Decreasing Overt Discrimination Increases Covert Discrimination: Adverse Effects of Equal Opportunities Policies

Building on the moral licensing literature, this paper examines whether highlighting the successful implementation of an equal opportunities policy in a company leads to covert forms of discrimination in hiring decisions (i.e., expressing a preference for a white candidate over an equally qualified black/Moroccan candidate in an ambiguous context). Furthermore, moral self‐image is indirectly tested as a possible underlying mechanism. Two scenario studies first revealed that covert discrimination is more likely after highlighting a successful implementation of an equal opportunities policy in the company (study 1) and that elevated levels of moral self‐image are related to covert discrimination (study 2). Subsequently, a field study revealed that the presence of successful equal opportunities policies positively related to employees’ moral self‐image (study 3).


| INTRODUC TI ON
A challenging debate in the recruitment and selection literature concerns the issue of employment discrimination. This debate started more than 50 years ago in the United States and led to the development of the Civil Rights Act of 1964. This Act rendered discrimination illegal and made prosecution possible. The Act was refined in 1991, translated into different laws and directives (e.g., European legislation) and picked up by organizations, who developed policies for equal employment opportunities, regardless of race, cultural background, and color or religion (Wilson & Iles, 1999).
In this study, we focus on racial discrimination as racial minorities remain a strongly targeted group for discrimination in the workplace (Deitch et al., 2003). In response to this, equal opportunity policies were developed and showed some success. Policy implementation changed the behavior of gatekeepers (i.e., recruiters) within companies, resulting in an increase in minority members hired in personnel selection (e.g., Colarelli, Poole, Unterborn, & D'Souza, 2010). Yet, recent studies warned for an important possible flipside to this success.
While overt forms of discrimination (e.g., openly expressing prejudices toward certain groups) are reduced (Schuman, Steeh, Bobo, & Krysan, 1997), more covert or subtle forms of discrimination (e.g., covering the expression of prejudices as neutral or even moral behavior) are not (Sue, 2010;Sue et al., 2007). This may be due to two reasons. First, since covert forms of discrimination are less visible and more difficult to attribute to prejudices, they are easily overlooked and cannot (all) be covered by policymaking. As a result, covert discrimination still allows prejudiced judgment and decision-making without the fear of (legal) prosecution. Second and more importantly, we argue that the presence of a successfully implemented policy may in itself provoke covert discrimination and thus be responsible for an increase in covert types of discrimination. Specifically, we bring forward the counterintuitive idea that covert discrimination can be even more likely to occur when successfully installed equal opportunities policies are made salient. In this way, a decrease in overt forms of discrimination could in itself stimulate more covert forms of discrimination.
We build our theorizing on the mechanism of moral licensing, stating that a prior moral deed can increase the moral self-image of the actor, which subsequently allows the actor to relax moral strivings, and thus to engage in less moral behavior on the next occasion (Monin & Miller, 2001). We focus in this study on covert discrimination, defined as discriminatory behavior that can be justified by the context as neutral or even moral behavior. This definition is derived from the moral credentials model of moral licensing which states that in contexts in which two competing explanations (i.e., one moral and one less moral) are possible for immoral behavior, the actor can easily give a moral rationale for acting immoral. In the moral licensing literature, this kind of behavior is also labeled "ambiguous behavior" since it is enacted in an ambiguous context (Merritt, Effron, & Monin, 2010).
We add to the moral licensing literature in several ways. First, we lift the idea of moral licensing from the individual level to the policy level. We argue that not only individual (Monin & Miller, 2001) or group past behavior (Kouchaki, 2011) but also the policy of a company can offer a moral license to express prejudices toward a black/ Morrocan candidate in a hiring scenario (study 1). This may have important implications for organizations who have successfully implemented an equal opportunities policy in their company. Second, we try to offer indirect evidence for the underlying mechanism of the moral self-image. Besides testing whether an increased moral self-image can lead to expression of prejudice (study 2), this study is the first to test whether the presence and success of equal opportunities policies in an organization can increase the moral self-image of employees (study 3). Understanding the underlying mechanism is important in order to develop possible interventions and direct the behavior of employees toward more moral behavior.

