I recently commented on the UK report “Commission on Race and Ethnic Disparities”, which dared to suggest that systemic racism was not the major cause of race differences in Britain. The wave of criticism continues. One line of attack is that the Commission did not give sufficient weight to studies which show that racial minorities are not invited for job interviews at the same rate as locals. In fact, the commissioners did discuss the issue, and are in favour of name-blind job applications. They discuss UK research on this matter at page 121 in the report, and to their credit they do not regard the results as conclusive of widespread bias. However, the methods do seem to indicate bias, so I thought I would look at the conventional way of researching the issue of racial discrimination in job applications, and contrast it with anonymous job applications.
To see whether racial discrimination exists, researchers send the same CV to employers with the same level of qualifications but different names attached, to see if the foreign-sounding names lead to a greater degree of rejection. They often find that to be the case.
Other alterations are carried out on these identical CVs to see whether other aspects, like religion, country of origin and so on are also a cause of rejection. The different rejection rates are then cited as evidence of racism. That might be the case, but there are other possible explanations.
Is there anything wrong with this method?
One thing: it assumes that different racial or religious group members write CVs which are just as good as anyone else. What if one group is over-confident, and boasts in a way which turns off employers? The effects of that boasting would not show up with this method. What if one group is not very organised, and their CVs reveal that? Again, that would not be detected if the standard CV method is used. Indeed, the “standard CV” method is designed to keep all those real-life effects out of the picture.
Does putting a foreign-sounding name on a standard CV really mean that applicants are rejected unfairly? Not quite. Employers may have actually found that some groups make poor employees. They may have learned that qualifications from some universities give an exaggerated picture of skills. They may want to take a chance with a particular minority, but have found that they have been disappointed with the candidates they chose. Furthermore, depending on the laws of the land, they may find that minorities are better placed in law to sue employers for discrimination than would be the case for majority locals. The latter cannot easily claim racial discrimination against them, and will be judged simply by their work record. A minority candidate, plausibly or implausibly, can claim that he was not treated fairly, not given chances open to others, not promoted as quickly as others, and so on.
Is there a better method of judging whether applicants are subject to racial bias from employers? Of course. Get people from the relevant minority groups to write actual job applications themselves, and then send them in to different employers under different names or anonymously, to see whether the employers are responding to the names or to the actual quality of the applications. This technique is usually called anonymous application procedures (AAP) and it is intended to prevent racial bias against women and minority candidates.
Åslund, O., & Skans, O. N. (2012). Do Anonymous Job Application Procedures Level the Playing Field? ILR Review, 65(1), 82–107. doi:10.1177/001979391206500105
This was a large-scale study of a trial of anonymous applications to local authority positions in Sweden. Crucially “Information on schools/universities was explicitly prohibited since it would reveal the ethnicity of many immigrant applicants.” Of course, it would remove any data about the quality of education received, and its relevance to the skills required in the advertised jobs. Educational standards, as shown by PISA type studies, differ considerably in different countries, and are usually lower than Western standards. Real data will be lost here.
The authors found that anonymous application procedures led to more interviews than conventional applications which disclosed sex and race. However, once the applicants from the anonymous method got to the interview stage, they did not do so well.
For women, the results concur with Goldin and Rouse’s (2000) finding of a positive effect on the final outcome of anonymous hiring procedures. However, we fail to find a corresponding effect for the non-Western applicants. In fact, the job offer differentials relative to other applicants are nearly identical under the different regimes. Although the statistical uncertainty is substantial, the results, if taken at face value, mean that the positive effects on interview offers are undone once origin, personal traits, and full credentials are disclosed.
One interpretation is that, once the mask of anonymity is taken away, then prejudicial attitudes come into play at interview. Another interpretation is that in a wide-ranging interview the minority candidates perform poorly, and lack the wider skills required. The authors lean to the first interpretation, but do mention that minorities may lack the “social capital” possessed by native Swedes. In my view the lack of school data in the anonymous applications could have been a big factor in later poor performance at interview, where candidates must think on their feet about job-related demands and problems.
Krause, A., Rinne, U. & Zimmermann, K.F. Anonymous job applications in Europe. IZA J Labor Stud 1, 5 (2012). https://doi.org/10.1186/2193-9012-1-5 This is a review of European experience with anonymous job applications.
The French government initiated an experiment in 2010 and 2011 which was implemented by the French public employment service. It involved about 1,000 firms in eight local labor markets and it lasted in total for about ten months (Behaghel et al., 2011). The experiments’ main findings can be summarized as follows. First, women benefit from higher callback rates with anonymous job applications—at least if they compete with male applicants for a job. However, for roughly half of the vacancies included in the experiment only female candidates or only male candidates applied. Second, migrants and residents of deprived neighborhoods suffer from anonymous job applications. Their callback rates are lower with anonymous job applications than with standard applications.
Why do minorities suffer when applications are anonymous? They should shine in such circumstances. The answer seems to lie in them not getting a sympathetic interpretation of their circumstances and their lower scholastic and occupational attainments:
Besides, context-specific information may be interpreted differently if information about the identity of the candidate is not available—and this can result in disadvantages for the applicant. For example, if recruiters are not aware of the applicant’s family situation, migration background or disadvantaged neighborhood, this information cannot be taken into account to explain, e.g., below-average education outcomes, labor market experience or language skills.
Paradoxically, in Europe a migrant may get the benefit of the doubt for low school grades, which would not be accorded to locals.
