Abstract
Extant research suggests subtle, interpersonal forms of discrimination, though often normalized and overlooked, may be just as detrimental to targets as compared to more traditional, overt forms of discrimination. To further examine this question, we meta-analyzed the current literature to estimate the relationship between discrimination and a host of psychological, physical health, and work-related correlates as a function of its form (subtle or overt). Analysis of 90 effect sizes suggested that subtle and overt forms of discrimination hold relationships of comparable magnitude with a host of adverse correlates. By demonstrating that these two forms of discrimination are not differentially related to relevant outcomes, our findings call into serious question the pervasive belief that subtle discrimination is less consequential for targets as compared to overt discrimination (Landy, 2008; McWhorter, 2008). Taken together, our results suggest that subtle discrimination is at least as important to consider and address as its overt counterpart. Implications for organizational scholars and practitioners are discussed.
Workplace discrimination is said to occur when individuals from a stigmatized group “are put at a disadvantage in the workplace relative to other groups with comparable potential or proven success” (Dipboye & Halverson, 2004: 131). Until recently, the term discrimination conjured up images of workplace behaviors that reflect “blatant antipathy, beliefs that [members of stereotyped groups] are inherently inferior, [and] endorsement of pejorative stereotypes” (Cortina, 2008: 59). These old-fashioned acts of prejudice, commonly referred to as overt discrimination, can be contrasted with subtle discrimination (e.g., Benokraitis, 1997; Brief & Barsky, 2000; Brief, Buttram, Elliott, Reizenstein, & McCline, 1995; Deitch, Barsky, Butz, Brief, Chan, & Bradley, 2003; Dipboye & Halverson, 2004), which encompasses actions that are ambiguous in intent to harm, difficult to detect, low in intensity, and often unintentional but are nevertheless deleterious to target employees (Cortina, 2008; Rowe, 1990).
Given the significant costs associated with workplace discrimination including worsened employee attitudes and increased turnover intentions (King, Hebl, George, & Matusik, 2010) and litigation costs (Goldman, Gutek, Stein, & Lewis, 2006), many organizations have undertaken initiatives to reduce the experience of discrimination for their employees (McKay, Avery, & Morris, 2008; Wentling & Palma-Rivas, 2000). Nevertheless, racial and gender inequalities in organizations persist (e.g., Benokraitis, 1997; Brief et al., 1997). One reason for this continued inequality may stem from the tendency of these diversity initiatives to target overt, easily recognizable forms of discrimination and to overlook subtle, interpersonal discrimination (Shih, Young, & Bucher, 2013). While instances of overt discrimination can be quelled with organizational or legal policies, subtle and seemingly benign occurrences of discrimination may prove particularly pernicious by both compromising an organization’s effort to facilitate a supportive diversity climate and remaining ambiguous, thus escaping castigation. In an effort to pinpoint the root of inequity between stigmatized and nonstigmatized employees, organizational researchers (e.g., Cortina, 2008; Deitch et al., 2003; Dovidio & Hebl, 2005) have called for a closer examination of subtle discrimination.
We contend that the ramifications of subtle discrimination are at least as substantial, if not more substantial, than the consequences of overt discrimination for three reasons. First, subtle discrimination is particularly deleterious given the difficulty in its identification and assessment (Hebl, Foster, Mannix, & Dovidio, 2002; King, Dunleavy, et al., 2011). Indeed, attributional ambiguity theory predicts that negative feedback will be attributed to prejudiced evaluators in clear but not ambiguous situations. Thus, targets of overt discrimination can easily externalize the negative experience to discrimination (e.g., “it’s not my fault they are prejudiced”) whereas targets of subtle discrimination may instead make internal attributions (e.g., “it’s not them, it’s me”). Experimental evidence confirms this expectation, showing that Black targets attribute negative feedback to prejudiced evaluators when their own race is observable but not when it may be hidden. This phenomenon has implications for self-esteem, self-regulation, and task performance (Crocker, Voelkl, Testa, & Major, 1991; Salvatore & Shelton, 2007; Singletary, 2009). Thus, harmful actions with ambiguous intent might be even more confusing and stressful for targets as compared to explicitly discriminatory actions.
Second, because subtle discrimination is often more difficult to detect than overt discrimination (Hebl et al., 2002), targets may experience subtle discrimination more negatively than overt discrimination as a function of the sheer fact that there are not as many clear options for reporting and/or remedying this type of treatment. Many organizations have formal policies in place for reporting overt discriminatory behaviors (Hebl et al. 2002); however, the means by which subtle discrimination can be reported and addressed are less clear.
Third and finally, subtle discrimination may be more damaging for targets because of its higher frequency and thus the chronic nature of its effects. Indeed, extant research has argued that one reason for the particularly damaging impact of subtle discrimination lies in its pervasiveness, whereas overt discriminatory behavior may occur less often (Van Laer & Janssens, 2011). This notion is consistent with research showing that chronic stress is a stronger predictor of depressive symptoms as compared to acute stress (McGonagle & Kessler, 1990). Furthermore, studies that have examined both subtle and overt forms of discrimination tend to show that participants report experiencing subtle discriminatory behaviors more frequently overall as compared to overt discrimination (Utsey, Chae, Brown, & Kelly, 2002; Utsey & Ponterotto, 1999; Yoo, Steger, & Lee, 2010).
In order to build an understanding of the comparative effects of overt and subtle discrimination, we use a meta-analytic approach to examine the relationship between each form of discrimination and a host of important correlates relevant to both targets and their organizations. Specifically, we investigate the magnitude of the link between subtle discrimination and psychological, physical health, and work-related correlates in comparison to the magnitude of the relationship between overt discrimination and these correlates. While our primary focus is discrimination that manifests in workplace settings, we broaden the scope of our review beyond studies that have been conducted in workplace settings to include studies that have measured discrimination in nonwork (e.g., educational, health care) settings. This decision was driven by evidence from the work-life literature suggesting events occurring outside of work are likely to spill over and influence experiences that take place inside the workplace and vice versa (Ford, Heinen, & Langkamer, 2007; Grzywacz & Marks, 2000). Thus, the physical and psychological impact of a discriminatory experience outside of work likely carries over to the workplace to affect physical, psychological, and job-related outcomes. In the following sections, we review the literature on overt and subtle discrimination and in doing so, clarify the definition of each form of discrimination, providing examples of the various ways in which subtle and overt discrimination were operationalized in our primary studies. Finally, we situate our meta-analysis in the context of recent meta-analyses on discrimination and describe the process of conducting the current meta-analysis.