| EFFEC TIVENE SS OF D IVER S IT Y P OLI CIE S
The Civil Rights Movement in America and Europe resulted in legislation that renders hiring discrimination of racial minorities illegal (Besley & Payne, 2013). Many organizations responded to these laws by implementing diversity policies, also referred to as affirmative action plans (AAPs, Kravitz, 2008). The purpose of these policies is to offer equal employment opportunities to all individuals, regardless of demographic variables, such as gender, age, and race among other.
These policies vary in terms of strength. Whereas some policies only focus on disregarding these demographics in hiring decisions (i.e., equal opportunities policies), others explicitly focus on giving a preferential treatment to individuals from the disadvantaged group (i.e., opportunity enhancement policies). There is much controversy with respect to the latter types of AAPs. Not only is strong preferential treatment only legal under certain conditions (see Pyburn, Ployhart, & Kravitz, 2008), applicants and employees also tend to respond more negatively toward these types of policies (Harrison, Kravitz, Mayer, Leslie, & Lev-Arey, 2006).
Equal opportunities policies are less debated and seem to yield modest positive effects on increasing diversity. For example, Leonard (1990) concluded on the basis of a review of decades of equal opportunity regulation that it helped to increase the participation of minorities. Also, Colarelli and colleagues (2010) provided experimental evidence for an increase in minority representation in the equal opportunities conditions compared to a neutral condition.

| AMB I G U IT Y OPEN S THE DOOR FOR COVERT DISCRIMINATION
Scholars, however, warned that one should not relax strivings for equality as prior studies may have overlooked covert forms of discrimination, which may still be prevalent (Sue, 2010;Sue et al., 2007). Covert discrimination can take two forms. On the one hand, it pertains to less noticeable behaviors, such as being less friendly toward and limiting interaction duration with minority members (King & Ahmad, 2010), or placing chairs at a larger distance when conversing with minority members compared to majority members (Madera & Hebl, 2013). These behaviors are ambiguous in the sense that they are so subtle that they cannot clearly be related to prejudices. Other forms of covert discrimination are more noticeable immoral behaviors that are enacted in an ambiguous context, allowing the actor to explain the behavior in an alternative but equally moral way, and thus to appear moral (Monin & Miller, 2001). The fact that individuals feel they can safely engage in these behaviors while at the same time upholding the impression that they are "nonprejudiced" (Deitch et al., 2003), clears the way for immoral behavior through the mechanism of moral licensing and more specifically moral credentials. These behaviors are the focus of this study.
Moral licensing refers to the idea that prior moral behavior may liberate individuals to subsequently engage in morally dubious behaviors, without fearing to feel and/or appear discredited (Miller & Effron, 2010). A moral license can be acquired through two different mechanisms: moral credits and moral credentials. These mechanisms differ in terms of the underlying rationale and their implications. The moral credits mechanism uses a bank account as a metaphor: past good deeds create credits on the account that can be used to engage in subtle but also overt forms of discriminatory behavior. People feel entitled to do this because they have earned the credits. More support has been found for the moral credentials model, which is also the focus of this study. This model shows that people are mostly more prudent in engaging in discriminatory behaviors and only do this when they can morally rationalize this behavior which means that this behavior is seen as less immoral. In this model, the previous moral behavior thus changes the way subsequent immoral behavior is construed. This credential can be offered by a previous moral act where people could establish oneself as an unbiased person. However, this mechanism may be further facilitated in ambiguous contexts where multiple interpretations of the same behavior are possible, for example, hiring a white applicant in an ostensible hostile work environment for black employees (Monin & Miller, 2001). Not only the previous moral behavior but also this context allows to justify a less moral act as moral, and literature shows that people are especially then more likely to license (Brown et al., 2011;Merritt et al., 2010;Monin & Miller, 2001).
For example, Monin and Miller (2001) showed that participants who employed a black (vs. a white) candidate in a first selection task were subsequently more likely to express stronger prejudices toward black candidates for a position as police officer. Notably, the case presented a hostile work environment for black employees, which allowed participants to morally rationalize their decision. Brown et al. (2011) extended this evidence by including both an ambiguous (i.e., short time slot to solve math equations) and a nonambiguous condition (i.e., long time slot to solve the same math equations). These authors showed that having a moral credential led to more cheating behavior, but only in the ambiguous condition, when cheating could easily be rationalized.
In this study, we argue that employees can derive a moral credential from "good" behavior of the company, such as having successfully installed an equal opportunities policy. Not only are legal rules followed but this also evidences that employees within the company do not discriminate and thus align their behavior with generally accepted social norms (Treviño, Weaver, & Reynolds, 2006). Highlighting this "good behavior" to the employees within the company, may offer them a license to engage in less moral behavior and express more prejudices, especially when it can be rationalized. This idea is also supported by Kouchaki (2011) who showed that merely pointing out to students that they belong to a morally superior group or informing students about prior nondiscriminatory behavior of members of their group, offered them a license to express prejudice toward a Hispanic applicant when the context was ambiguous. Based on this literature, we formulated the first hypothesis.
Hypothesis 1 Highlighting the successful implementation of an equal opportunities policy leads to a stronger expression of racial prejudices compared to a neutral control condition, but only when the context is ambiguous.

