship programmes targeting specific groups with the aim of building a workforce that is more representative of the public they serve.

Most outreach campaigns for the public sector tend to be concentrated in education, targeting pupils and recent or soon-to-be university graduates. In Australia, specific programmes seek to encourage Aboriginal and/or Torres Strait university graduates, as well as graduates with a disability into the Australian Public Service (APS). Similarly, the Swiss public administration cooperates with specific universities to recruit more women. In some countries, such programmes begin as early as during secondary education. In Estonia, campaigns in Russian language schools seek to encourage pupils of Russian nationality to apply for the public service. The French public service offers 15 000 internships in the public administration to students around age 15 who attend schools with high shares of disadvantaged students (schools in the réseau d’éducation prioritaire renforcé).

Recruiters for the public administration use a wide range of channels to reach diverse groups. The Irish Public Appointments Service, for example, has engaged in a number of activities to communicate job opportunities to underrepresented groups over social media including Facebook, Twitter and LinkedIn, targeting a range of nationalities and Travellers, LGBT people, disabled people and Irish language groups. In the Netherlands, public authorities work with professional migrant or diversity networks and student associations to encourage highly educated young people with a non-Western migration background to apply for jobs in the public administration. Integrating such local stakeholders into efforts to approach hard-to-reach groups is crucial, especially with regard to youth from migrant families (OECD, forthcoming[47]).

Anonymous or “blind” job applications are aimed at reducing bias in the application process by removing information that could activate recruiters’ stereotypes, such as ‘foreign-sounding’ names, gender, age or address. Used only by 6% of firms that have diversity measures in place, anonymous applications were not a common instrument in the HR survey sample. Those firms that did hide characteristics in the hiring process, most often removed an applicant’s age, followed by their name and address (see Figure 3.5). Another commonly masked characteristic was civil or marital status. Almost 19% of those firms implementing diversity measures stated that their application process does not require a photograph of the candidate, despite this being common practice within the industry or country concerned.

In several European countries including Sweden, France, the Netherlands and Germany, the effects of anonymous job applications have been tested in field experiments. The French government has conducted an experiment with jobseekers registered with the public employment service in 2010 and 2011 involving around 1 000 firms for a period of ten months. The result was that women benefited from higher call-back rates. Surprisingly, the second finding was that immigrants and residents of deprived neighbourhoods suffered from anonymous job applications, as their call-back rates were lower with anonymous job applications than with standard applications (Behaghel, Crépon and Le Barbanchon, 2011[49]). The gap in interview rate between minority and majority candidates widened by around 10 percentage points when CVs were anonymised. The effect seemed to persist beyond the application screening stage, as the hiring gap also widened by four points (only significant at the 10% level).

A second experiment was conducted by the Swedish city of Gothenburg between 2004 and 2006, which introduced a pilot project to test the impact of anonymous CVs on hiring outcomes. Based on a difference-in-differences analysis of around 3 500 applications made to around 100 job openings in the local administration, Åslund and Skans (2012[50]) find that anonymous job applications increased the chances of an interview invitation for both women and applicants of non-Western origin when compared to standard applications, by the same order of magnitude for both groups. These increased chances for candidates in the first stage translated into a higher job offer rate for women, but not for migrants, suggesting that barriers at the interview stage are more difficult to remove for the latter group. The authors speculate that the high importance placed on gender equality in the Swedish context may have influenced the findings.

Studies on further trials, including small-scale pilots within single organisations, did not yield statistically significant effects. Nevertheless, they give indications of how diverse the effects of anonymous applications can be depending on the groups concerned (see for example Krause, Rinne and Zimmermann (2011[51])). A common explanation for the contradicting findings in the literature is the crucial importance of the initial situation in the respective industry or organisation. While effective in the face of bias and discriminatory practices, anonymity prevents employers committed to diversity from favouring minority applicants when credentials are equal. In situations where such affirmative action practices are in place, the introduction of anonymous job applications may cause interview invitation probabilities of minority groups to decrease. Anonymous applications might be helpful with regard to biases, but unproductive when it comes to structural disadvantages, since they remove characteristics that could serve to explain subpar labour market outcomes from the process.

Current evidence does therefore not support the introduction of anonymous job applications in every context. Anonymous hiring might be sensible in a specific sector or for a certain company where there is sizeable discrimination, but be futile where there is no discrimination, and even counterproductive where practices of favouring women or minorities are in place. This finding relates to the limitation of the studies discussed above: the firms and organisations in these experiments were participating on a voluntary basis. There has likely been firm selection into such programmes by those already concerned with diversity issues, who are more likely to already have affirmative action in place, which might explain some of the negative results.