In the Netherlands, two experiments took place in the public administration of one major Dutch city in 2006 and 2007. The experiments focus on ethnic minorities. More specifically, a distinction is made between applicants with and without foreign (i.e., non-Western) sounding names.
Bøg and Kranendonk (2011) emphasize in their study the lower callback rates for minority candidates with standard applications, but their analysis also reveals that these differences disappear with anonymous job applications. With regards to job offers, however, the authors do not detect any differences between minority and majority candidates—irrespective of whether or not their resumes are treated anonymously. This indicates that even with standard applications, discrimination against minorities in interview invitations disappears at the job offer stage.
They also review a more detailed experiment in Germany.
Both, the results of the various European experiments and of the German experiment predominantly show that anonymous job applications can lead to the desired effect of increasing the interview invitation probabilities of disadvantaged groups. However, there are indications for exactly the opposite effect, namely that anonymity prevents employers from favoring minority applicants. In particular, our analysis of the heterogeneous data from the German experiment shows that the initial situation is crucial. Three different conditions can initially exist: discrimination, affirmative action, and equality of opportunity. Not surprisingly, the effects of anonymous job applications are as heterogeneous as the initial situation to be changed. This result is in line with findings from the various European experiments. It often appears that the introduction of anonymous job applications is beneficial for a particular minority group in a given experiment, whereas another minority group does not benefit to the same extent—although the setting is the same.
A study in Australia comes to similar conclusions, and shows that the effects of anonymous job applications partly depend on whether employers have a prior disposition to see women or minorities unfavourably. If they don’t, then anonymous applications can stop those with positive attitudes from implementing them via affirmative action (which is itself a bias, I think).
The authors explain how they set up their research project (page 9):
There are many potentially irrelevant characteristics that could be screened out from reviewers in order to remove biases. Besides gender, race or ethnic status, we might also consider any information about age, health status or conditions, disability, sexual orientation, political views, and socioeconomic status (reflected for instance, by address and education background).
Now let us put that into clear text: “There are many potentially relevant characteristics which employers would like to know about: gender, race, age, health, disability, sexual orientation, politics and economic success, so as to reduce their risks when hiring someone. They want to choose whom to hire. We are here to stop them”.
The researchers found that, when actual names were used, assessors were slightly more, not less, likely to employ women and minority candidates. There was a 6% advantage for women when people could show their names, and also an advantage for minorities when they could reveal their names, but not significantly so. One reason is that the reviewers were very much in favour of short-listing the indigenous Australian women (up 22%).
To my mind this shows that the reviewers who volunteered to participate in this study about bias were probably those who were biased in favour of women and minorities. At the very least, there was no strong support for anonymous applications, it would seem.
Overall, the results indicate the need for caution when moving towards ’blind’ recruitment processes in the Australian Public Service, as de-identification may frustrate efforts aimed at promoting diversity.
One of the pleasures of research is that an investigation of bias against immigrants can reveal biases in favour of immigrants.
If we look at employment with a wider perspective, here is a little thought experiment. Imagine a world in which employers recruited the candidates they wanted to employ. In such a world, employers would compete for the best candidates, and would use every single datum to help them make good decisions. If employers made any false positive mistakes, they would burden themselves with failures of selection (under-performance, costs of further training, costs of terminating contracts, and massive opportunity costs). If they made false negative mistakes, they would miss out on superb performers, and would let rivals pick them up at favourable prices, eventually risking their own company’s success in the market place.
In seeking the most suitable employees, employers would certainly look at group differences, because when there are big differences in ability and behaviour then those affect the interpretation placed on all the measures, including IQ measures. The reason is that if you set a cut-off of say IQ 120 for your recruits then you must expect that some of the results will have a chance element to them. Actual scores can be conceived of as being composed of a true score (one which could be determined by several different tests taken over a period of time) and a chance element, corresponding to luck. Not only can candidates select which achievements to talk about (all of which may have a luck element) but in some circumstances they can give results from re-examinations. Employers would like a margin of safety to guard against flukes.
Consider 3 candidates, all with an achieved score of IQ 120, but drawn from three different genetic groups A, B, C which have mean IQs of 110, 100 and 85 respectively.
If drawn from Group A of mean IQ 110, then 25.24% are above that level. Big margin of safety. The person’s score is very unlikely to be a fluke, so he is safe to hire, and 4 times safer than a Group B hire.
If drawn from Group B of mean IQ 100 then 9.12% of the group above that level. Fluke unlikely, so pretty safe. 9 times safer than a Group C person.
If drawn from Group C of mean IQ 85 then only 1% of that group above that level. Risky. Not much margin of safety.
Incidentally, if the standard deviation for Group C is not 15 but 13, then there will be only 0.35% above the IQ 120 cut-off and it will be 25 times safer to pick someone of IQ 120 from Group B.
A final observation is that the old standby, identical CVs with different names attached, seems to confirm biases which account for different outcomes in occupational success. On closer examination, that technique adopts the null hypothesis that different racial groups are the same in ability and employment potential, but employer prejudice blights their prospects. Employers already know that groups differ, but aren’t allowed to make use of that fact, which would considerably simplify their hiring processes. In fact, in some cases recruiters are in favour of affirmative action, so their preference for that approach is thwarted by anonymous applications. On further examination, granting anonymity to actual job applications reveals that the real crunch often comes at interview.
Fascinating, that a technique intended to detect bias in others reveals an unconscious bias in the researchers, in the form of a persistent blank-slatism. Whatever the actual results, these researchers keep wanting to assume that there are no group differences.
There should be a compulsory course for all researchers, to warn them about the dangers of this egregious bias bias.