| THE ROLE OF AN IN CRE A S ED MOR AL S ELF-IM AG E
The moral licensing literature suggests that the moral licensing effect can be explained by increases in an individual's moral self-image. Behaving morally highlights the moral self-image of an individual (Blasi, 1980;Monin & Miller, 2001). In the same way, employees who are informed about a successful equal opportunities policy may experience a heightened level of moral self-image.
This elevated feeling of being moral may provide a lens through which the neutral or even moral interpretation of less moral behavior in an ambiguous context becomes more likely (Mullen & Monin, 2016).
Few studies have provided evidence for increases in moral selfimage as an underlying mechanism of moral licensing. Sachdeva, Iliev, and Medin (2009) primed a treatment group with nine moral traits to make them feel more moral, which subsequently reduced donation amounts compared to people that were not primed with moral traits. Cornelissen, Bashur, Rode, and Le Menestrel (2013) showed that people who were asked to recall own moral actions were more likely to cheat on the next occasion, which was explained by a higher moral self-image compared to people who recalled immoral behavior. Kouchaki (2011) showed that an increased moral self-image could explain the discriminatory behavior of the morally superior group compared to the control group.
Based on this literature, we formulated the following hypotheses: Hypothesis 2 An increased moral self-image leads to a stronger expression of racial prejudices.

Hypothesis 3 Employees perceptions of the extent to which their organization successfully implemented an equal opportunities policy is positively related to their moral self-image.
We designed three studies to test our hypotheses. In study 1, we set up a 2 × 2 design, manipulating information about a successful implementation of an equal opportunities policy as well as ambiguity of the context. We suggested that the expression of racial prejudices would be highest when highlighting a prior successful policy implementation in combination with high contextual ambiguity (hypothesis 1). In study 2, we tested whether higher levels of moral selfimage led to a stronger expression of racial prejudices (hypothesis 2). In study 3, we tested whether the presence of a successful policy within the organization increases the moral self-image of participants (hypothesis 3).