Defining Subtle and Overt Discrimination
To build understanding of the broader distinction between subtle and overt discrimination, Hebl and colleagues’ (2002) distinction between interpersonal discrimination and formal discrimination provides a useful starting point. These authors define interpersonal discrimination as more subtle in nature, involving “nonverbal, paraverbal, and even some of the verbal behaviors that occur in social interactions” and are not prohibited by law (Hebl et al., 2002: 816). Interpersonal discrimination was operationalized by interaction length, word count, perceived interest of potential employer, and the extent to which a potential employer was “helpful,” “standoffish,” “nervous,” “motivated to end the conversation prematurely,” “avoidant of eye contact,” and “hostile” (Hebl et al., 2002: 819). These types of behaviors can be contrasted with formal discrimination, which they define as “discrimination in hiring, promotions, access, and resource distribution … that in many states is illegal … [and for which] there are often organizational laws, company policies, or social norms against” (Hebl et al., 2002: 816).
In addition to interpersonal discrimination (Hebl et al., 2002), subtle discrimination has been examined under a variety of labels including microaggressions (Sue, Bucceri, Lin, Nadal, & Torino, 2009), incivility (Andersson & Pearson, 1999), everyday racism (Essed, 1995), everyday sexism (Swim, Hyers, Cohen, & Ferguson, 2001), and benevolent sexism (Glick & Fiske, 1997). Scholars have conceptualized subtle discrimination as encompassing behaviors that are seemingly normal, natural, or acceptable (Benokraitis, 1997), often unintentional, perceived as trivial and harmless, and not unlawful (Hebl et al., 2002; King, Dunleavy, et al., 2011). Subtle discrimination has further been described as “interpersonal discrimination that is enacted unconsciously or unintentionally and that is entrenched in common, everyday interactions, taking the shape of harassment, jokes, incivility, avoidance, and other types of disrespectful treatment” (Van Laer & Janssens, 2011: 1205). In light of previous literature on this topic, we define subtle discrimination as negative or ambivalent demeanor and/or treatment enacted toward social minorities on the basis of their minority status membership that are not necessarily conscious and likely convey ambiguous intent. Compared to overt discrimination, subtle discrimination is less likely to be unlawful. In addition, uninvolved bystanders would exhibit more hesitation and experience more affective and cognitive ambivalence when labeling subtle discrimination as “discrimination” relative to its overt counterpart. Examples of subtle discrimination measures used in our primary studies include items such as “you were treated with less courtesy than others,” “people acted as if they were better than you,” “others expected your work to be inferior,” “others reacted to you as if they are afraid or intimidated,” “being mistaken for someone who serves others (e.g., janitor, maid, etc.),” and “sales people/clerks did not say thank you or show other forms of courtesy and respect (e.g., put your things in a bag) when you shopped at some White/non-Black business” (Harrell, 1997; Utsey & Ponterotto, 1996; Williams, Yu, Jackson, & Anderson, 1997; the full list of discrimination measures used in the primary studies is available as an online supplement).
In contrast, overt discrimination has been labeled as bullying (Fox & Stallworth, 2005), old-fashioned racism (Virtanen & Huddy, 1998), hostile sexism (Glick & Fiske, 1997), and formal discrimination (as reviewed above, Hebl et al., 2002). According to scholars, overt discrimination occurs when “differential and unfair treatment is clearly exercised, with visible structural outcomes” and takes the form of behaviors that are unconcealed, intentional, and easily recognizable and are directed at a target on the basis of his or her stigmatized characteristics (Van Laer & Janssens, 2011: 1205). Engaging in these behaviors is generally considered unacceptable by society and is often proscribed in employment contexts as echoed in Hebl and colleagues’ (2002) conceptualization of formal discrimination. Thus, in light of previous research, we define overt discrimination as explicitly negative demeanor and/or treatment enacted toward social minorities on the basis of their minority status membership that are necessarily conscious. Furthermore, as compared to subtle discrimination, overt discrimination is more likely to be unlawful and less likely to produce hesitation and ambivalence within an uninvolved bystander to label the treatment as “discrimination.” Examples of overt discrimination measures used in our primary studies include items such as “someone at work makes derogatory comments about your ethnicity,” “someone at work uses ethnic slurs to describe you,” “someone at work fails to give you information you need to do your job because of your ethnicity,” “someone made unwanted attempts to stroke or fondle you,” “you have been unfairly fired or denied a promotion,” “for unfair reasons, you have not been hired for a job,” “you have been unfairly stopped, searched, questioned, physically threatened, or abused by the police,” “you were refused an apartment or other housing; you suspect it was because you were Black,” “you were passed over for an important project although you were more qualified and competent than the White/non-Black person given the task,” and “you have discovered that the White/non-Black person employed in the same capacity as you with equal or less qualifications is paid a higher salary” (Schneider, Hitlan, & Radhakrishnan, 2000; Schneider, Swan, & Fitzgerald, 1997; Thompson, 1999; Utsey & Ponterotto, 1996).
Outcomes of Overt and Subtle Discrimination
Although recent literature has begun to focus primarily on manifestations and consequences of subtle discrimination, research continues to demonstrate that explicit, discriminatory acts still occur and have damaging effects on stigmatized individuals’ work-related outcomes and attitudes (Hughes & Dodge, 1997), physical well-being (Guyll, Matthews, & Bromberger, 2001), and mental health (Motoike, 1995; Noh, Kaspar, & Wickrama, 2007; Schneider et al., 1997). Empirical studies have documented the negative effects of explicit discrimination on work-related issues such as lower earnings and decreased job satisfaction (Hughes & Dodge, 1997). In addition, psychological outcomes, such as decreased positive affect, self-esteem (Motoike, 1995; Noh et al., 2007), and psychological well-being (Schneider et al., 1997) and increased depression have been related to overt discrimination (Crouter, Davis, Updegraff, Delgado, & Fortner, 2006). Further, physical health outcomes, such as cardiovascular problems (Guyll et al., 2001; Thompson, 1999), have been positively linked to overt discrimination. Taken together, empirical research and theoretical research indicate that although the incidence of overt discrimination may be decreasing, its continued negative impact on stigmatized individuals has not been eradicated.
