Judul: Discriminating in hiring on the basis of statistics: Conceptual and ethical issues
Penulis: Vanessa Scholes
Cite as: Scholes, V (2011). Discriminating in hiring on the basis of statistics: Conceptual and ethical issues. Unpublished manuscript. Retrieved from: https://www.academia.edu/1327603/Discriminating_in_hiring_on_the_basis_of_statistics_Conceptual_and_ethical_issues
Discriminating in hiring on the basis of statistics: conceptual and ethical issues
ABSTRACT Assessing job applications with the aim of improving the productivity of an organization will involve attempting to predict the likely performance of applicants. A focus on efficiency in forecasting performance promotes group-level discrimination if there is group information that is statistically linked to productivity. What are the ethical concerns with employers aiming to identify and hence exclude applicants who engage in behaviours that are statistically correlated with lower performance? This paper examines the differences between lawful statistical discrimination and other ways of using information to make choices in hiring. Conceptual and ethical analyses are drawn from consideration of the practical nature of statistical discrimination in hiring, with reference to the perspectives of both applicants and employers. The paper concludes that it is not necessarily unfair for employers to use statistical discrimination in hiring, despite its potential unfairness for some applicants.
KEY WORDS: discrimination, employment, ethics, hiring, privacy, statistical discrimination
An employer has sorted through application forms and resumes from 50 applicants for a particular vacancy and has come up with a list of twelve candidates, all of whom seem good prospects in terms of qualifications, experience and suitable ambitions expressed for the position. This group must be further whittled down to a short list of five applicants for interviews. The employer then sees that two of the twelve are smokers. The employer has read some research suggesting that non-smokers are, on average, more productive employees than smokers (see e.g. Chadwick, 2006; Lecker, 2009). A particular smoker, of course, might happen to be much more productive than your average non-smoker. Nonetheless, the employer now believes that smokers, as a group, are less productive employees than non-smokers. What, if anything, is wrong with the employer dismissing the applications of the two smokers because of the statistic that the group of people who smoke is less productive than the group of non-smokers?
Statistical discrimination occurs when there is a characteristic that can be correlated with an increased or reduced risk of something across the group of persons that has it. The group characteristic is taken as a proxy for information about an individual, and used as the basis for discriminating between individuals. Statistical discrimination in hiring occurs when an employer takes a group characteristic (believed to be) correlated with higher or lower employee productivity and uses it as a proxy for productivity information about individual applicants; for example, to discriminate against applicants who belong to a group with increased risk of lower performance.
All employers discriminate of necessity whenever they have to choose one or more applicants over others when they are hiring. Usually we judge whether the discrimination is fair by whether we think there is a fair basis for the choice. This is often fleshed out in terms of whether the choice is based on job-relevant criteria, such as having higher qualifications or more experience. By way of contrast, if an employer rejected a man who scored well on the job-relevant criteria in favor of a woman who scored less well, because the employer simply preferred female employees, this discrimination would be judged unfair (and illegal in most countries). This type of unfair discrimination is "taste"-discrimination; that is, choosing between people on the basis of a personal taste or prejudice regarding a particular group.
Statistical discrimination in hiring is distinguished from taste-discrimination by its motive and its rationale (the agent's conscious justification for the action). In the smoking example, the employer's motive for discriminating does not necessarily involve any personal distaste for smokers or support for a general social prejudice against smokers. Instead, the motive is to increase the productivity of the workforce, and thereby increase profits for the firm. On the face of it, this does not seem an unreasonable motive for an employer to act on when hiring.
Similarly, the employer's rationale is not overtly based in a personal taste prejudice. The employer is not acting on the belief that it is appropriate to offer less favorable hiring terms to people with a group characteristic the agent personally finds less appealing. The employer does not thinks it somehow socially appropriate that people belonging to this particular group be afforded less opportunity in society. An employer's practice of statistical discrimination need not have anything to do with such beliefs. Instead, the rationale could be to carry out a fiduciary responsibility to shareholders to increase profits through legally permissible actions; and it is not illegal to discriminate against smokers. So we can see that the sorts of motives and rationales for taste-discrimination that we might find ethically objectionable do not apply in the case of statistical discrimination.
After further introducing the concept of statistical discrimination in hiring, a brief examination of the literature on statistical discrimination will pick out some key points salient to the ensuing discussion. The differences between statistical discrimination and other ways of using information to make lawful choices in hiring will be analyzed to distinguish the key factors identifying a hiring practice as an instance of statistical discrimination. A hiring scenario will illustrate some of the ethical implications for applicants and employers. The evaluation of the nature of the practice shows statistical discrimination's potential for producing a disturbing conflict between privacy and liberty values for applicants and employers. The final section of the paper considers the issue with reference to some privacy principles that are commonly enshrined in legislation applicable to employers, leading to the conclusion that although statistical discrimination is potentially unfair for some applicants, it may not be reasonable to characterize employers as unfair if they engage in statistical discrimination.
Influences on the use of statistical discrimination in hiring
There is a significant body of research discussing how to hire to improve productivity (see, e.g., Le, Oh, Shaffer and Schmidt, 2007; Sackett and Lievens, 2008; Schmidt and Hunter, 1998), that includes consideration of what features of employees are associated with higher and lower job performance. To what end? The natural assumption is that employers can use the results to increase their chances of hiring an applicant whose performance is higher quality; or at least, reduce their chances of hiring an applicant whose performance is lower quality. The group characteristics that could be investigated for links to productivity tend to fall into two categories: private activities and personal features. Private activities could include, for example, smoking or taking other drugs (legal or illegal); caring for children or elderly parents; religious activities; sports activities, or other leisure activities. Personal features could include, for example, race, gender, age, personality traits, other aspects of physical appearance. Some characteristics may cross the two categories; a person's weight might be partly due to genetics, whilst also influenced by private activities such as eating and exercise (Roehling, 2002, p.183).