Removing bias from the recruitment process has also been a topic of interest in the context of data-driven HR practices. HR analytics aim to use statistical (data mining) techniques to predict uncertain outcomes, i.e. the performance of potential employees, and monitor the success of recruitment and retention. Through predictive analytics in recruitment, HR can draw upon data-driven predictive models, instead of relying on gut feeling and soft science. 

Such new technologies are already impacting the labour market, with the potential to counteract discrimination and unconscious bias in hiring. Two recent studies provide experimental evidence on the impact of algorithms on the probability that atypical job candidates are hired. In both cases the software favoured “non-traditional” candidates, in contrast to human screeners. The algorithm exhibited significantly less bias against candidates that were underrepresented at the firm, such as those without personal referrals or degrees from prestigious universities (Cowgill, 2018[52]; Hoffman, Kahn and Li, 2018[53]). Another recent study finds that compared to those candidates chosen by humans, board directors selected by algorithms were less likely to be male and had smaller networks (Erel et al., 2018[54]).

However, it must be kept in mind that algorithms are not necessarily free of bias, as they are written and maintained by people, and machine learning algorithms adjust what they do based on people’s behaviour. Although not designed with the intent of discriminating against certain groups, algorithms are likely to incorporate historical prejudices if the training data reflect existing social biases against a minority (Goodman, 2016[55]). Nevertheless, the more relevant question for practitioners and policy makers concerned with diversity and anti-discrimination is to ask how algorithms compare with the status quo – and the evidence suggests that HR analytics has the potential to make some decisions exhibiting less bias than humans do.

Creating an inclusive workplace is a crucial step in making the most of diversity. All efforts put into hiring diverse candidates are fruitless if these workers cannot flourish or end up leaving the organisation for feeling excluded at the workplace. To this end, diversity training encompasses a wide array of programmes intended to train managers’ and employees’ attitudes, reduce stereotypes and enhance their skills in dealing with diversity. Such trainings were found to be in widespread use in the HR survey – 44% of firms with measures in place have implemented trainings for all employees, and 20% have specific training for HR personnel. The overwhelming majority of HR representatives surveyed stated that a wider availability of training could help their company increase diversity (see Figure 3.6). In the academic literature, diversity training has generally been found to have no impact on the representation of disadvantaged groups (Kalev, Dobbin and Kelly, 2006[12]), while the impact on attitudes towards diversity varies from beneficial in some contexts to no or a relatively modest negative effect in others (Kulik and Roberson, 2008[56]).

Kulik et al. (2007[57]) find that voluntary diversity training likely attracts those employees who need training the least. Because employees who are sceptical towards diversity are usually not able to gauge their own competence in dealing with diversity, they assess their need to develop those skills inadequately. This means that voluntary training often misses those employees with lowest skill competencies. While pre-training competence of (potential) participants matters, demographics have non-significant effects on interest in training and participation. This means that women, ethnic minorities and older employees do not seem more open towards voluntary training on diversity issues than are other employees.

Implementing mandatory diversity training may avoid ‘preaching to the choir’, yet it is unclear whether they are more effective, as external control and perceived societal pressure were found to have a negative effect on training outcomes (Legault, Gutsell and Inzlicht, 2011[58]).

Several studies examine how training should be designed to affect participants’ attitude and skill outcomes positively, showing that diversity training needs to stimulate intrinsic motivation rather than framing it around social norms. Legault et al. provided participants with instructions that either encouraged autonomous motivation to regulate prejudice by emphasising the value of non-prejudice, or imposed control of prejudice, framing it as a societal requirement to comply with norms. The results show that external control to reduce prejudice produces more bias than not intervening at all. (Kulik, Perry and Bourhis, 2000[59]) find that under certain circumstances, instructing participants to suppress age-related stereotypes activated them in the first place, thus creating unintended negative consequences for the evaluation of older applicants. Similarly, instructing participants to simply discard their bias as “wrong” without adequately addressing and deconstructing it seems to leave participants more confused and more biased towards diverse groups (Dobbin and Kalev, 2018[60]).