As with overt discrimination, subtle acts of discrimination have been shown to negatively affect psychological health (Gifford, 2009; Lim & Cortina, 2005; Utsey et al., 2002), physical health (Lewis et al., 2006; Miscally, 2009; Thompson, 1999), and work-related outcomes (Gifford, 2009; King, Shapiro, Hebl, Singletary, & Turner, 2006; Stewart, King, Botsford, Gilrane, Hylton, & Jones, 2010). Several studies have demonstrated that subtle forms of discrimination are associated with psychological outcomes such as decreased well-being (Lim & Cortina, 2005), self-worth (Gifford, 2009), and quality of life (Utsey et al., 2002). Similarly, with regard to health outcomes, research has shown that subtle discrimination may increase alcohol and illicit drug use (Miscally, 2009) and negatively influence cardiovascular health (Lewis et al., 2006; Thompson, 1999).
Evidence also suggests that subtle discrimination can have job-related implications for stigmatized employees and their organizations. For instance, one study found that in response to subtle discrimination, female police officers disengaged from their jobs in an effort to protect self-esteem (Tougas, Rinfret, Beaton, & de la Sablonniére, 2005). Furthermore, employees who attributed negative interpersonal interactions to their stigmatized identity experienced decreased organizational commitment, lower confidence in their ability to achieve professional goals, and poorer relationship quality with supervisors (Gifford, 2009). In addition, extant research suggests benevolent sexist behaviors that are seemingly subtle may negatively influence performance (Dardenne, Dumont, & Bollier, 2007; Vescio, Gervais, Snyder, & Hoover, 2005). Finally, research has shown that interacting with stigmatized customers in an interpersonally negative manner may affect the organization’s bottom line by decreasing purchasing behavior and customer loyalty (King et al., 2006). In addition to the studies that solely examine subtle discrimination, recent research has begun to directly compare the two types of discriminatory treatment.
Comparison of Subtle and Overt Discrimination
Given postulations that subtle discrimination may be equally or even more damaging than overt discrimination (Cortina, 2008; Dovidio, 2001; Motoike, 1995), researchers have conducted experimental studies to contrast the two forms of discrimination (Hebl et al., 2002; Hebl, King, Glick, Singletary, & Kazama, 2007; Hughes & Dodge, 1997; Salvatore & Shelton, 2007; Singletary, 2009; Singletary & Hebl, 2009). Some research suggests that both types of discrimination may have equally harmful consequences for stigmatized individuals. For instance, Hughes and Dodge (1997) found that interpersonal prejudice (i.e., subtle discrimination) and institutional (i.e., overt) discrimination were both negatively related to job satisfaction for African American women. More recent studies, however, provide evidence that subtle forms of discrimination may be particularly harmful for targets in performance settings. For example, Singletary (2009) demonstrated that women exhibited detriments in performance when treated with subtle discrimination; however, overt discrimination showed no influence on performance. In addition, Salvatore and Shelton (2007) examined the impact of exposure to either ambiguous or blatant, racially biased hiring recommendations on cognitive performance. Their findings indicated that, for African American individuals, viewing subtle discrimination was more detrimental to performance than viewing overt discrimination. Singletary (2009) and Salvatore and Shelton (2007) both cite attributional ambiguity theory (Crocker & Major, 1989) as an explanation for their findings. Specifically, they argue that subtle discrimination may be particularly problematic because deciphering ambiguous discriminatory cues is more cognitively and emotionally taxing than comprehending overt discrimination, which can be easily attributed to prejudiced motives. Together, these studies suggest that the effects of subtle discrimination may be even more damaging to targets than overt discrimination.
Prior Meta-Analyses
As an answer to the call from organizational scholars to address contemporary discrimination (e.g., Cortina, 2008; Deitch et al., 2003; Dovidio & Hebl, 2005), five meta-analyses on discrimination and their outcomes have been conducted within the past 15 years (Bowen, Swim, & Jacobs, 2000; Davison & Burke, 2000; Lee & Ahn, 2011, 2012; Pascoe & Smart Richman, 2009; see also Colella, McKay, Daniels, & Signal, 2012, for a comprehensive review of meta-analytic research on this topic). In contrast to the current study, these works adopted a homogeneous perspective of discrimination, aggregating across its subforms. In other words, potentially differential correlates associated with subtle versus overt forms of discrimination were neither hypothesized nor examined.
Although these meta-analyses advance knowledge of the relationships of overall exposure to discrimination and correlates of importance, they fail to consider the influence of discrimination as conceptualized by frameworks and theory that are more representative in recent years. For instance, Lee and Ahn (2011) examined the mental health outcomes of (general) discrimination for Asians, and a second meta-analysis by Lee and Ahn (2012) investigated the effects of (general) discrimination on mental health, physical health, employment, and educational outcomes for Latina/o populations. Further, two separate meta-analyses focused on (general) discrimination directed at targets solely on the basis of gender. Davison and Burke’s (2000) meta-analysis of simulated employment contexts revealed that women were rated higher than men for female-typed jobs, whereas men were rated higher and offered more compensation than women for male-typed jobs. A complementary meta-analytic study illustrated the existence of formal (i.e., overt) discrimination in field settings by showing that performance ratings were biased against women when men served as raters (Bowen et al., 2000). While providing informative results concerning the work-related outcomes of overt gender discrimination, these meta-analyses do not form a comprehensive review of the multitude of correlates (e.g., mental and physical health) associated with overt and subtle discrimination targeted at a variety of stigmatized individuals.
An additional meta-analysis filled some of the gaps left by the previous meta-analyses by investigating the effects of perceived discrimination—which they defined as the target’s personal experience of discrimination—on psychological and physical well-being without limiting the target population to only one group (Pascoe & Smart Richman, 2009). The primary studies used in this meta-analysis included those that assessed both subtle and overt discrimination; however, the researchers did not distinguish between the two forms. Their results suggested that the experience of discrimination is associated with detrimental mental and physical health outcomes for targets. Specifically, targets were more likely to report suffering from symptoms of depression and psychiatric distress and less likely to report positive well-being as compared to those who did not view themselves as being discriminated against. Furthermore, cardiovascular disease and diabetes were positively linked with perceived discrimination. Although Pascoe and Smart Richmans’s (2009) meta-analysis indicates the alarming need to consider how discriminatory actions affect targets, our study makes two important contributions beyond their work. First, we directly compare the relationship between subtle discrimination and a host of important outcomes to the relationship between overt discrimination and those same outcomes. Second, in addition to examining relationships between discrimination and both physical and psychological outcomes, we consider a range of meaningful workplace outcomes. Given existing research on subtle and overt discrimination, the primary goal of this study is to provide a meta-analytic review of the relationship between discrimination and psychological, physical, and work-related outcomes as a function of its form (i.e., subtle versus overt).