Whether a group characteristic actually does correlate with productivity is a matter to be decided by empirical research. Currently, if empirical research on a characteristic either has not been done or has produced contradictory results, employers are faced with an uncertainty when it comes to making decisions using the characteristic. But once this research can assign a statistical probability to the relation between the characteristic and productivity at the group level, this creates a rational basis for statistical discrimination. This will effectively turn the uncertainty into a situation of quantifiable risk – thus creating a measurable incentive for employers to engage in statistical discrimination. So it is to be expected that the very pursuit of research into characteristics correlated with productivity will prompt increasing use of statistical discrimination in the future (Gandy, 2010). This is already occurring in other contexts such as admissions to tertiary study, where institutions may try to discourage or prevent students who have characteristics that are correlated with a lower chance of passing from taking courses (Simpson, 2009).
Through various forms of pre-employment testing, employers could seek information about a characteristic that is statistically correlated with a risk of lower productivity, in order to discriminate against applicants who have such a characteristic. But will they? Practicality is one factor controlling the potential practice of statistical discrimination in employment – whether the practice is an efficient means of producing the desired outcome. If a group characteristic is found to correlate with productivity, there is still the question of whether the cost to the employer of testing for it outweighs the benefits of knowing about it. An example of a low cost test is a question on an application form, such as "Do you smoke?" A pre-employment drug test, by contrast, imposes a significantly higher cost.
The other factor controlling the practice of statistical discrimination in employment is legislation. Human rights legislation commonly prohibits employment discrimination on certain grounds, and if any of these grounds are potential proxies for statistical discrimination, then statistical discrimination will be legally prohibited for these. These grounds usually include sex, race, religious beliefs, ethnicity and national origin (Gandy, 2010, p.35), amongst others. The presence of legal prohibitions against employment discrimination suggests both that societies have made an ethical determination that (the members of) certain groups suffer prejudice that unfairly harms their employment participation in that society, and that making such discrimination illegal is socially appropriate. For the purposes of this paper, I will presume that these determinations are correct, and hence that employers ought not to discriminate on the basis of sex, race, or other grounds commonly prohibited in law. Given this presumption, this paper will not focus on discussing the ethics of statistical discrimination in these areas. If the application of statistical discrimination on the grounds of sex was found not to be ethically objectionable; the practice of discriminating against job applicants on the grounds of sex would still be wrong. Accordingly, in order to focus on a "live" ethical issue, the paper will discuss only the lawful application of statistical discrimination by employers. In particular, smoking, drug use and the traditional hiring ground of academic qualifications will be considered.
How to statistically discriminate in hiring
There are several ways in which differences between groups can make it potentially beneficial for an employer to engage in statistical discrimination when hiring. We can use the example of pre-employment drug testing to consider these. Probably the most common basis for discrimination is where there is (believed to be) a difference in the average productivity of one group compared with another group. Say Bryan is expanding his data-entry business and needs to hire staff. He knows there will be a large pool of applicants as the work is not highly skilled, and the job offers a comfortable environment with flexible working hours. Bryan has read research suggesting the performance of the group of people who take some types of mind-altering drugs is more impaired than the performance of the group of non-users. (Say, for the sake of argument, the studies found the group of drug users tends to be sick more often and take more time off work, compared with the group of non-users). Bryan decides to drug-test applicants. 'After all,' thinks Bryan, 'it's not illegal to drug-test applicants, and I have a right to take actions to safeguard productivity in my business.' He would require applicants who pass the typing test to take a drug-test. Applicants who test positive or refuse the test will be weeded out. If asked directly, Bryan would agree that he is discriminating against some individuals on the basis that they belong to a group with a particular characteristic - some drug use - that has (hypothetically) a lower average productivity in comparison with the group lacking that characteristic.
A second type of difference that could be a basis for statistical discrimination is where the average productivity for two groups is the same but there is a higher variance in the productivity of members across one group compared with another. Make Group X the group of drug users, and Group Y the group of non-users. Say the two groups have the same average productivity. However, the group of drug users is estimated to contain one third users who effectively focus some of their drug use for work purposes; one third recreational drug users; and one third heavy users. The users who focus some of their drug use for work purposes have a higher productivity than non-users; the recreational users have the same productivity distribution as non-users; and the heavy users have a lower productivity than non-users. An employer who drug tests applicants will simply get information as to whether those who take the test have recently used drugs or not; it won't say which of the three types a user is. Ex hypothesi, an employer who selects an applicant from Group X rather than Group Y has a higher chance of getting a higher productivity employee, but also a higher chance of getting a lower productivity employee. Albert and Cabrillo (2000, p.6) suggest employers tend to be risk-averse. So a risk-averse employer would have an incentive to choose a member of Group Y, the non-users, over a member of Group X, the users.
Albert and Cabrillo (2000, p.8) outline a third type of group difference that could be a basis for statistical discrimination. This involves variability in the results for groups due to the testing tool, rather than due to variability in the groups themselves. So one group does not have a lower average productivity than the other, nor does productivity distribution vary wildly across each group. Rather, in this case the test for predicting productivity produces more accurate predictions for one group than for another. Bertrand, Chugh and Mullainathan (2005) suggest another form this discrimination can take, where cognitive factors affecting decision makers influence them to rely on their perceptions of a group characteristic. These factors might include being rushed, having multiple obligations competing for attention, and performing a task with a nonverbal automated response as its output - such as a manager considering resumes and putting them in a "Yes" pile or a "No" pile (p. 96). The researchers term this "implicit" discrimination rather than statistical discrimination. However, the results of their exploratory experiments to test for this, if extrapolated to real hiring situations, would fit the profile of statistical discrimination if making quicker decisions using the group characteristic was more profitable under these circumstances.
In summary, statistical discrimination in hiring can occur when an employer takes into account information about i) a characteristic that correlates with higher or lower productivity across the group of persons that has it; or ii) a high variance in the distribution of a productivity-correlated characteristic in one group; or iii) an uncertainty or pressure regarding measurement for a productivity-correlated characteristic for one group. The information is taken into account when considering employees from one group compared with those from other groups, in an attempt to reduce the risk of hiring a lower productivity applicant. Employers thus use the group characteristic as a proxy for productivity information on individual applicants.