| Sample and procedure
Participants were approached in two ways. Students who enrolled in graduate courses on business administration were invited to participate during class. Other participants were invited via social networks (Facebook and LinkedIn). All participants completed the experiment online, either in Dutch or in English. Translational equivalence was ensured by the translation back-translation method (Beaton, Bombardier, Guillemin, & Ferraz, 2000). Two native Dutch speakers fluent in English translated the English study materials to Dutch. Both translators then discussed differences and agreed on a Dutch version. After that, two native English speakers fluent in Dutch "back-translated" the Dutch version into the English version and discussed differences in the final translation. Notably all translators were unfamiliar with the study.
In order to encourage participation, people could take part in a raffle to win cinema tickets. In total, 115 people (44 male; M age = 29.95, SD age = 10.05, 86.1% Caucasian/white vs. 13.9% other racial groups) participated in this study. Most of the participants were students (42%) or employees (24%).
Participants were informed that the purpose of the study was to improve decision making in personnel selection. In order to ensure ecological validity, participants were asked to imagine working for a recruitment team and were presented with a description of a job vacancy (i.e., trainee). Participants were then randomly assigned to one of four experimental conditions.

| Experimental design and materials
We set up a 2 (experimental vs. control) × 2 (nonambiguous vs. ambiguous) between subjects randomized controlled trial. The experimental condition of the policy manipulation reads as follows: Imagine you work as a personnel recruiter for the HR Interim company "Personnel Success." Your agency acts worldwide and provides recruitment services to companies of all sizes and in all industries. The agency recruits jobseekers from 23 different countries to find the best match for the customer's vacancies. You and your colleagues were successful in implementing an equal opportunities policy and improved diversity.
In the control condition, the last sentence was omitted. Participants were asked to imagine working for this company and to recruit a new trainee with the following competencies: be responsible and trustworthy, have a high performance, and show quick intelligence enabling them to make quick decisions in difficult situations. Trainees were expected to be highly motivated and to have a strong performance orientation.
The manipulation of contextual ambiguity was inspired by the police hiring scenario of Monin and Miller (2001)  In the low-ambiguity control condition the text in bold was omitted.

| Measures
To measure the expression of prejudices, we followed the approach of Monin and Miller (2001), asking participants to indicate the suitability to either employ a black or a white candidate for the job. Participants answered the question "Do you feel that this specific position is better suited for a black or a white candidate?" on a 7-point Likert-scale (−3 = certainly better for a black person, 3 = certainly better for a white person). If participants do not want to express prejudices toward either black or white candidates, they would indicate an answer around the middle point of the scale. To prevent first-order effects, the order in which "white" and "black" were mentioned in the question was randomized. All answers were coded so that a higher score meant a stronger preference for a white candidate.

| Analysis
To test hypothesis 1, we fitted a Univariate General Linear Model with the expression of prejudice as dependent variable. The manipulated variables policy (0 = no policy, 1 = equal opportunities policy) and ambiguity (0 = nonambiguous, 1 = ambiguous) and their interaction term were included in the analysis. We reported effect sizes in terms of partial eta-squared. According to Ferguson (2009), values of 0.04, 0.25, and 0.64, respectively, indicate small, medium, and large effect sizes.
Although participants were randomized across conditions, we first wanted to rule out the possibility that confounding variables could have influenced the results. We therefore checked for possible effects of language and the administration mode, since the survey was administered in English (=0) and Dutch (=1), and via social networks (=0) and in the classroom (=1) on the expression of prejudices. We also checked for effects of racial group of the participants

| Manipulation checks
We first checked for possible sequence effects of the words "black" and "white" in the measure of the dependent variable. We did not find a significant effect of the position of the words "black" or "white" on the expression of prejudices, F(1, 114) = 1.26, p = 0.27, 2 p = 0.01, indicating that this randomization did not distort the findings. Second, we checked whether the scenario was easy to understand, in order to ensure that the participants could easily comprehend the manipulations and answer the questions accordingly. The results showed a slightly above "average" level of comprehensibility (M = 6.14, SD = 2.65, MIN = 1; MAX = 10). We ran the analysis with and without this variable as a covariate but it did not change the results. For reasons of parsimony, we therefore only report the results without this variable included.