Additional Moderators
Though our primary research question involves the investigation of the form of discrimination (i.e., subtle, overt) as a moderator of relationship between discrimination and its psychological, physical, and work-related correlates—a secondary goal of this study is to examine the presence of additional moderators including minority characteristic, publication date, and study setting. Given the breadth of characteristics along which individuals discriminate (e.g., race, sex, weight, sexual orientation, disability status), it seems reasonable to suggest individuals may experience discrimination differently as a function of the specific marginalized group to which they belong. Indeed, extant research has demonstrated that stereotypes vary in the extent to which they evoke negativity, pity, anger, and fear as a function of the social threat that a particular group is perceived to pose (Cottrell & Neuberg, 2005; Neel, Neufeld, & Neuberg, 2013; Schaller & Neuberg, 2012). This minority characteristic moderator analysis was conducted across all four correlate domains for the observed categories of minority groups in which at least three samples were present. Furthermore, since about half of our primary studies were conducted in nonworkplace settings, we examined the study setting (i.e., workplace, nonworkplace) as a potential moderator of the relationship between discrimination and psychological, physical, and work-related correlates. Nonworkplace studies included participants who either were undergraduates or were recruited from the community using snowball sampling methods. In these cases, the data collection usually involved survey or interviewing methods that took place on a college campus, in a laboratory, online, or via phone. We found this moderator examination particularly important given our primary focus was on the experience of discrimination in the workplace. Finally, scholars have argued that subtle forms of discrimination tend to represent more recent phenomena, whereas overt forms are considered to be more prevalent in past years (Dipboye & Colella, 2005; Dovidio & Gaertner, 1998; Virtanen & Huddy, 1998). This prompted our examination of the effect of publication date of the primary studies investigated, comparing the link between discrimination and its correlates in three arbitrary five-year intervals to examine, for example, whether the relationship between discrimination and its correlates has been increasing, decreasing, or stable over time.
Method
Literature Search
The literature search procedure for this meta-analysis was designed to locate both published and unpublished research. The primary method of identifying relevant studies was through electronic databases, including PsycINFO, PsycExtra, ERIC, and Medline. Given there has not been a major quantitative analysis examining aspects of subtle and overt discrimination and their effects, publication date was not restricted. Keyword searches included terms such as covert discrimination, implicit discrimination, modern discrimination, everyday discrimination, explicit discrimination, traditional discrimination, blatant discrimination, and hostile discrimination. These keywords were included with various outcomes of interest such as health, career success, stress, life satisfaction, well-being, academic achievement, depression, and psychological distress. We reviewed the reference list of each article to identify additional citations that were not revealed by other search means. In addition, we mined the citations of several articles highly relevant to the analysis topic, including recent meta-analyses examining general exposure to discrimination and subsequent outcomes. Finally, we sent messages to e-mail lists (e.g., the Society for Personality and Social Psychology, the Society for Industrial Organizational Psychology) that appeal to frequent contributors to the discrimination research literature requesting copies of in-press or unpublished articles.
Criteria for Inclusion
To be included in the research synthesis, an article needed to meet three criteria. First, the article had to contain data representing the relationship between discrimination targeted at an individual and a correlate that fell into at least one of the following four domains: psychological health (e.g., psychological distress, anxiety, depression), physical health (e.g., smoking habits, cardiovascular health, blood pressure), individual work-related correlates (e.g., satisfaction, attachment, stress), and organizationally relevant correlates (e.g., employee turnover intentions, employee performance, organizational performance). Second, sufficient information had to be available for coders to identify each discrimination-correlate relationship as either subtle or overt in nature. Third, each study must have reported the sample size and a test statistic that could be converted into a correlation between discrimination and the correlate variable. When possible, attempts were made to contact study authors to obtain usable statistics. The final tally of effect sizes reflecting the relationship between subtle discrimination and a correlate or overt discrimination and a correlate (k = 90) were obtained from a total of 44 samples, including 26 journal publications, 11 dissertations, 1 conference presentation, and 1 unpublished study. Note that a few multipart studies provided effect sizes from separate samples.
Coding of Studies
Depending on the relationships reported by particular studies, coded statistics included sample size, correlations, variable means and standard deviations, t tests, or F tests. To ensure that accurate data were retrieved from each study, studies were coded independently by two of the study authors (94.5% agreement). This overall index of agreement was calculated by dividing the number of times two coders actually agreed by the number of opportunities provided for the two coders to agree. Though rare, sources of disagreement stemmed from deciding how many and which discrimination measures to use (e.g., when studies took multiple measures of subtle discrimination) and how many and which outcome measures to use (e.g., when studies assessed both depression and loneliness). All discrepancies between coders were resolved through consensus discussions after which 100% agreement was obtained.
Constructs of Interest
Correlates
In line with previously published meta-analyses (e.g., Eby, Allen, Evans, Ng, & DuBois, 2008; Valentine, DuBois, & Cooper, 2004), variables that were conceptually similar were combined into one of our four correlate domains. This was critical in order in examine the relationships between both types of discrimination and each domain of correlates. Table 1 lists the four categories of correlates investigated, which include individual work correlates, organizationally relevant correlates, physical health, and psychological health. Within these categories are listed specific outcomes examined in primary studies and examples of how these outcomes were operationalized. Note that individual work correlates refer to those outcomes that are primarily relevant to shaping an individual’s experience at work (e.g., satisfaction, attachment, stress), whereas organizationally relevant correlates encompass those outcomes that are largely relevant to the organization’s bottom line (e.g., employee turnover intentions, employee performance, organizational performance).