Theorizing the economic impacts of statistical discrimination in hiring
Discussion of statistical discrimination in employment has occurred predominantly in the economics literature and has tended to focus on the likely difference in wages between a group that suffers from statistical discrimination and a group that benefits from it. Seminal papers by Phelps (1972) and Arrow (1973) are credited with originating discussion of statistical discrimination; and Aigner and Cain (1977) also produced an early influential treatment (Fang and Moro, 2011, p.135). Typically theorists propose a scenario with applicant groups A and B. Members of Group A are rated by employers as having lower average productivity than Group B, either because Group A has lower average productivity, or because the employer finds it harder to estimate the productivity of Group A members. Using mathematical models that randomly assign applicants to employers, theorists found that the hypothesized statistical discrimination of a lower wage offer to Group A members would occur, and would persist over time. This finding was of significant interest. Previous discussion of employment discrimination in the economics literature focused on taste-based discrimination and argued that it was not supported by a competitive market system so market forces would penalize employers for this and work toward eliminating it (Becker, 1957, 1971). With statistical discrimination, however, the unequal treatment and economic discrimination are actually generated by a competitive market system (Sattinger, 1998, p229).
Gathering empirical data on the impact of statistical discrimination is difficult (Moro and Norman 2003). Evidence on widespread group discrimination in hiring has been presented (e.g. Bertrand and Mullainathan, 2004), but determining whether it is statistical discrimination or taste discrimination that is operating – or perhaps, how much of the discrimination is due to taste and how much due to using a perceived proxy for productivity– is considerably harder (Bertrand and Mullainathan, 2004; Harford 2008). However, part of the problem is that researchers tend to focus on traditionally widespread factors for discrimination in societies, such as race and sex, for which we have good historical reason to suspect there may be significant taste discrimination (conscious or unconscious - see Banarji, Bazermann and Chugh, 2003). If researchers were to focus on factors which are correlated with lower average productivity but do not have a long-standing history of prejudice attached, this problem, at least, would be mitigated.
It is common in the economics literature to model the effect that statistical discrimination has on the level of wage an employer will offer an applicant. On the whole, there has been much less discussion of employers engaging in statistical discrimination to make a binary decision on whether or not to screen applicants out of consideration (see however, Berk, 2001; Harford, 2008).Yet this latter situation is both more typical, and more worrying if discrimination occurs (Sattinger, 1998, p.229). While it is bad to be offered a lower wage for a job due to statistical discrimination, it is worse not to be offered a job at all – or to be excluded from further consideration for a job.
Theorizing the ethics of statistical discrimination in hiring
Outside of the economics literature, Baumle and Fossett's (2005) conceptual analysis of statistical discrimination in employment draws a contrast with taste-discrimination, considers some of the implications of statistical discrimination in employment, and suggests that policies used to combat taste-discrimination are not likely to be effective against statistical discrimination. Baumle and Fossett compare some differences in how the law has dealt with statistical discrimination in the employment context and in the insurance context. They draw on this when discussing why statistical discrimination seems to be accepted in some circumstances but not in others, eventually suggesting statistical discrimination may appear more acceptable the less it impinges on essential individual rights. This suggests the view that being able to work is a fundamentally important part of what it is to be a human being, whereas having insurance is more of an optional extra. So if a society considers the right to insurance less basic than a right to work, then statistical discrimination might be acceptable to that society in insurance but not in hiring (Baumle and Fossett, 2005, p.1270).
Lippert-Rasmussen (2007, p.386) suggests that "…a moral case for statistical discrimination …[would consist] in the fact that the use of the relevant proxy indicator is an efficient way of pursuing valuable outcomes, for example hiring those best qualified." Lippert-Rasmussen (2007, 2011) has argued against viewing statistical discrimination as inherently morally objectionable, although he allows that there are some special problem cases; one such is where statistical discrimination becomes a self-fulfilling prophecy. Decker (2010, p.799) provides a historical example from West Africa. British managers promoted African staff, but then withdrew responsibilities from the posts to which the staff were promoted, based on the British managers' low performance expectations of these staff - thereby undermining opportunities for staff to perform or develop in the posts. The discriminatory practice thus stimulates and perpetuates the problem. Setting aside such cases, Lippert-Rasmussen (2011) argues that complaining that statistical discrimination fails to treat persons as individuals does not automatically indicate a moral problem with it. This point will be taken up in a later section of the paper.
Finally, Maitzen (1991, p.23) is interested in whether statistical discrimination in general, or a given instance of statistical discrimination, is morally justified as part of effecting good or permissible social policies. These policies could include, for example: car insurance policies (discriminating against all young males, regardless of whether a particular young male is a safe driver); race-based affirmative action programs; and sex-based pension plans. Maitzen defines moral justification purely in terms of the benefits and harms for the aggregate of those affected. He draws on his utilitarian definition of morality to claim that the level of statistical discrimination used in a social policy "is justified if and only if the information-cost of further statistical refinement equals or exceeds the net social utility to be gained by such refinement" (Maitzen, 1991, p.26), and that a social policy involving statistical discrimination is justified if it produces an overall positive outcome (p.30). Statistical discrimination is thus morally justified simply if it produces equal or more social utility than not engaging in statistical discrimination.
This paper eschews the "top-down" approach of applying one normative criterion (such as utilitarianism) to the principles or practices of statistical discrimination. Instead, it aims to take a "bottom-up" approach, to provide a sense of what the lawful practice of statistical discrimination might mean to those directly affected by it. The disadvantage of this is that a simple formula for when statistical discrimination is ethically acceptable in hiring is unlikely to emerge. The advantage is that the paper should speak to people across different ethical ideologies. The paper will give a sense of what the issues are for those involved; this can be drawn on when applying one's preferred normative criterion. Following an exploration of how statistical discrimination differs from other bases of selection in hiring that will determine the practical nature of the issue, the paper draws out the ethical implications of statistical discrimination for the choices facing applicants and employers. These will be discussed from the perspectives of the agents involved, to add to our ethical understanding of the issues.