| Pretest
In a separate pretest, we tested whether our policy condition increased the moral self-image of participants compared to the neutral condition as this is supposed to be the underlying mechanism of the effect of the policy on covert discrimination. 80 participants (35 male; M age = 30.24, SD age = 8.96, 91.3% Caucasian/white vs. 8.7% other racial group, 54% employees, 14% had experience with taking employment decisions) were recruited via social network sites (Facebook and LinkedIn). Participants were randomly as-

| Descriptive statistics
The mean level of the expression of prejudices was 0.33 (SD = 0.81).
This indicates that on average participants revealed a very weak preference for a white candidate. Seventy-five percent of the participants indicated to have no preference at all and chose the midpoint of the scale.

| Hypothesis testing
The main effect of the policy on the expression of prejudice was not significant, F(1, 111) = 0.58, p = 0.45, 2 p = 0.01., while the main effect of ambiguity was significant, F(1, 111) = 8.00, p < 0.01, 2 p = 0.07. Participants in the ambiguous condition (M = 0.56, SD = 0.11) more strongly expressed racial prejudices than participants in the nonambiguous condition (M = 0.14, SD = 0.10). We also found a significant interaction effect between policy and ambiguity, F(1, 111) = 4.02, p < 0.05, 2 p = 0.03. Figure 1 suggests that the policy only leads to a stronger expression of prejudices in the ambiguity condition

| Discussion
The results of this study supported the idea that highlighting the presence of a successful equal opportunities policy may have an adverse effect. Indeed, the findings illustrate that in the presence of a policy, participants were more likely to express prejudices but only when the context allowed to morally rationalize this behavior. In the next studies, we further (indirectly) test whether higher levels of moral self-image are positively associated with the expression of prejudice.

| Sample and procedure
We approached HR departments of large companies in the Netherlands, asking for their participation in this study. In addition, HR representatives within the professional network of the researcher were contacted through email, LinkedIn, and Facebook.
Finally, we sent emails to contact persons in job ads, advertised on F I G U R E 1 Interaction between policy and ambiguity on the expression of prejudices nationalevacaturebank.nl. After deleting incomplete answers as well as three respondents who did not complete the informed consent, we collected data from 62 recruiters (21 male; M age = 40.44, SD age = 12.23, M years experience with recruitment = 11.93, SD years experience with recruitment = 9.86).
In order to align this study with the first study and to increase ecological validity, we introduced a scenario where participants were asked to image working for a recruitment agency (i.e., Telecom BV) with an equal opportunities policy. We showed the policy, adopted from Kaiser et al. (2013). We did not mention whether this policy was successful or not, in order to allow substantial variation in moral self-image. The policy reads as follows: Telecom BV believes that creativity, growth and innovation can only be achieved through collaboration between people with different experiences, perspectives and cultural backgrounds. We have based our policy and our practices on this philosophy. In order to better serve our customers and to create a unified work environment, we strive to: • promote trust, mutual respect, and dignity among employees • attract, develop, and reward talented employees • encourage collaboration between employees with different backgrounds, cultures, and ethnicities.
In accordance with our philosophy, Telecom BV motivates its employees to do their best and provide us with a competitive advantage in the market. Telecom BV does not discriminate against employees on the basis of race, religion, gender, sexual preference, nationality or age.
Respondents were then asked to complete the moral self-image scale, measured in the same way as in study 1 (Cronbach's α = 0.93). Next, they were randomly assigned to an ambiguous versus neutral condition.
This manipulation was identical to the manipulation in study 1. At the end of the survey, we measured the expression of prejudices with the same question as in study 1, followed by a measure of moral identity and demographic variables. We replaced the black candidate by a Moroccan candidate in this study as we know that prejudices are especially strong with respect to this group in the Netherlands (Derous, 2011).

| Measurements
Since the literature suggests that people with a strong moral identity may not show licensing effects (Mullen & Monin, 2016), we included moral identity as a control variable in the analysis. We used the 10item scale of Aquino and Reed (2002) to measure moral identity, rated on a 7-point Likert-scale (1 = totally disagree, 7 = totally agree).