Examples of Correlates of Discrimination
Grouping variable
The form of discrimination (i.e., overt, subtle) was coded to enable examinations of whether the relationship between discrimination and its correlates varied across the experience of subtle and overt discrimination. As such, positive correlations indicated that higher levels of discrimination were positively associated with adverse outcomes. Specifically, we coded discrimination as subtle when it reflected negative or ambivalent demeanor and/or treatment enacted toward social minorities on the basis of their minority status membership that was not necessarily conscious and for which uninvolved bystanders would exhibit more hesitation and experience more affective and cognitive ambivalence to label the treatment as “discrimination.” Conversely, we coded discrimination as overt when it reflected explicitly negative demeanor and/or treatment enacted toward social minorities on the basis of their minority status membership that was necessarily conscious and less likely to produce hesitation and ambivalence within an uninvolved bystander to label the treatment as “discrimination.” For example, studies reporting effect sizes of the relationship between correlates and microaggressions, selective incivility, interpersonal, covert, implicit, modern, and everyday discrimination were coded as subtle, whereas studies reporting effect sizes of the relationship between correlates and formal, explicit, traditional, blatant, and hostile discrimination were coded as overt. Most primary studies used measures of discrimination that were continuous in nature, ranging on a scale that reflected the frequency with which one was exposed to discriminatory treatment. That is, most studies used measures of discrimination for which “low levels” indicated discrimination never or rarely occurred whereas “high levels” indicated discrimination occurred frequently (over some specified amount of time such as the past 12 months). It is important to note that discrimination was operationalized slightly differently in the lab studies (9% of the primary studies) wherein discrimination was manipulated by the researcher. In these cases, discrimination either occurred (in the discrimination condition) or it did not (in the control condition), so by default the discrimination variable took on a value of either 1 or 0, reflecting the extreme end points of the frequency continuum.
Meta-Analytic Method
Using the Hunter and Schmidt (1990) method, effect sizes were meta-analyzed for discrimination in relation to each correlate domain when there were at least three samples present. Effect size estimates were corrected for predictor and criterion reliability with the mean attenuation factor, which was calculated using sample-size weighted means of the available reliability data. Additionally, 95% confidence intervals were constructed around each meta-analytic effect size, which can be interpreted as the range of values within which we are 95% certain the corresponding population parameter lies. Thus, a confidence interval excluding zero indicates a 95% likelihood that the population parameter is nonzero.
To avoid issues associated with independence of associations, we used a shifting unit-of-analysis approach (Cooper, 1998). Associations within studies were first coded as though they were independent estimates and subsequently averaged to provide one overall estimate from each study. For example, if a single sample reported relationships between subtle discrimination and both income level and career advancement, we first calculated two effect sizes, one representing the relationship between subtle discrimination and income level and the other representing the relationship between subtle discrimination and career advancement. We then estimated the overall effect by averaging these correlations such that the sample provided only one estimate of the relationship between subtle discrimination and individual work correlates.
There were, however, a few studies that examined the impact of both subtle discrimination and overt discrimination. For example, Crouter and colleagues (2006) examined the impact of both interpersonal (subtle) discrimination and institutional (overt) discrimination on depressive symptoms. Thus, we averaged these two correlations in our overall assessment of discrimination on psychological health correlates; however, we separated these two correlations in our subgroup analyses of subtle versus overt discrimination (which is why the k’s and N’s from subtle and overt subgroup analyses do not sum perfectly to the k’s and N’s in the overall analysis). The implication of this decision is that the meta-analytic correlation reflecting the relationship between subtle discrimination and psychological correlates is not completely independent of the meta-analytic correlation reflecting the relationship between overt discrimination and psychological correlates. Extant evidence suggests treating dependent correlations as independent produces similar (and if anything, more conservative) results and is appropriate for rough inferences especially in cases such as ours in which there are only a few examples of this (Hedges, Tipton, & Johnson, 2010; Hunter & Schmitt, 2004). Moreover, this decision is justified in light of our primary research question—the comparison of subtle and overt discrimination. It did not make sense to penalize primary studies for pursuing this same question.
Moderator analyses
Although moderator analyses were conducted on all theoretical relationships of interest, the Q statistic (Hedges & Olkin, 1985) was used to indicate whether reductions in variability of effect sizes accompanied investigations of subgroup differences. For relationships suspected to be influenced by moderators, meta-analytic correlations were computed separately for samples at differing levels of the moderators (Hunter & Schmidt, 2000). As with the overall analyses, three or more samples were required to conduct a moderator analysis. In the case of our primary moderator variable (subtle versus overt discrimination), we report tests of statistical significance for the sake of completeness; however, our primary focus in comparing these effect sizes concerns practical significance, which we discuss as well. To determine if the effect sizes associated with the particular moderators differed significantly, we computed t-statistics following the procedure outlined by Aguinis, Sturman, and Pierce (2008) for evaluating the significance of meta-analytic moderator variables.
Publication bias
Rothstein, Sutton, and Borenstein (2005) suggested that confidence in the robustness and validity of meta-analytic findings is related to the extent to which publication bias affects the results of the study. The software package Comprehensive Meta-Analysis (Borenstein, Hedges, Higgins, & Rothstein, 2005) was used to complete Duval and Tweedie’s (2000a, 2000b) trim and fill analysis (using the symmetry of a distribution of correlations to detect bias and impute “missing” correlations to reestimate the overall effect size). For the trim and fill estimates, Kepes, McDaniel, Banks, and Whetzel (2012) classified bias as negligible when the difference between the meta-analytic mean and a trim and fill adjusted mean estimate is less than 20%, moderate when the difference is between 20% and 40%, and severe when the difference is greater than 40%. These analyses were conducted on observed, not corrected, correlations.
Results
For each of the relationships we investigated in this meta-analysis, we have reported the number of samples on which the estimate is based (k), the total sample size aggregated across studies (N), the mean sample-size-weighted uncorrected correlation (ro), the mean sample-size-weighted corrected correlation (rc), and the upper and lower 95% confidence interval. Additionally, significant Q statistics for relationships between discrimination and all four of the correlate domains suggested the presence of between-study moderators: for individual work correlates, Q(13) = 65.28, p < .01; for organizationally relevant correlates, Q(13) = 62.62, p < .01; for physical health correlates, Q(10) = 30.98, p < .01; and for psychological health correlates, Q(31) = 170.63, p < .01. Finally, the results of our publication bias examination indicated negligible bias on all reported analyses.