How statistical discrimination differs from other lawful discrimination in the hiring process
In many hiring situations employers will use applicants' personal information to discriminate in favor of some applicants and against some other applicants when they hire. In fact, employers discriminate of necessity in hiring whenever they have to choose one or more applicants over others. Discrimination is simply (non-randomly) choosing one thing over another. The mere fact that a practice involves discrimination in this sense does not show that it is bad, nor is it obvious there should be any ethical concerns with hiring practices that involve legal discrimination. We will need to consider how statistical discrimination may differ from other legal discriminations used as a basis for hiring decisions.
Traditionally, common methods used to gain information from applicants in the hiring process include application forms and/or CVs, interviews and reference checks (Levashina and Campion, 2009, pp.236-239; Taylor, Keelty and McDonnell, 2002, pp.9-10). Huffcut and Youngcourt (2007, p.181) suggest that application blanks and interviews are the most widely-used selection methods. These methods typically produce information about qualifications; past work experience; skills gained from other activities; endorsements from previous employers (referees); communication and social skills (interview); and job applicants' stated ambitions and expectations of the position. Employers clearly believe that this constitutes relevant information about an applicant's likely productivity in the job. But does using this information also involve statistical discrimination? Consider: it is very probable that there is a strong statistical correlation showing that applicants with, say, an engineering qualification perform better in an engineering job than applicants without that qualification; or that someone with engineering experience will perform better in an engineering job than an applicant with no practical experience in that line of work. If employers are assuming this statistical correlation when they seek this information, on the face of it, it seems they are engaging in the same practice of statistical discrimination as outlined earlier. However, digging deeper reveals two main differences with the use of this sort of statistical correlation.
First, there is a degree of difference in the control of the information provided. The examples above involve the applicant volunteering personal information of their choice, such as their qualifications, past work experience or their chosen referees. Short-listed applicants typically attend an interview where they can explain how they think they will perform in the job they are applying for, and provide evidence drawing on past activities to support their claims. As applicants respond to questions, they have the opportunity to assess how they are making themselves understood and to provide further information if necessary to showcase their skills and experience. With information requested by an employer for the purposes of statistical discrimination, however, the applicant has little understanding of or control over the presentation or interpretation of the information they are asked to provide. Consider the difference between a section of an application form that asks applicants to list relevant work experience, with a section that asks applicants to check a YES / NO answer about whether they smoke.
Second, and most importantly, there are two ways in which there is a difference in the locus of assessment of actions. First, information about qualifications, work history and past or current work experience is information that focuses on past or current job-relevant performance. Second, the information links to the individual applicant. What the particular applicant has performed, is performing, or can perform are features of the individual that show that individual's potential productivity in the job. Discrimination between applicants on these grounds is not statistical discrimination as the information is not a proxy for productivity information, but instead gives a direct indication of the individual's capabilities. So while initially employers ask about factors such as qualifications, work history and work experience on the basis of their statistical correlation with productivity in the job, the information provided says something concrete about that individual in relation to the job.
By contrast, the information that factor X is correlated with lower average productivity across the group with X (or that Xes) does not tell us what a particular applicant with X (or who Xes) has done, is doing or can do. This is the crux of the issue. Information about a group of which an individual is a member, where the group feature is a proxy for productivity information, bears only a statistical link to the individual; "statistics are summaries that do not apply to particular individuals" (Ding and Stillman, 2005, p.145). With statistical discrimination, applicants are judged on a feature that relates to productivity at the group level, but has no necessary connection to the productivity of any particular individual applicant.
Baumle and Fossett (2005) point out that decisions made on some of the traditional legal bases of discrimination, such as academic qualifications, could still involve statistical discrimination (p.1255). They give the example of an employer who screened out applicants for a job on the basis of whether the applicants had a college degree or not. So there are some applicants who are good candidates for the job but do not have a degree; these applicants will miss out on being considered for the job (p.1255). Say Baumle and Fossett have in mind a situation where Employer A believes that being a college graduate is statistically correlated with a higher level of productivity relative to the group of non-graduates. If Employer A screens out applicants who do not have any college degree based simply (or principally) on the belief that graduates, as a group, are more productive than non-graduates, this would indeed count as statistical discrimination.
Screening job applicants based on college degrees does not necessarily involve statistical discrimination, however. Say Employer B is looking to hire someone to fill a position as a lawyer, and that having a law degree is necessary to practice law in the position. If Employer B screens out applicants who do not have a law degree, this does not count as statistical discrimination against those applicants. By definition, no applicant without the qualification could be a good candidate; having the degree is an essential attribute for being able to do that job. Here the employer's discrimination is not based on a statistical correlation with productivity across a group, but rather a group feature that identifies a minimum threshold for being able to do the job.
We have just outlined two possibilities with regard to group discrimination concerning college degrees. The first involved the rationale that college degrees are statistically correlated to increased productivity in a job, and is statistical discrimination; the second, where a college degree was necessary for doing the job, is not statistical discrimination. However, there is a third possibility: Employer C has a vacancy for a job and is currently only seeking applicants with a college degree. For this job, the degree qualification is not a legal requirement, but the degree-type skills that a person gains as part of achieving a college degree are necessary. In this case, it is the degree-type skills that are job-relevant, rather than the degree certification itself. If there are some applicants who possess the degree-type skills (and are thus potentially good candidates), but do not have a degree, these applicants will miss out on being considered by Employer C for the job.
Let us say Employer C is relying on a statistical correlation between a degree and having degree-type skills; how is this not also statistical discrimination for productivity? I suggest that it is not a morally objectionable form of statistical discrimination if the employer is using the method as a proxy for identifying a minimum threshold of skill necessary to do the job and there is no other method reasonably available to do this. The issue is how to determine whether an applicant has adequate degree-type skills, in order to be suitable for the job; and this involves the availability and reliability of information about the applicant. Obviously, a fast and reasonably reliable way of checking for this particular information is for the employer to look at whether someone has a degree.