The duration of training is another important factor. To be effective in the long-run, training must be sustained over an extended period, as effects are found to dissipate within short time periods, often within days (Paluck and Green, 2009[61]). Devine et al. (2012[62]) develop a multi-faceted prejudice habit-breaking intervention to produce long-term reductions in implicit race bias. The 12-week longitudinal study comparing a group of people who completed the intervention to a control group who did not showed that sustained training leads to increased concern about discrimination and personal awareness of bias. For most participants, the effect on implicit bias control lasted throughout the eight week interval observed in the aftermath of the intervention. Such repeated, longer-term trainings, however, are a costly undertaking that firms are unlikely to implement.

In view of the volatile nature of training effects, diversity training can only be one element of a wider set of policies. As a one-time, standalone measure, it is unlikely to have any substantial impact on attitudes. Diversity training might get the conversation on existing bias started, but must be combined with other, more structural measures (Dobbin and Kalev, 2018[60]) suggest that firms must actively engage managers in reducing bias within their workforce, including them as “diversity champions” in an overall positive message. The close interaction with women and minorities in mentoring and college recruitment schemes, for example, seem to be more effective in reducing managers’ bias than diversity training.


[7] Acemoglu, D. and J. Angrist (2001), “Consequences of Employment Protection? The Case of the Americans with Disabilities Act”, Journal of Political Economy, Vol. 109/5.

[50] Åslund, O. and O. Skans (2012), “Do Anonymous Job Application Procedures Level the Playing Field?”, ILR Review, Vol. 65/1, pp. 82-107.

[3] Balestra, C. and L. Fleischer (2018), “Diversity statistics in the OECD: How do OECD countries collect data on ethnic, racial and indigenous identity?”, OECD Statistics Working Papers, No. 2018/09, OECD Publishing, Paris, https://dx.doi.org/10.1787/89bae654-en.

[49] Behaghel, L., B. Crépon and T. Le Barbanchon (2011), Evaluation de l’impact du CV anonyme, Paris School of Economics.

[45] Beicht, U. and M. Granato (2010), Ausbildungsplatzsuche: Geringere Chancen für junge Frauen und Männer mit Migrationshintergrund – BIBB-Analyse zum Einfluss der sozialen Herkunft beim Übergang in die Ausbildung unter Berücksichtigung von Geschlecht und Migrationsstatus, Bundesinstitut für Berufsbildung, BIBB Report: Forschungs- und Arbeitsergebnisse aus dem Bundesinstitut für Berufsbildung, No. 15.

[26] Bertrand, M. and S. Mullainathan (2004), “Are Emily and Greg More Employable Than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination”, American Economic Review, Vol. 94/4, pp. 991-1013.

[37] Boockmann, B. (2015), “The effects of wage subsidies for older workers”, No. 189, IZA World of Labor.

[25] Bunel, M. and P. Petit (2016), “Discrimination based on place of residence and access to employment”, Urban Studies Journal Limited, Vol. 53/2, pp. 267-286.

[38] Butschek, S. and T. Walter (2014), “What active labour market programmes work for immigrants in Europe? A meta-analysis of the evaluation literature”, IZA Journal of Migration, Vol. 48/3.

[32] Cancian, M. (1998), “Race‐based versus class‐based affirmative action in college admissions”, Journal of Policy Analysis and Management, Vol. 17/1, pp. 94-105.

[33] Carnevale, A. and S. Rose (2003), “Socioeconomic Status, Race/Ethnicity, and Selective College Admissions”, Century Foundation.

[14] Carrington, W., K. McCue and B. Pierce (2000), “Using Establishment Size to Measure the Impact of Title VII and Affirmative Action”, The Journal of Human Resources, Vol. 35/3, pp. 503-525.

[15] Chay, K. (1998), “The Impact of Federal Civil Rights Policy on Black Economic Progress: Evidence from the Equal Employment Opportunity Act of 1972”, ILR Review, Vol. 51/4, pp. 608-632.

[9] Cialdini, R. and N. Goldstein (2004), “Social Influence: Compliance and Conformity”, Annual Review of Psychology, Vol. 55/1, pp. 591-621.

[52] Cowgill, B. (2018), “Bias and Productivity in Humans and Algorithms: Theory and Evidence from Résumé Screening”, Columbia University Working Paper.

[1] Crenshaw, K. (1989), “Demarginalizing the Intersection of Race and Sex: A Black Feminist Critique of Antidiscrimination Doctrine, Feminist Theory and Antiracist Politics”, Univesity of Chicago Legal Forum, Vol. 1, Article 8.

[41] Deuchert, E. and L. Kauer (2017), “Hiring subsidies for people with a disability: Evidence from a small-scale social field experiment”, International Labour Review, Vol. 156/2, pp. 269-285.