Individual Work Correlates
Table 2 lists meta-analytic results for samples investigating the relationship between discrimination and individual work outcomes. The mean sample-size-weighted corrected correlation between overall discrimination (i.e., including both subtle and overt forms) and adverse individual work correlates is .30.
Meta-Analytic Relationships Between Discrimination and Adverse Individual Work Correlates
Note: CI = confidence interval; k = number of samples; N = total number of data points; ro = uncorrected sample-size-weighted mean correlations; rc = weighted sample size mean correlations corrected for reliability; L = lower; U = upper.
Insufficient cases for analysis.
Furthermore, as can be seen in Table 2, the mean sample-size-weighted corrected correlations were .31 for subtle forms of discrimination and .28 for overt forms of discrimination. Despite the fact that these two effect sizes were not statistically significantly different from one another, we focus on the practical significance of the comparison of these two values. At the very least, the lack of statistical significance suggests that both forms of discrimination are similarly damaging, and thus neither form should be trivialized or overlooked. Finally, results suggested that discrimination was similarly related to adverse individual work correlates whether it was race based (.18) or sex based (.29) and regardless of publication date (1996-2000 = .27, 2001-2005 = .26, 2006-2011 = .31).
Organizationally Relevant Correlates
The meta-analytic results for samples including organizationally relevant correlates are reported in Table 3. As shown in the table, the mean sample-size-weighted corrected correlation between overall discrimination and adverse organizationally relevant correlates was .24.
Meta-Analytic Relationships Between Discrimination and Adverse Organizationally Relevant Correlates
Note: CI = confidence interval; k = number of samples; N = total number of data points; ro = uncorrected sample-size-weighted mean correlations; rc = weighted sample size mean correlations corrected for reliability; L = lower; U = upper.
Insufficient cases for analysis.
Furthermore, moderator analyses indicated the mean sample-size-weighted corrected correlation for subtle forms of discrimination was .25, whereas overt forms of discrimination resulted in a mean sample-size-weighted corrected correlation of .22. As with individual work correlates, results of significance testing did not suggest that subtle and overt discrimination were differentially related to adverse organizationally relevant correlates. However, showing that these two forms of discrimination are not differentially related to relevant outcomes is perhaps even more meaningful given that subtle discrimination is often dismissed relative to its overt counterpart. In this case, it was possible to examine only date of publication (i.e., 2001-2005 versus 2006-2011) as a potential moderator. However, the findings did not suggest the relationship between discrimination and organizationally relevant correlates generated from studies published between 2001 and 2005 (.20) differed from effect sizes generated in studies published after 2005 (.25).
Physical Health Correlates
Meta-analytic estimates of the relationship between discrimination and adverse physical health correlates are displayed in Table 4. The overall mean sample-size-weighted corrected correlation was .16, which closely corresponds with Pascoe and Smart Richman’s (2009) meta-analytic estimate of −.13, reflecting the relationship between perceived discrimination and improved physical health.
Meta-Analytic Relationships Between Discrimination and Adverse Physical Health Correlates
Note: CI = confidence interval; k = number of samples; N = total number of data points; ro = uncorrected sample-size-weighted mean correlations; rc = weighted sample size mean correlations corrected for reliability; L = lower; U = upper.
Insufficient cases for analysis.
Furthermore, moderator analyses yielded a mean sample-size-weighted corrected correlation of .17 between subtle forms of discrimination and physical health correlates and .16 between overt discrimination and physical health correlates. Again, statistical significance testing suggested these meta-analytic estimates were similar in magnitude. Additionally, our results suggested the relationship between discrimination and adverse physical health correlates was not significantly different across work (.16) and nonwork (.18) settings, and race-based discrimination yielded a similar average effect size (.19) as compared to sex-based discrimination (.15). Finally, effect sizes reflecting the relationship between discrimination and physical health correlates from studies published between 2001 and 2005 (.13) did not significantly differ from those published after 2005 (.16).
Psychological Health Correlates
In Table 5, we present the meta-analytic relationships between discrimination and adverse psychological health correlates. Here, it can be seen that the mean sample-size-weighted corrected correlation between overall discrimination and psychological distress is .30, which is noticeably larger than the effect size of −.16 found by Pascoe and Smart Richman (2009), reflecting the relationship between discrimination and psychological well-being.
Meta-Analytic Relationships Between Discrimination and Adverse Psychological Correlates
Note: CI = confidence interval; k = number of samples; N = total number of data points; ro = uncorrected sample-size-weighted mean correlations; rc = weighted sample size mean correlations corrected for reliability; L = lower; U = upper.
In addition, moderator analyses reveal average effect sizes of .31 and .28 for subtle and overt forms of discrimination, respectively. Again, significance testing suggested these two estimates were comparable in magnitude, which as we argued above is even more practically meaningful as it suggests subtle discrimination is at least as psychologically damaging as overt discrimination.
Furthermore, moderator analyses revealed similar relationships between discrimination and psychological correlates for sex-based (.33) and race-based (.30) discrimination. Regarding the setting of the study, studies conducted in work settings yielded similar effects (.30) as compared to studies conducted in nonwork settings (.29). Finally, studies published after 2001, across both intervals, yielded significantly stronger effects than those published prior to 2001. Specifically, studies published between 2001 and 2005, t(13) = 1.98, p < .05, and studies published after 2005, t(19) = 5.30, p < .01, held significantly more positive relationships with psychological distress (.29 and .31, respectively) than those published before 2001 (.20). The effect sizes of the two later interval groups did not differ.
Discussion
Taken together, our findings advance the existing literature in three primary ways. First, we have provided the first meta-analysis examining the differential correlates of subtle versus overt discrimination. While only recently have there been enough primary studies to support a meta-analytic investigation, the findings here suggest that subtle discrimination is at least as positively related to adverse correlates as its overt counterpart. Specifically, the current study indicates that the relationships between subtle discrimination and all correlates examined were comparable to the relationships between overt discrimination and the correlates examined. Second, the current meta-analysis extended the work of prior meta-analyses on outcomes associated with general forms of discrimination by examining not only psychological and physical health outcomes but also work-related outcomes and by expanding our focus to include multiple (as opposed to one single) minority groups. Third and finally, we have further clarified the distinction between the constructs of subtle and overt discrimination, providing researchers with a more defined framework for future research.