| Analysis
To test hypothesis 2, we fitted a Univariate General Linear Model to the data with the expression of prejudice as dependent variable. The manipulated variable ambiguity (0 = nonambiguous, 1 = ambiguous) was entered as a fixed term and moral self-image was entered as a covariate. Also the interaction term between ambiguity and moralself-image was included in the analysis. Moral identity was entered as a covariate. Like in the previous study, we assessed effect sizes in terms of partial eta-squared.

| Manipulation checks
We again checked for possible sequence effects of the words "Morrocan" and "white" in the measure of the dependent variable. We did not find a significant effect of the position of the words "Morrocan" or "white" on the expression of prejudices, F(1, 60) = 2.67, p = 0.11, 2 p = 0.04, indicating that this randomization did not distort the findings.

| Descriptive statistics
The mean level of the expression of prejudices was 0.31 (SD = 1.02).
This indicates that on average participants revealed a very weak preference for a white candidate. Sixty-three percent of the participants indicated to have no preference at all and chose the midpoint of the scale.

| Hypothesis testing
The results of this study supported H2, suggesting that higher levels of moral self-image related to a stronger expression of prejudices.

| D ISCUSS I ON
The results of this study supported the idea that increased levels of moral self-image may have a negative effect in the sense that these participants were more likely to express racial prejudices, particularly in an ambiguous context. In the next study, we test whether and when an equal opportunities policy may increase the moral self-image of employees. We tested this in a field study where employees were asked to reflect upon the presence and successfulness of an equal opportunities policy within their company. We assessed whether the presence and the success of equal opportunities policies were positively associated with the moral self-image of employees (hypothesis 3). This study added ecological validity to the previous scenario studies. 9 | S TUDY 3 9.1 | Method

| Procedure and participants
Participants were recruited via Prolific, an online tool to collect data.
All questionnaires were administered in English. We again framed the questionnaire as a study on decision-making in selection procedures. One hundred and two people participated in the study (52 male; M age = 35.02, SD age = 9.50, all white). All participants were employees and 32.4% indicated that taking employment decisions (i.e., selection and promotion decisions) was part of their job. After completing the informed consent, respondents were asked to indicate whether the company they worked for had an equal opportunities policy and the extent to which this policy was successful, followed by the same measure of moral self-image as in the previous studies (Cronbach's α = 0.89).

| Presence of equal opportunities policies
To assess whether the company the respondents work for had implemented equal opportunities policies, we first offered a definition of an equal opportunities policy, followed by three items. The definition read as follows: Equal opportunities policies are all organizational initiatives aimed at giving the same opportunities for employment, pay and promotion to everyone, without discriminating against particular groups (e.g., ethnic minorities, women, handicapped, older people, etc.).
Respondents were then asked the following questions: "My organization has implemented equal opportunity policies aimed at …," followed by: " … giving the same opportunities for employment to everyone without discriminating against particular groups," "… giving the same opportunities for promotion to everyone without discriminating against particular groups," and "… giving the same opportunities for pay to everyone without discriminating against particular groups." All three items were rated on a 7-point Likert scale (1 = strongly disagree, 7 = strongly agree). Cronbach's α for this scale was 0.84.

| Successfulness of equal opportunities policy
The successfulness of the policy was assessed with five items, answered on a 7-point Likert scale (1 = reduced to a large extent, 7 = increased to a large extent). Since in some companies, an equal opportunities policy did not exist and/or employees were not aware of it, we also offered the option "I do not know." The items were: "In your perception, have equal opportunity policies changed the share of the following groups in your organization …." We asked this question with respect to women, older people, racial minorities, and religious minorities. We also included an item on "other groups" but since 46% of the respondents marked the option "I do not know" for this question, we left it out of the scale. Cronbach's α for this fouritem scale was 0.80, yielding 86 valid responses.
Since the literature showed that this variable could be related to a lower expression of racial prejudices (Chao, Wei, Spanierman, Longo, & Northart, 2015), this could be a confounding variable in this study. As in the previous study, we also controlled for moral identity, measured with the same scale of Aquino and Reed (2002, Cronbach's α = 0.75).