[62] Devine, P. et al. (2012), “Long-term reduction in implicit race bias: A prejudice habit-breaking intervention.”, Journal of Experimental Social Psychology, Vol. 48/6, pp. 1267-1278.

[60] Dobbin, F. and A. Kalev (2018), “Why Doesn’t Diversity Training Work? The Challenge for Industry and Academia”, Anthropology Now, Vol. 10/2, pp. 48-55.

[5] Donohue, J. (2005), “The Law and Economics of Antidiscrimination Law”, Yale Law & Economics Research Paper, No. 318.

[54] Erel, I. et al. (2018), “Selecting Directors Using Machine Learning”, NBER Working Paper, No. w24435.

[48] Ghelli, F. and J. Pross (2019), “Interkulturelle Öffnung. Die Polizei wird vielfältiger”, Mediendienst Integration, https://mediendienst-integration.de/artikel/die-polizei-wird-vielfaeltiger.html.

[55] Goodman, B. (2016), Economic Models of (Algorithmic) Discrimination.

[17] Groeneveld, S. and S. Verbeek (2012), “Diversity Policies in Public and Private Sector Organizations: An Empirical Comparison of Incidence and Effectiveness”, Review of Public Personnel Administration, Vol. 32/4, pp. 353-381.

[30] Harris, A. and M. Tienda (2010), “Minority Higher Education Pipeline: Consequences of Changes in College Admissions Policy in Texas”, The ANNALS of the American Academy of Political and Social Science, Vol. 627/1, pp. 60-81.

[10] Heath, Liebig and Simon (2013), Discrimination against immigrants – measurement, incidence and policy instruments, OECD Publishing, Paris, https://dx.doi.org/10.1787/migr_outlook-2013-en.

[53] Hoffman, M., L. Kahn and D. Li (2018), “Discretion in Hiring”, The Quarterly Journal of Economics, Vol. 133/2, pp. 765-800.

[21] Holzer, H. and D. Neumark (2000), “What does affirmative action do?”, Industrial and Labor Relations Review, Vol. 53/2, pp. 240-271.

[19] Horst, E. and K. Wrohlich (2018), “Spitzengremien großer Unternehmen: Geschlechterquote für Aufsichtsräte greift, in Vorständen herrscht beinahe Stillstand”, Deutsches Institut für Wirtschaftsforschung: Managerinnen-Barometer: Unternehmen.

[23] Kahlenberg, R. (2012), A Better Affirmative Action: State Universities that Created Alternatives to Racial Preferences, Century Foundation.

[12] Kalev, A., F. Dobbin and E. Kelly (2006), “Best Practices or Best Guesses? Assessing the Efficacy of Corporate Affirmative Action and Diversity Policies”, American Sociological Review, Vol. 71, pp. 589-617.

[31] Kane, T. (1998), “Racial and Ethnic Preferences in College Admissions”, Ohio State Law Journal, Vol. 59/3, pp. 971-996.

[51] Krause, A., U. Rinne and K. Zimmermann (2011), “Anonymous Job Applications of Fresh Ph.D. Economists”, No. 6100, IZA Discussion Paper.

[57] Kulik, C. et al. (2007), “The rich get richer: predicting participation in voluntary diversity training”, Journal of Organizational Behavior, Vol. 28/6, pp. 753-769.

[59] Kulik, C., E. Perry and A. Bourhis (2000), “Ironic Evaluation Processes: Effects of Thought Suppression on Evaluations of Older Job”, Journal of Organizational Behavior, Vol. 21/6, pp. 689-711.

[56] Kulik, C. and L. Roberson (2008), “Common Goals and Golden Opportunities: Evaluations of Diversity Education in Academic and Organizational Settings”, Academy of Management Learning & Education, Vol. 7/3, pp. 309-331.

[58] Legault, L., J. Gutsell and M. Inzlicht (2011), “Ironic Effects of Antiprejudice Messages: How Motivational Interventions Can Reduce (But Also Increase) Prejudice”, Psychological Science, Vol. 22/12, pp. 1472-1477.

[11] Leonard, J. (1990), “The Impact of Affirmative Action Regulation and Equal Employment Law on Black”, The Journal of Economic Perspectives, Vol. 4/4, pp. 47-63.

[16] Leonard, J. (1985), “What Promises Are Worth: The Impact of Affirmative Action Goals”, Source: The Journal of Human Resources, Vol. 20/1, pp. 3-20.