Presumably, an applicant who has the skills but not the degree could take steps to make information about their skills available to an employer, perhaps through volunteering some demonstration of their skills, or obtaining a degree equivalency certification through some process of Recognition of Prior Learning. The applicant needs to make an extra effort to notify the employer of the relevant information about their skills, information that other applicants are able to provide simply through noting their degree. This might be unfair, but perhaps this unfairness needs to be addressed by authorities outside of employers. When it comes to the disciplinary approaches and generic higher thinking skills such as can be developed through a college degree, for example, the role and responsibility for credentialing these belongs with the State, not with an individual employer. As long as the employer does not rule out in principle the consideration of alternative credentials that are shown to be equivalent, then the employer would not be engaging in statistical discrimination in asking whether applicants had a college degree.
In conclusion, what distinguishes whether a particular hiring process counts as statistical discrimination is the relationship between the information the process seeks, and what the job requires. It is not statistical discrimination if the process identifies a minimum threshold for being able to perform the job. It is not statistical discrimination if the process seeks information about factors that indicate that individual's productivity with regard to the job requirements. It is statistical discrimination when a process seeks information about a statistical proxy for productivity (a factor that correlates with higher or lower productivity across a general group of people compared with another group) and does not identify a minimum threshold required for performing the job.
The ethics of statistical discrimination – from concept to practice
The mere fact that statistical discrimination categorizes individuals as part of a group does not show that an employer who practises statistical discrimination is failing to respect an applicant's individuality. Lippert-Rasmussen (2011, p.53) notes that it is possible to use information that categorizes an individual in this way and also take into account other information on the same factor that is particular to that individual. He suggests that "X treats Y as an individual if, and only if, X's treatment of Y is informed by all relevant information, statistical or non-statistical, reasonably available to X" (2011, p.54). For example, an employer could consider the fact an applicant belongs to a group that has a lower-than-average productivity but also consider the applicant's work history that indicates continued promotion within their last place of employment, which is not suggestive of lower-than-average productivity. This dual consideration means the applicant's individuality is still taken into account.
The point above is intended to show that the consideration of information gained through statistically-backed generalizations will not necessarily result in decisions that fail to treat people as individuals. However, Lippert-Rasmussen does allow that there are "morally faulty ways of relating to available information, including that contained in generalizations, that we can helpfully describe as failing to treat people as individuals" (2011, p.53); this would include, for example, using some information gained from a general categorization while ignoring other relevant information. This leaves open the possibility that the practice of statistical discrimination on its own is in some sense morally faulty. Does a decision made purely on the basis of statistical information necessarily treat a person badly (even if, for other reasons, this discrimination is ultimately judged the right thing to do)?
We already have a practical example of statistical discrimination where people are judged purely on group categorizations: insurance. The insurance industry is set up to work specifically on the principle of discriminating between people on the basis of statistical correlations for groups to which they belong. Insurance companies make money by charging groups of consumers more over an extended period than they are likely to pay those groups of consumers over that period. Insurers work out how much they are likely to have to pay using statistical correlations. Consumers from groups who are statistically less likely to require a payout are charged lower premia. Baumle and Fossett (2005, pp.1268-9) state that considerations of fairness in insurance have focussed on fairness at the group level. Say that the group 'male drivers under 25 years old' is involved in more car crashes than any other group. As long as the insurance premia accurately reflect the average crash risk for that group, those premia are considered fair.
Baumle and Fossett (2005) suggest that the approach of the North American legal system to statistical discrimination tends to differ depending on whether it occurs in the insurance context or the employment context. As mentioned earlier, they suggest that perhaps economic efficiency is more acceptable in insurance than in employment, as insurance is less fundamental to our lives than is our work. As evidence of this, Baumle and Fossett (2005, p.1270) note that in states in the US where car insurance is mandatory - so not having insurance impacts more fundamentally on people's lives, preventing them from being able to run a car - some states have prohibited the usual practice in the automobile industry of statistical discrimination based on sex.
Drawing on some legal rulings and policies of some North American states, Baumle and Fossett speculate that perhaps society finds statistical discrimination more or less acceptable depending on whether the good sought has more or less impact on the person subjected to the discrimination. We can consider the experience of people subject to insurance: customers. Customers know that the information they provide when applying for insurance is mapped against group statistics for particular characteristics to form an overall risk profile that is used to determine whether they will be offered an insurance policy, and at what price. Having an insurance industry allows customers the opportunity to accept to be treated as an average of some collective statistics and pay premia in order to trade off the financial risk posed by potential future adversities. The insurance context utterly fails to treat persons as individuals, but it is not obvious that it necessarily acts unethically toward customers in doing so. As customers, we want to be able to protect ourselves in case of the sudden, huge costs that can be associated with accident, illness or other future misfortune, and we explicitly trade off our individuality for this protection; that is what it is to get insurance.
My point with the insurance example supports Lippert-Rasmussen's conclusion (2011, p.58) that we cannot specify what (if anything) is wrong about statistical discrimination solely in terms of the practice failing to treat people as individuals. However, the hiring context differs from the insurance example. Unlike applicants for insurance, applicants for employment are not usually aware that the personal information they provide may form the basis for some practices of statistical discrimination. While applicants for insurance are, in a sense, choosing the discriminatory practices in order to gain the insurance benefits, applicants in the hiring context simply have the practice imposed on them. I propose that the lack of awareness, and hence reduced freedom to consent or otherwise act on the matter, is what makes statistical discrimination less acceptable (and perhaps unacceptable) in employment.
Applicants for insurance, in actively choosing the statistical discrimination that is an open and necessary part of a non-compulsory insurance industry, are able to act as agents in the insurance application process. In hiring situations involving statistical discrimination, it is not the loss of individuality for applicants that is ethically problematic, it is the loss of their individual agency in the process. The concept of statistical discrimination has been analyzed as involving an intersection of two aspects: the information the process seeks and what the job requires. These are the two aspects of statistical discrimination that we will subject to further ethical interrogation. In what follows, these will be explored primarily from the viewpoints of the agents directly involved in the practices, namely applicants and employers, with particular focus on privacy and liberty issues.
