Curbing Discriminating Human Resource Practices—A Microfounded Perspective

Discrimination constitutes a sticky phenomenon in corporations despite decades of anti-discrimination initiatives. We argue that this stickiness is related to the complex relations between various factors on the micro level in organizations, which determine and stabilize each other. Based on a systematic literature review comprising empirical studies on discrimination due to age, gender, race, and ethnicity/nationality, we find eight general mechanisms which can be further clustered into an economic, a behavioral, and a socio-structural domain. While mechanisms in the behavioral domain form the roots of discrimination, the economic and the socio-structural mechanisms stabilize each other as well as the behavioral ones. Thus, the analysis shows that the various building blocks on the micro level are entangled with each other and suggests a structured way by identifying a problem hierarchy to manage this complexity.


Introduction
Discrimination, which can be understood according to Allport's (1954) widely accepted definition as activities "that deny to individuals or groups of people equality of treatment which they may wish" (Allport 1954, p. 51), has been the subject of social debate for decades. A broad social movement across countries, like for example in the U.S. the civil rights or the women's liberation movements (Kochan et al. 2003), fought and are fighting for its containment across the globe. In many regions, the various movements have led to a broad body of formal regulation in a wide variety of areas, such as the International Covenant on Civil and Political Rights (Article 26) or the European Convention for the Protection of Human Rights (Article 14) (Plummer 2003). As a result, also firms are faced with a growing number of equal opportunity legislation acts, which oblige them to avoid any activities that could result in the discrimination of certain groups within their potential and actual workforce (e.g., Jackson and Joshi 2004;Prasad et al. 2006;Shen et al. 2009).
Moreover, beside the need to fulfill these regulatory obligations firms also get under pressure to address the prevention of discrimination for strategic reasons, as it is a major inhibitor to fruitfully exploiting the heterogeneity of a company's workforce along different dimensions, like age, gender or nationality (Armstrong et al. 2010), which is a major disadvantage in times of labor shortage and changing work forces: For example, the aging populations in industrialized countries face employers with the need to adapt their business processes and their perception of older employees (e.g., Aaltio et al. 2014;Armstrong-Stassen and Templer 2005;Billett et al. 2011). Additionally, due to increasing worldwide migration firms have to integrate a more and more ethnically diverse workforce (e.g., Sharma 2016; Williams and O'Reilly III 1998). However, despite this legal and economic pressure to curb discrimination it remains an everyday phenomenon in society in general and in firms in particular worldwide.
This has fueled a considerable body of research with highly insightful findings particularly in gender-based and race-or ethnicity-based discrimination. For example, one very broad stream of research discusses gender-based discrimination and resulting gender-pay gaps as well as misrepresentation of women in higher positions (e.g., Deschacht 2017;Kolev and Robles 2010). In this context, especially the Social Role Theory developed by Eagly (1987) has to be mentioned. It relates existing inequalities between men and women in the working sphere to stereotypes and prejudices and the social context (Eagly 1987). Also, the concept of inequality regimes can be mentioned here, which are interlocking practices that can lead to inequality in organizations. The concept is useful in analyzing the success or failure of organizational change processes (Acker 2006). Other scholars focus on the effects of race and ethnicity on discrimination in the workplace. For example, already Dickens and Kane (1999) found detrimental effects of race-blind screening processes on the chances of people of color to be selected for a job. Further research deals with effects of intersectionality between the mentioned characteristics on discrimination, e.g., Smith (2005Smith ( , p. 1157 observed "promotion gaps between white men and their female and minority counterparts." Thus, overall, extant literature already provides profound insights into reasons behind discrimination in the workplace. K However, we think it is reasonable to add one further perspective to this literature: We posit that discrimination in firms is such a sticky phenomenon not only because it is hard wired in factors underlying Human Resource (HR) practices to select, promote, train, and retain staff, but also because these underlying factors-or microfoundations (e.g., Eisenhardt et al. 2010;Felin et al. 2015;Grigoriou and Rothaermel 2014;Minbaeva et al. 2012)-are additionally interconnected with each other, which further complicates to change them. However, we observe that simultaneously the very broad and rich empirical research stream that has emerged on topics related to discrimination in a corporate context due to HR practices and that provides an undoubtedly insightful knowledge base is fragmented. While large parts of literature focus on stereotypes and prejudices (e.g., Chiu et al. 2001;Kaplan et al. 2016;Li et al. 2017), other scholars discuss issues as diverse as human capital (e.g., Aubert et al. 2006;Heywood et al. 2010;Fryer et al. 2013), networks (e.g., Beaman et al. 2018;Blum 2015;Mooney and Ryan 2009) or organizational practices (e.g., Baker and French 2018;Deschacht 2017;Goldberg et al. 2013). Partly, disciplinary boundaries seem to inhibit an integrating perspective on these already existing insights and thus scholars seldom relate these issues to each other. This fragmentation prevents a comprehensive perspective on the decision maker at the organizational micro-level in companies, who ultimately consciously or unconsciously affects discrimination in the workplace, because it neglects the network of micro-level mechanisms that influence his/her decision making behavior.
We strive to address this research gap by providing a model, which provides the ground to integrate the existing empirical insights on discrimination in companies due to HR practices and thereby reconceptualizes the present issue with "new relationships and perspectives that have not been fully explored" (Torraco 2005, p. 364). Thereby it also helps to advance theory building in an innovative way and to derive paths for future research.
Building on literature on HR practices and a systematic, albeit limited literature review focusing on empirical research related to discrimination in corporations, we identify building blocks for a comprehensive model that bridges the gap between the organizational macro, meso and micro levels. However, while the model's building blocks on the macro and the meso levels are directly identifiable within extant literature, research on the microfoundation underlying these levels and directly linked to individual decisions is very fragmented and broad. Consequently, to identify major building blocks on the micro level we perform a systematic literature review, driven by the dimensions identified on the macro and meso levels. To assure that the identified research findings are related to empirically observable phenomena, we focus on empirical research (both quantitative and qualitative). On this basis, we carve out the relations between the identified microfoundations across disciplinary boundaries and thereby uncover several mechanisms underlying the stickiness of discriminating practices within corporations. Finally, we analyze the significance of these mechanisms and their connections for the individual decision making in the HR context. Based on this analysis we derive starting points for further research.
Existing research on mechanisms underlying discrimination is very broad, heterogeneous, and numerous. Thus, it is not possible to merge it in a single step to form an overarching theoretical framework within the limited scope of a single pa-per. Therefore, we need to concentrate on a specific area of existing research. Here, we focus on the literature found in the ABI/Inform database in the period from January 1990 to March 2019. This literature base is naturally, although it spans over 30 years, limited and biased. For example, it is a business-oriented database that includes primarily journals based in the U.S. and Europe. In addition, the search is limited to journal articles. Therefore, the underlying literature overview is explicitly not to be understood as a comprehensive literature review covering all existing research. Rather, it is a sample that includes a broad spectrum of existing research with different perspectives which allow for a first step to carve out the mechanisms we are looking for. Consequently, the attempt made to design an integrated model made in the present paper is just a first step to provide a nucleus which can be enriched by future research along several directions, be it diversity dimensions or the scope of the incorporated research streams including impactful research which was certainly missed due to the necessary restrictions regarding the search procedure. With impactful research we mean in this context, research which provides further insights into the mechanisms on the micro levels which have intended to be identified.
The paper proceeds as follows: The second section contains the development and explanation of the integrated model. The procedure of the systematic literature review is presented in the third section. Section four is dedicated to the presentation of the results of this literature review. In the fifth section we discuss these results and derive implications for theory building and future research. The paper closes with a final conclusion as well as the discussion of the limitations.