[43] Li, Y., M. Savage and A. Warde (2008), “Social mobility and social capital in contemporary Britain”, British Journal of Sociology, Vol. 59/3, pp. 391-411.

[22] Maida, A. and A. Weber (2019), “Female Leadership and Gender Gap within Firms: Evidence from an Italian Board Reform”, Discussion Paper Series, IZA Institute of Labor Economics.

[4] Makkonen, T. (2016), European handbook on equality data 2016 revision, Publications Office of the European Union, Luxembourg.

[44] McDonald, S., N. Lin and D. Ao (2009), “Networks of opportunity: Gender, race, and job leads”, Social Problems, Vol. 56/3, pp. 385-402.

[13] Miller, C. (2017), “The Persistent Effect of Temporary Affirmative Action”, American Economic Journal: Applied Economics, Vol. 9/3, pp. 152-190.

[28] Nunn, A. et al. (2010), Postcode selection? Employers’ use of area-and address-based information shortcuts in recruitment decisions, Department for Work and Pensions.

[24] OECD (2018), A Broken Social Elevator? How to Promote Social Mobility, OECD Publishing, Paris, http://dx.doi.org/10.1787/9789264301085-en.

[34] OECD (2017), Government at a Glance 2017, OECD Publishing, Paris, https://dx.doi.org/10.1787/gov_glance-2017-en.

[35] OECD (2017), “OECD Framework to Promote the Strategic Use of Public Procurement for Innovation”, in Public Procurement for Innovation: Good Practices and Strategies, OECD Publishing, Paris, https://dx.doi.org/10.1787/9789264265820-6-en.

[46] OECD (2017), The Pursuit of Gender Equality. An Uphill Battle, http://dx.doi.org/10.1787/9789264281318-en (accessed on 5 October 2017).

[39] OECD (2014), International Migration Outlook 2014, OECD Publishing, Paris, https://dx.doi.org/10.1787/migr_outlook-2014-en.

[36] OECD (2013), “A strategic role for public procurement”, in Implementing the OECD Principles for Integrity in Public Procurement: Progress since 2008, OECD Publishing, Paris, https://dx.doi.org/10.1787/9789264201385-7-en.

[42] OECD (2010), Sickness, Disability and Work: Breaking the Barriers: A Synthesis of Findings across OECD Countries, OECD Publishing, Paris, https://dx.doi.org/10.1787/9789264088856-en.

[6] OECD (2008), OECD Employment Outlook 2008, OECD Publishing, Paris, https://dx.doi.org/10.1787/empl_outlook-2008-en.

[40] OECD (2003), Transforming Disability into Ability: Policies to Promote Work and Income Security for Disabled People, OECD Publishing, Paris, https://dx.doi.org/10.1787/9789264158245-en.

[47] OECD (forthcoming), Making Integration Work: Young people with a migrant background, OECD Publishing, Paris.

[61] Paluck, E. and D. Green (2009), “Prejudice Reduction: What Works? A Review and Assessment of Research and Practice”, Annual Review of Psychology, Vol. 60/1, pp. 339-367.

[29] Sowell, T. (2004), Affirmative action around the world: An empirical study, Yale University Press.

[18] Storvik, A. and M. Teigen (2010), Women on board - the Norwegian experience, Friedrich Ebert Stiftung.

[8] Tankard, M. and E. Paluck (2017), “The Effect of a Supreme Court Decision Regarding Gay Marriage on Social Norms and Personal Attitudes”, Psychological Science, Vol. 28/9, pp. 1334-1344.

[27] Tunstall, R. et al. (2014), “Does poor neighbourhood reputation create a neighbourhood effect on employment? The results of a field experiment in the UK”, Urban Studies, Vol. 51/4, pp. 763-780.

[20] Valfort, M. (2018), “Do anti-discrimination policies work?”, No. 450, IZA World of Labor.

[2] Valfort, M. (2015), “Religious discrimination in access to employment: a reality”, No. October 2015, Institut Montaigne.

Metadata, Legal and Rights

This document, as well as any data and map included herein, are without prejudice to the status of or sovereignty over any territory, to the delimitation of international frontiers and boundaries and to the name of any territory, city or area. Extracts from publications may be subject to additional disclaimers, which are set out in the complete version of the publication, available at the link provided.


© OECD 2020

The use of this work, whether digital or print, is governed by the Terms and Conditions to be found at http://www.oecd.org/termsandconditions.