The information the process seeks
The practice of statistical discrimination involves asking applicants to give personal information about themselves on the basis of which a decision affecting that individual will be made. Requiring people to reveal personal information in this way raises privacy issues. Benn (1978) argues that privacy is ethically important because some privacy is essential for us to create and maintain our own personalities, and it is our individual personalities that allow us to be individual rights-bearing moral agents. The point of having individual moral rights would be considerably diminished if these were not supporting us to be able to live and act as agents with individual personalities.
Privacy theorist Ferdinand Schoeman (1984, p.3) recommends the following definition of privacy: a state or condition of limited access to a person. Hence, if others have unlimited access to you as a person, you have no privacy. You have a degree of privacy insofar as access to you by others is limited in some way. We are interested in privacy just as it pertains to information (rather than any broader concept of privacy per se). We will define private information as: personal information to which the person concerned has an interest in controlling access.
Rossler (2005, p.113) proposes that "If informational privacy consists in a person in principle being able to control or at least know or estimate who has what information about her, then a violation of informational privacy consists ... in the person no longer having this control". Applicants clearly have an interest in having some control over the employer's access to at least some of their personal information. This is especially the case if employers might seek the information to use as a proxy for information statistically correlated with productivity at a group level. Applicants will want to ensure the information about them that is provided to the employer is accurate; that it is used to form an accurate picture of themselves (or at least, a picture that is not less favorable than an accurate picture would be); that only job-relevant information about them is used to make a determination of their suitability; and that, in general, the personal information required of them is kept to a minimum, to best protect their privacy in their lives outside work.
Another way of stating this informational privacy concern is as a claim of a liberty right: applicants want the liberty to keep personal information private, where "personal" refers to information outside that which presents their (public) job performance. However, employers could claim their own right to liberty grounds a right to take actions to improve the productivity of their businesses – actions such as, for example, requiring applicants to provide information that employers want and are not legally prohibited from asking about, so employers can try to increase their success at selecting a high-performing applicant. These liberty rights claims seem to oppose each other: the liberty to access private information about persons versus the liberty to keep personal information private. But is this a genuine conflict? Each party in the employment process can still exercise their liberty in this situation. The applicant can exercise their liberty by refusing to provide some private information; and the potential employer can exercise their liberty in their choice between either considering just the information that is volunteered, or refusing to consider the applicant.
Whilst neither party has a knife to their throat, a scenario will help illustrate the nature of the constraints on the applicant's liberty. Imagine once more our employer who has read studies showing a statistical correlation between smoking cigarettes and lower productivity in employees. The employer makes sure that whenever she advertises a vacancy, the application form asks whether the applicant smokes. When the forms come in, she screens out all the applicants who haven't answered "No" to the question about smoking. Say a smoker applying for the job realizes such discrimination could occur. What could the applicant do? Well, the applicant could give up smoking, then truthfully answer "No" about smoking and remain in the running for the job. Or the applicant could remain a smoker, and lie about it, again answering "No" to the question about smoking. Or the applicant could remain a smoker and truthfully answer "Yes" to the question, or they could simply refuse to answer the question; either way getting struck off the list.
In exercising their liberty over providing this private information, the applicant is forced to choose between: 1) stopping their private activity to maintain the same chance of getting the job; 2) lying about their private activity to maintain the same chance of getting the job; 3) telling the truth and probably losing the chance to get this job; or 4) refusing to answer and probably losing the chance to get this job. Every option in this situation appears unpalatable. The first option imposes a serious restriction on the applicant's liberty. The second requires the applicant to tell an outright lie; this would be uncomfortable for many applicants, and leaves them vulnerable in their job if the lie is later exposed. The third and fourth options both have the outcome that the applicant will no longer be considered for the job.
Thus we have a situation where applicants are pressured to tailor not their working activities, but their private activities to fit with information about statistical correlations to productivity that an employer has, just to be considered for a job. Such an employment market appears to have too much control over an applicant's private life. Statistical discrimination with regard to information about personal activities is a concern in terms of the constraints it may impose on people engaging in their choice of personal activities, or in forcing them to choose between lying or revealing them to their detriment. Furthermore, consider the position of an applicant vis-à-vis the job market as a whole. If employers are able to demand and give preference to those who reveal personal information (believed to be) statistically correlated with higher productivity, then job applicants who protect their privacy by refusing to reveal this information will not have the same liberty of job opportunities.
What the job requires
The employer's liberty to treat job applicants in ways designed to increase workforce productivity raises fundamental questions about how we should conceive of a job. There is a tension if employers see themselves almost entirely as owners or managers of companies set up for private ends, but society considers the jobs they offer to be social goods as well as private goods. Participation in a job is very important for most people's wellbeing and this participation is facilitated by society. As well as providing developed social structures and institutions, societies allow people particular benefits when setting up their businesses, such as a legal system with which to enforce business contracts, and limited liability in the event of failure. When hiring, an employer who publicly advertises a job attempts to benefit from having a group of appropriately-trained persons from which to source an employee. These applicants will have drawn on social resources for their education and training; hence the employer (amongst others) profiting from society's facilitation of their "human resources".
If a job is properly considered at least in part a social good, then individual members of the society may have rights to access it in certain ways, such as without being subject to certain types of discrimination. We see this expressed currently in human rights legislation concerning employment. However, human rights legislation will be sensitive to what employers need to know to successfully hire employees (Gandy, 2010, p.36). For example, while legislation may prohibit discrimination on the basis of religious belief and hence make it illegal for employers to ask questions about religious belief, exceptions are made for jobs where the religious belief is central to the work, such as being a vicar, priest or rabbi. Basically, employers are allowed to request information that is required for hiring purposes with regard to the job.
Our question is, could the information about a factor that is statistically correlated with productivity be required for hiring purposes or hiring functions? To answer this, we will need to consider the characterization of the employer's hiring activity and purpose, as well as the hiring environment. Levashina and Campion (2009) suggest it is a "fundamental truism in HR management … that hiring procedures should be based on the job requirements", and that knowledge of job requirements ought usually to be obtained through a job analysis (p.235). Thompson and Smith (2009, p.260) suggest that "the purpose of hiring labor power is to expand capital". Together these suggest that the employer should monitor their workforce with an eye to the vacancy-filling or job creation that will expand capital, and set in place hiring procedures that reflect the requirements for the performance of those work positions. However, as it stands, this characterization is too broad to help answer the question, as it is compatible with two different descriptions of the hiring purpose.