Interestingly, individual work correlates, organizationally relevant correlates, and psychological correlates held the most positive relationships with both forms of discrimination, whereas the relationship between physical health correlates and discrimination was smaller in magnitude (rc = .16), consistent with the meta-analytic estimate of the relationship between discrimination and favorable physical health correlates reported by Pascoe and Smart Richman (2009; rc = −.13). There was, however, a noticeable difference between our meta-analytic estimate of the relationship between discrimination and psychological distress (rc = .30) and that reported by Pascoe and Smart Richman (2009; rc = −.16). One potential explanation for this difference is that our estimate was based on a much smaller number of primary studies (k = 32) as compared to Pascoe and Smart Richman’s (2009; k = 105), which is likely a function of our inclusion criteria requiring that a study provide enough information to code discrimination measures as either subtle or overt in nature and could not include discrimination measures that were mixed in nature. Furthermore, there appeared to be substantial variance in Pascoe and Smart Richman’s (2009) meta-analytic estimate of the relationship between discrimination and psychological health (Qmental = 7480.44, p <.001). In fact, moderator analyses found that the absence of social support exacerbated the negative relationship between discrimination and psychological health (Pascoe & Smart Richman, 2009). Thus, it is possible that the 32 primary studies used to generate our meta-analytic estimate of the relationship between discrimination and psychological health reflected samples in which participants tended to be at the lower end of the spectrum on social support. Nevertheless, our meta-analytic estimate of the relationship between discrimination and psychological correlates was more similar to the meta-analytic estimates reported by Lee and Ahn (2011; rc = −.23) and Lee and Ahn (2012; rc = −.28).
Though significance tests did not suggest any of the subtle estimates were statistically significantly different from their respective overt estimates, it is notable that across all correlate domains, effect sizes for subtle discrimination were larger in absolute magnitude relative to those for overt discrimination. Indeed, the discrepancy between subtle and overt effect sizes was larger for organizationally relevant correlates (.31 vs. .28), individual work correlates (.25 vs. .22), and psychological correlates (.31 vs. .28) and smaller for physical health correlates (.17 vs. 16). Taken as a whole, these results are particularly provocative given that instances of subtle discrimination are often perceived as less offensive and are easily minimized or dismissed, likely due to the ambiguous nature of subtle discrimination. At the very least, our results suggest subtle discrimination is at least as important to pay attention to as overt discrimination.
Although the current investigation primarily focused on the comparison of effects of subtle and overt discrimination, additional moderators were examined. Specifically, we evaluated the possibility that the relationship between discrimination and its correlates may depend on minority status membership (i.e., race-based, sex-based), setting of the study (i.e., workplace, nonworkplace), and date of study publication (i.e., 1996-2000, 2001-2005, 2006-2011). On the whole, our results did not provide support for any moderating effects of the variables examined. However, the one exception was for psychological health correlates, for which we found that studies published in more recent years (i.e., after 2001) produced stronger negative effect sizes relative to studies published in earlier years (i.e., before 2001). One possible reason for this finding could be increased prejudiced attitudes toward Arabs and those of Middle Eastern descent following the events of 9/11 (Merskin, 2004). Finally, it is worth noting that although we did not restrict the date of publication in our search for primary studies, we found no usable studies prior to 1996. This highlights the critical need to further this line of relatively new research and continue building our base of primary studies for future meta-analytic investigations.
Furthermore, in light of potential concerns that correlates of discrimination function differently depending on whether they are examined in work or nonwork settings, it is worth mentioning that when it was possible to compare work and nonwork study settings, the relationship between discrimination and its correlates was not significantly different across study settings.
Theoretical Implications
Our findings yield several important theoretical implications with regard to experiences of discrimination from the target’s perspective. For all examined outcomes, subtle discrimination exhibited a similar relationship with correlates as compared to overt discrimination, demonstrating that subtle discrimination is at least as detrimental to targets. Thus, our findings are consistent with both stress and coping models (Clark, Anderson, Clark, & Williams, 1999; Lazarus & Folkman, 1984) as well as attributional ambiguity explanations (Crocker & Major, 1989; Crocker et al., 1991) in accounting for the potentially damaging impact of subtle discrimination. Specifically, our results build on attributional ambiguity theory in demonstrating that subtle discrimination may be particularly damaging to targets as a function of its inherently ambiguous nature. The ambiguity inherent in subtle discriminatory behaviors decreases the target’s ability to attribute the negative behavior externally and increases the likelihood the target will blame himself or herself for the negative experience, taking a toll on psychological well-being (Crocker et al., 1991). Furthermore, this process of deciding whether to attribute subtle discrimination externally or internally takes more time as compared to situations of explicitly overt discrimination, which enable targets to make quicker, easier external attributions to prejudice. Thus, our results are consistent with prior research that has shown that the heavier cognitive load required to make an attribution in the case of subtle discrimination impairs cognitive performance as compared to the lighter cognitive load associated with experiencing explicitly overt discrimination (Salvatore & Shelton, 2007; Singletary, 2009).
Finally, our meta-analytic results bolster the notion that subtle discrimination may be particularly harmful for targets because of its higher frequency and thus chronic nature. This is consistent with the notion of “accumulation,” or the idea that seemingly trivial microaggressions can accumulate over time to produce substantial negative impact on those who incur them (Cortina, 2008). Indeed, one simulation study illustrated this phenomenon by showing that seemingly inconsequential instances of male-female bias can accumulate over time, and it is the sum of these instances that coalesce into a single significant situation of bias (Martell, Lane, & Emrich, 1996).
Practical Implications
Our findings echo those of other scholarly research illustrating that subtle slights and microaggressions in the workplace likely undermine organizational efforts to emanate prodiversity climates and lead to worsened job attitudes, decreased performance, and increased turnover (Gifford, 2009; King et al., 2006; Stewart et al., 2010; Tougas et al., 2005). Given recent research in the diversity management literature demonstrating the benefits of organizational climates that are supportive of diversity (Gonzalez & DeNisi, 2009; McKay et al., 2008; McKay, Avery, Liao, & Morris, 2011), organizations must take action by increasing awareness and identification of these insidious types of behaviors perhaps by incorporating information about their potentially detrimental impact into diversity training programs.