| Analysis
To test hypothesis 3, we ran a linear regression analysis with moral self-image as the dependent variable. In step 1, openness to diversity and moral identity were entered. The measurements of presence and successfulness of equal opportunities policies within their company were entered as independent variables in step 2.

| Descriptive statistics and intercorrelations
The mean level of the presence of equal opportunities policies was 5.91 (SD = 0.98), indicating that for most respondents, the company they worked for had equal opportunities policies. The mean level of success of equal opportunities policies was 3.61 (SD = 0.76). 49% of the respondents indicated that the equal opportunities policies in their company increased diversity at least to some extent. The mean level of moral self-image of the respondents was 5.64 (SD = 1.12).
We correlated the measures of presence of equal opportunities policies and successfulness of the policies with moral self-image. The presence of policies did not show a significant correlation with moral self-image (r = 0.04, p = 0.70), but working for a company that had increased the share of minority members with equal opportunities policies was positively correlated with the moral self-image of the employees (r = 0.26, p = 0.02).

| Hypothesis testing
The results of the linear regression analysis supported hypothesis 3, in the sense that the relationship between the successfulness of the policy and moral self-image was significantly positive, ß = 0.25, p < 0.05, 2 r = 0.06. The relationship with the presence of equal opportunity policies was not significant, ß = 0.19, p = 0.08, 2 r = 0.03, and also not with the control variables openness to diversity, ß = −0.06, p = 0.59, 2 r = 0.00, and moral identity, ß = 0.07, p = 0.49, 2 r = 0.00.

| Discussion
In this field study, we tested whether the presence of (successful) equal opportunities policies increases the moral self-image of employees in a real field context, adding ecological validity to the first 2 scenario studies. The results showed that not the mere presence but rather the successful implementation of policies correlated positively with the moral self-image of the employees. We based our reasoning on the concept of moral licensing (Monin & Miller, 2001) and suggested that highlighting to employees that their organization was successful in increasing the diversity of a workforce due to an equal opportunities policy licenses covert forms of discrimination. Moreover, we argued that this effect may be driven by an increased moral self-image of the employees.

| Main findings and theoretical implications
We found empirical evidence for these expectations in three studies. In line with the literature on moral licensing (e.g., Kouchaki, 2011;Monin & Miller, 2001) we showed that highlighting the successful implementation of an equal opportunities policy increased the expression of prejudices in an ambiguous (as compared to nonambiguous) context (see Brown et al., 2011). We therefore show that moral licensing is not only driven by own moral deeds of individuals, or their membership in morally superior groups, but can also occur due to successful equal opportunities policies. Moreover, we provide additional evidence for the moderating role of the contextual ambiguity. This finding also adds to the literature on the effectiveness of equal opportunity policies, by showing that even successful equal opportunities may backfire, because they can promote covert forms of discrimination.
The literature on moral licensing suggests that this may be due to elevated levels of moral self-image following the manipulation of the policy. We therefore tested in a second study whether higher levels of moral self-image led to an increased expression of prejudices and found evidence for this positive relationship, particularly in the ambiguous condition as predicted.
We conducted an additional third study to compensate for the lack of ecological validity of studies 1 and 2 and also to disentangle Although we did not predict a main effect of contextual ambiguity, we found in study 1 that participants were generally more likely to express prejudices when the context was described as hostile for black employees, independent from the presence versus absence of a policy. This finding is in line with the justification-suppression model of prejudice (Crandall & Eshleman, 2003).
This model suggests that behaving morally is costly to individuals because they need to suppress immoral thoughts. However, people will cease to suppress these thoughts as soon as they can justify the expression of it. Ambiguous contexts allow for these justifications as people can interpret the behavior in these contexts in multiple ways, allowing also for a moral interpretation. In the scenario in this study, participants could interpret the expres-