One description of the hiring purpose might simply be to source an employee who is suitable for the position. On the face of it, this description does not appear unreasonable. Employers want an employee who is suitable for the work position, and are anxious to avoid hiring someone who is unsuitable. If we use this description of the hiring purpose, the question of whether the information about a factor that is statistically correlated with productivity is required for the hiring purpose can be answered with a clear "No". The standard hiring methods provide plenty of relevant information to identify a suitable applicant. While employers may nonetheless be interested in undertaking statistical discrimination as part of assessing applicants, the information could not be said to be required under this definition of the hiring purpose.
For vacancies that are difficult to fill, sourcing any suitable employee might be the extent of the hiring purpose. However, other employers may not want just any suitable employee. A second description of the hiring purpose is to hire the best – most productive - applicant for the position. If collecting some information to carry out statistical discrimination helps employers select a better employee than if it were not used, then employers could argue that it is necessary for the purpose or for carrying out the activity. This raises the question of whether employers should be considered to have a right to require information not strictly necessary for trying to find a suitable person for the job, but that could be necessary or useful for trying to find the best person for the job.
Employers could appeal to their general right to liberty to support a claim of a right to try to source the most suitable person for the job, rather than simply any person who could perform adequately in the position. The employer owns the business and is the direct cause of the availability of the job. Provided employers are not discriminating on the basis of grounds prohibited in human rights legislation, why should they not have a right to whatever information they deem necessary to finding the person likely to be best for the job? This may also produce better consequences. Besides employers feeling happier about having fewer restrictions, society has an interest in the employer hiring the person who seems most likely to be the best applicant for the job. Having the best available people in jobs means an increase in productivity for businesses; that may mean an expansion in business, creating more jobs and potentially increasing the taxes businesses pay to Government to spend on social issues.
We have noted that human rights legislation that sets out prohibited grounds for discrimination places restrictions on what information an employer may legally use when making hiring decisions. But this will only cover statistical discrimination that occurs on one of those prohibited grounds. Unless, say, weight or smoking is classed as a disability or illness, such legislation will not make it illegal to discriminate on these factors. However, human rights legislation is not the only legislation covering the hiring process. In particular, privacy legislation sets out rules for the collection and use of personal information. This may bear on whether employers have a right to require information that they judge necessary for finding the best person for the job.
Privacy legislation will commonly take account of both the purpose for which information is collected, and the epistemic qualities of the information collected. For example, it may be specified that the information must be necessary for one or more of the activities or functions of the organization (Australian Privacy Act 1988, S.1.1); and that with regard to the purpose for which it is used, the information is accurate, complete, up-to-date, relevant and not misleading (NZ Privacy Act 1993, S. II, p.8). Such stipulations make the employer's characterization of their purpose central to assessing its legitimacy. If an employer states their purpose as "to get a picture of the individual applicant X as a basis for making a hiring decision", then because the information pertains to a group, the resulting information may not be accurate or relevant and so the picture may be misleading. However, say the purpose were restated as: "to gain a risk profile of applicant X as a basis for making a hiring decision". As compiling statistics on factors correlated with risk elements is exactly how risk profiles are built, the employer could claim that the information is accurate, relevant and not misleading.
Finally, if the practice of statistical discrimination were to become widespread, then any employer might argue that they needed to engage in such discrimination to avoid being at a competitive disadvantage. For example, if the large companies in a particular industry adopted pre-employment drug-testing, small companies without such testing would have reason to fear that the applicants for their own vacancies would include a higher proportion of workers who would fail (or perhaps have failed) these drug-tests. If this were the case, we might have to qualify our description of the employer's purpose, including a factor such as 'finding an employee who is no more likely to be a risky bet than the employees of one's competitors'; or, 'engaging in a hiring practice that is no less risky than one's competitors in terms of both the employee sourced and the cost of sourcing'. If these conditions were to eventuate, an employer would have strong grounds for claiming that statistical discrimination was necessary for his or her purpose.
From the viewpoints of the agents involved, the main ethical issue raised by statistical discrimination in hiring can be characterized as a conflict of liberties: the liberty to access information to try to improve your workforce's productivity versus the liberty to access employment opportunities as an individual agent. Employers can require personal information from, and use it to discriminate against, a job applicant solely on the basis of a statistical link between such information and decreased productivity. This disrespects key features of the applicant as an agent, including consent and choice. Statistical discrimination also imposes a potentially heavy cost on applicants who value their privacy regarding some of their personal information. As such, it appears to be a practice that is unfair toward some applicants. Against this, for employers to treat persons as individual agents at all stages of the hiring process may impose unreasonable costs; there may be competitive disadvantage to ignoring risk information; and the employer's goal of selecting the person likely to be most productive in the job has social benefits as well as self-interested benefits. Finally, statistical discrimination in employment sits uneasily with regard to principles of informational privacy that typically undergird privacy legislation. However, where these principles are based around the employer's purpose, hiring strategies aiming to select the best applicant appear to undercut these. Moreover, the competitive nature of the employment market means the use of statistical discrimination by some employers in a given industry or segment of the labor market may force its use on other employers. Thus, despite the potential unfairness of statistical discrimination toward some applicants, under such circumstances it is not clear that employers should be characterized as acting unfairly if they use statistical discrimination.
Aigner DJ and Cain GC (1977) Statistical Theories of Discrimination in Labor Markets. Industrial and Labor Relations Review, 30, 175–187.
Albert R and Cabrillo F (2000) Sex discrimination: Is the labour market a market for lemons? http://www.ucm.es/info/iudem/cabrillo.DT00-4.pdf (accessed 15 September 2005).
Arrow KJ (1973) The Theory of Discrimination. In Ashenfelter, O and Rees, A. (Eds), Discrimination in Labor Markets, Princeton, NJ: Princeton University Press.