To our knowledge, there are only two existing studies that have directly assessed the impact of diversity training participation on subsequent discriminatory behaviors (King, Dawson, Kravitz, & Gulick, 2012; Sanchez & Medkik, 2004). Whereas King and her colleagues (2012) found diversity training led to less ethnic minority discrimination, Sanchez and Medkik (2004) found that diversity training led to higher levels of differential treatment toward ethnic minorities after the training. The divergent results from these two studies suggest the impact of diversity training on discrimination likely depends on a number of moderating conditions, one of which can be gleaned from social psychological research suggesting the general lens through which diversity is portrayed (i.e., color-blind approach, multiculturalism approach) has an important impact on subsequent racially charged attitudes and behaviors.
The color-blind approach (the predominant approach to diversity training; Plaut & Markus, 2007; Thomas & Ely, 1996) underscores the importance of minimizing and ignoring subgroup differences. Generally, research has shown this approach may actually result in damaging interpersonal outcomes. For example, White confederates experimentally primed to avoid asking a Black partner about race (color-blind condition) exhibited more negative nonverbal behaviors (as coded by an independent coder watching a muted version of the interaction) toward their partners as compared to a control condition (Apfelbaum, Sommers, & Norton, 2008). Specifically, these participants’ behaviors were evaluated as more unpleasant, unfriendly, unlikeable, cruel, and cold: behaviors reminiscent of subtle discrimination. In contrast, the multiculturalism ideology, which recognizes and even celebrates group differences, has been found to predict psychological engagement and lower perceptions among ethnic minority employees that an organization’s diversity climate was racially biased (Plaut, Thomas, & Goren, 2009; Sue, 2004). Scholars argue these differential effects are likely a function of an inclusive, fair, and accepting diversity climate created by the multiculturalism approach (Plaut et al., 2009).
Taken together, the evidence described above suggests endorsement of the color-blind approach may actually lead to damaging interpersonal outcomes, particularly in the form of subtle discriminatory behaviors (Apfelbaum et al., 2008). In light of our meta-analytic findings linking subtle discrimination to a host of adverse outcomes that are both individually and organizationally relevant, an important practical implication of our study is that organizations carefully and cautiously consider their approach to diversity in various diversity initiatives (including diversity training). Specifically, emphasizing a multiculturalism approach to diversity may facilitate more positive interpersonal relationships while decreasing the likelihood of backlash in the form of both subtle and overt discrimination.
In addition to designing diversity training initiatives that are likely to reduce both subtle and overt forms of discrimination, organizations should consider creating formal policies and procedures outlining what subtle discrimination is, actions for recourse, and clear disciplinary consequences to match already existing policies and procedures focused on more blatant (and often illegal) forms of discrimination. Taken together, our findings point to the potentially damaging nature of both forms of discrimination and suggest that diversity management efforts aim to improve interpersonal relationships among subgroup members, thereby decreasing the incidence of both subtle and overt discrimination.
Limitations and Future Research Directions
The current findings should be interpreted in light of the study’s limitations. First, by and large, our primary studies were field studies and thus correlational in nature (only 4/44 primary studies (9%) were experimental lab studies). Thus, our meta-analytic effect sizes reflect the relationship between target’s perceptions of discrimination (as opposed to when discrimination is either absent or present because it is manipulated by the researcher) and the four correlate domains. To this end, as the body of evidence on discrimination continues to grow, further analysis of experimental research that directly manipulates subtle and overt discrimination is warranted. In line with this goal, our results highlight the need for subtle discrimination to be examined through a longitudinal lens to build understanding of the processes by which it emerges, in isolation as well as in conjunction with its more overt counterpart.
Second, because of the relatively small number of studies available, we were unable to determine the effects of specific subtypes (e.g., microinsults, benevolent sexism) for each form of discrimination (i.e., subtle and overt discrimination). For instance, it may be that microaggressions toward disadvantaged ethnic groups function differently than microaggressions directed toward women. Thus, whereas the subtle measures used in our primary studies generally fit our definition of subtle discrimination, they may not have reliably captured microaggressions and the more subtle aspects of subtle discrimination—at least not directly. Furthermore, though our definition of subtle discrimination identifies such behavior as “not necessarily conscious,” we may not know whether it is unconscious. In spite of this point, our definition still addresses subtlety at multiple levels: the nature of the behaviors, the perceptions and inferences made, and the context being outside of law or formal policy.
Third, single studies examining outcomes that did not conceptually fit one of our four criterion domains could not be included in this analysis. For example, one study that did not fit with our specified criterion domains examined the effect of mothers’ perceptions of discrimination on the relationship between mothers and teachers (Rowley, Helaire, & Banerjee, 2010). Indeed, as primary studies continue to build, the moderating impact of different subtypes of subtle discrimination, more varied outcomes, and more diverse targets on the relationship between discrimination and its correlates should be further examined.
Finally, given the potential impact the approach to diversity (i.e., color-blind, multiculturalism) may have on subsequent prejudiced attitudes and discriminatory behaviors, future evaluation research should strive to provide more information on the approaches taken in diversity training programs. Furthermore, given a reduction in discrimination toward socially disadvantaged group members is the most commonly cited goal of diversity training (Bendick, Egan, & Lofhjelm, 2001; Chrobot-Mason & Quinones, 2002), more careful attention to the measurement of these behaviors posttraining is warranted. Indeed, postdiversity-training behavioral measures that distinguish between subtle discriminatory and overt discriminatory behaviors represent a particularly fruitful direction for future research. In addition, given the potential impact of subtle discrimination, more attention should be given to the conditions under which subtle discrimination is most likely to occur. For instance, scholars should examine whether there are certain individual differences, organizational policies or climates, or situations that trigger its emergence.
Conclusion
In conclusion, our study represents the first attempt to meta-analytically investigate the correlates of subtle and overt forms of discrimination across work-related, psychological, and physical health domains. Our results provide support for the notion that discrimination is positively associated with adverse correlates across domains. Most importantly, the findings suggest that subtle forms of discrimination are at least as damaging as overt forms of discrimination in terms of meaningful outcomes. However, most initiatives implemented to reduce discrimination target overt rather than subtle behaviors. Thus, our work highlights the critical need to more sufficiently address subtle forms of discrimination and underscores the importance of developing initiatives that reduce even seemingly insignificant biases.
Footnotes
Acknowledgements
This article was accepted under the editorship of Deborah E. Rupp.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