| Limitations and future research
The present study is a first step toward understanding how successful equal opportunity policies can stimulate instead of constrain covert forms of discrimination. However, this study, is not without limitations.
First, we suggested that the mechanism of moral self-image may explain the link between the policy and the expression of prejudices. We offered indirect support for this mechanism by showing in two separate studies-that higher levels of moral self-image are positively associated with the expression of prejudice (study 2) and that successes of equal opportunities policies increases the moral self-image of employees (study 3).
We did not include a direct test of the mediation of moral selfimage. We believe that measuring moral self-image in between the manipulation of the independent variable and the measurement of the dependent variable, could distort the findings in the sense that the measurement of moral self-image in itself may manipulate moral self-image and influence subsequent judgment and decision-making (see Aquino, McFerran, & Laven, 2011). However, looking for ways to more directly test the underlying mechanism of moral licensing is an interesting avenue for future research.
Second, we described in the scenario that the company was successful at implementing an equal opportunities policy. In the policy manipulations that we used in study 1, it was not clear whether the policy just increased diversity among employees of "Personnel Success" or whether this policy also applied to its customers. This variation in possible interpretations of the scenario may have been a confounding variable in the design and should be controlled for in future studies.
Third, we investigated the effect of a successful equal opportunities policy on overt and covert forms of discrimination. We worked with this type of AAPs since they are least debated and yield modest positive results in terms of overt discrimination (Colarelli, Poole, Unterborn, & D'Souza, 2010;Leonard, 1990). Future research could consider other forms of AAPs, such as preferential treatment. This type of policy may induce even stronger effects as employees may interpret given a preferential treatment to minorities even as more moral behavior than treating them equally raising their level of moral self-image more strongly, and thus offering a stronger license for the expression of prejudices.
In studies 1 and 2, we relied on scenario studies in which participants were asked to imagine working as a recruiter for a company with(out) a successful equal opportunities policy and where there is (or is not) a hostile work environment for racial minorities.
We chose this design as it allows for a powerful manipulation of the experimental variables and to control for possible confounding variables and thus to make causal claims (Antonakis, Bendahan, Jacquart, & Lalive, 2010). In this way, we can start to understand how information on policies and contextual ambiguity influence the expression of prejudices in a controlled setting. We also offered some preliminary evidence that also in real organizations, the presence of a successful equal opportunity policy may increase the moral self-image of individuals. It is, however, important for future research to study the moral licensing effect in more complex and ecologically valid environments.
A related limitation is that the dependent variable in studies 1 and 2 only assessed the expression of prejudices, following the study of Monin and Miller (2001). We did not ask the participants to make a concrete choice between a black/Morrocan and a white candidate.
Including this variable would allow to draw more valid conclusions on discrimination in hiring decisions.

| Practical implications
The results of these studies have important practical implications for organizations who put a lot of effort in implementing an equal opportunities policy. Specifically, the results warn organizations to prevent putting too much emphasis on the success of their policies as this may encourage recruiters to engage in more covert forms of discrimination. The results suggest that the danger lies in an increased moral self-image of the employees. In order to prevent this perverse effect to happen, organizations could react in different ways. First, they can be more prudent with communicating the success of the policy and pursue investments in employees motivation to execute the policy.
Second, they may intensify efforts to reduce the ambiguity of the selection context. This can be done by structuring the selection context, for instance, by basing the selection decision on a job analysis, by defining clear competences that are required from the candidates, and by leaving less room for lenient decision-making toward nonstereotyped candidates (Kutcher & Bragger, 2004).

| CON CLUS ION
In this paper, we studied whether the successful implementation of an equal opportunities policy could lead to a stronger expression of racial prejudices. We further provided indirect evidence that this effect could be explained by an increase in the moral self-image of the participants.