Banarji MR, Bazermann MH and Chugh D (2003) How (un)ethical are you? Harvard Business Review, 81, 56-64.
Baumle AK and Fossett M (2005) Statistical Discrimination in Employment: Its Practice, Conceptualization, and Implications for Public Policy. The American Behavioral Scientist. 49 , 1250-1274.
Becker GS (1957, 1971, 2nd ed.). The Economics of Discrimination. Chicago: University of Chicago Press.
Benn SI (1978) Human rights – for whom and for what? In Kamenka E. and Tay AE-S (Eds) Human Rights. London: E Arnold.
Berk JB (2001). Statistical Discrimination in a Labor Market with Job Selection. http://faculty.haas.berkeley.edu/berk/papers/disc.pdf (accessed 26 September 2007).
Bertrand M, Chugh D and Mullainathan S (2005) Implicit Discrimination. American Economic Review, 95, 94-98.
Bertrand M. and Mullainathan S (2004) Are Emily and Greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination. American Economic Review, 94, 991-1013.
Blume LE (2006) The Dynamics of Statistical Discrimination. The Economic Journal, 116: F480–F498.
Chadwick KL (2006) Is leisure-time smoking a valid employment consideration? Albany Law Review, 70,117-141.
Decker S (2010) Postcolonial transitions in Africa: Decolonization in West Africa and present day South Africa. Journal of Management Studies, 47, 791-813.
Dickinson DL and Oaxaca RL (2006) Statistical discrimination in labor markets: An experimental analysis. IZA Discussion paper 2305. http://ftp.iza.org/dp2305.pdf (accessed on 1 July 2011)
Ding VJ and Stillman JA (2005) An Empirical Investigation of Discrimination Against Overweight Female Job Applicants in New Zealand. New Zealand Journal of Psychology. 34, 139-148.
Ewens M, Tomlin B and Wang LC (2011) Statistical discrimination or prejudice? A large sample field experiment. http://ewens.tepper.cmu.edu/wp-content/uploads/2011/04/statistical-discrimination2011.pdf (accessed 30 June 2011)
Fang H and Moro A (2011) Theories of statistical discrimination and affirmative action: a survey. In Benhabib, J, Jackson MO and Bisin A. (Eds), Handbook of Social Economics, 1A, 133-200.
Feltovich N and Papageorgiou C (2004) An experimental study of statistical discrimination by employers. Southern Economic Journal, 70, 837-849.
Gandy OH (2010) Engaging rational discrimination: exploring reasons for placing regulatory constraints on decision support systems. Ethics and Information Technology, 12 (1), 29-42.
Harford T (2008) The dangers of rational racism. In The Logic of Life, London: Abacus
Huffcut AI and Youngcourt SS (2007) Employment Interviews. In Whetzel DL and Wheatson GR (Eds), Applied measurement: Industrial psychology in human resource management. Mahwah, NJ: Lawrence Erlbaum.
Le H, Oh I-S, Shaffer J and Schmidt F (2007) Implications of methodological advances for the practice of personnel selection: How practitioners benefit from meta-analysis. Academy of Management Perspectives, 21 (3), 6-15.
LeBlanc G (1995) Discrimination in the labor market. Canadian Journal of Economics. 28, 702-17.
Lecker MJ (2009) The smoking penalty: Distributive justice or smokism? Journal of Business ethics, 84, 47-64.
Levashina J and Campion MA (2009) Expected practices in background checking: Review of the Human Resource Management Literature. Employee Responsibilities and Rights Journal, 21, 231-49.
Lippert-Rasmussen K (2007) Nothing personal: On statistical discrimination. The Journal of Political Philosophy, 15, 385-403.
Lippert-Rasmussen K (2011) "We are all different": Statistical discrimination and the right to be treated as an individual. The Journal of Ethics, 15, 47-59.
Lundberg S and Startz R (1983) Private discrimination and social intervention in competitive labor markets. American Economic Review. 73, 340-7.
Maitzen S (1991) The Ethics of Statistical Discrimination. Social Theory and Practice, 17, 23-45.
Moro A and Norman P (2003) Empirical Implications of Statistical Discrimination on the Returns to Measures of Skills. Annales d'Economie et de Statistique. 71-72,399-417.
Phelps ES (1972) The statistical theory of racism and sexism. American Economic Review, 62, 659-61.
Roehling MV (2002) Weight discrimination in the American workplace: Ethical issues and analysis. Journal of Business Ethics, 40, 177-89.
Rossler B (2005) The Value of Privacy. Cambridge, UK: Polity Press
Sackett PR and Lievens F (2008) Personnel Selection. Annual Review of Psychology, 59, 419-50.
Sattinger M (1998) Statistical Discrimination with Employment Criteria. International Economic Review. 39, 205-30.
Schoeman FD (Ed). (1984). Privacy: Philosophical dimensions of the literature. In Philosophical Dimensions of Privacy: An Anthology. Cambridge: Cambridge University Press
Schmidt FL and Hunter JE (1998) The Validity and Utility of Selection Methods in Personnel Psychology: Practical and Theoretical Implications of 85 Years of Research Findings. Psychological Bulletin. 124, 262-74.
Schwab S (1986) Is Statistical Discrimination Efficient?. The American Economic Review. 76, 228-34.
Simpson O (2009) Open to people, open with people: Ethical issues in open learning. In Demiray U and Sharma RC (Eds), Ethical Practices and Implications in Distance Learning: IGI Global.
Taylor P, Keelty Y and McDonnell B (2002) Evolving Personnel Selection Practices in New Zealand Organisations and Recruitment Firms. New Zealand Journal of Psychology. 31, 8-18.
Thompson P and Smith C (2009) Waving, not drowning: Explaining and exploring the resilience of labor process theory. Employee Responsibilities and Rights Journal, 21, 253-62.
Terimakasih telah membaca Discriminating in hiring on the basis of statistics: Conceptual and ethical issues. Gunakan kotak pencarian untuk mencari artikel yang ingin anda cari.