Macro Level: Composition of the Workforce
Our starting point for discussing the composition of the workforce is the concept of "workforce diversity". Literature provides a variety of definitions and conceptions (e.g., Harrison and Sin 2006;Williams and O'Reilly III 1998). We focus on the definition by Ely and Thomas (2001, p. 230), according to whom "diversity is a characteristic of groups of two or more people and typically refers to demographic differences of one sort or another among group members". Thus, workforce diversity can be defined as the degree of employees' heterogeneity within a given company and it is a phenomenon of a firm's macro level as it emerges from the joint observation of employees' individual characteristics, i.e., one employee cannot be "diverse". This heterogeneity can be examined along many dimensions, which are of varying importance within different contexts. For example, while in Europe especially gender-based discrimination is discussed, in the U.S. and in South Africa race-based heterogeneity is an important issue and in China the discrimination against members of the rural population as opposed to the urban inhabitants plays an important role (e.g., Shen et al. 2009). In literature different attempts have been made to categorize the relevant components of diversity along a manageable amount of dimensions to reduce the complexity of this research field. Several scholars have suggested to divide the components of diversity into two broader categories: Visible (readily detectable attributes) and invisible characteristics (underlying attributes) (e.g., Jackson et al. 1995;Mannix and Neale 2005). Visible types are those that can be noticed by meeting a person the first time, for example sex, race or age, whereas invisible attributes just become obvious after getting to know a person, like personality, values, education, job tenure or technical abilities (e.g., Jackson et al. 2003;Kirton and Greene 2015;Foldy 2002;Milliken and Martins 1996). Besides, there are some dimensions, which fall between these two types like disabilities or ethnicity (e.g., Prasad et al. 2006).
Due to the limitation of a set of variables that are operationalized by just one characteristic, other scholars criticized this two-dimensional typology and suggested various multifaceted approaches. These approaches use various groups of categories and their interactions (e.g., Mannix and Neale 2005). McGrath et al. (1995) were the first researchers who differentiated five categories. Mannix and Neale (2005) picked up this categorization and created a list with the following dimensions: (1) socialcategory differences (race, gender, religion, age etc.), (2) differences in knowledge or skills (education, functional knowledge, training etc.), (3) differences in values or beliefs (cultural background, ideological beliefs), (4) personality differences (cognitive style, affective disposition, motivational factors), (5) organizational-or communitystatus differences (tenure, title), and (6) differences in social and network ties (workrelated ties, community ties etc.).
Finally, one further perspective on diversity is introduced by Gardenswartz and Rowe (2003), who developed a four layers model. In their framework, personality is at the center and in the first layer, referred to as internal or primary dimensions according to Gardenswartz and Rowe (2003) and Loden and Rosener (1991), are aspects such as race, gender, or age. The second layer comprises external dimensions, which refer to how a person is shaped by the society and his/her experiences, e.g. income, educational background or parental status. The third layer contains the organizational dimensions that includes aspects such as seniority, work location or management status.
As the previous discussion shows, a model that explains the development of firms' workforce diversity to track discrimination should be open with respect to the specific diversity dimensions, because organizations in different contexts have to focus on different dimensions. However, as the present model also shall be applied to identify extant literature with relevance to discriminating mechanisms underlying an unbalanced workforce, several specific dimensions have to be selected to allow for a theory-driven literature review. For the purposes of the present paper, a multifaceted model seems to be the best fitting alternative, as it provides a high degree of flexibility with respect to the applied dimensions, while the four layer model seems to be more suitable to foster an individual centered analysis and the categorization in visible versus invisible characteristics seems to be too undifferentiated. Consequently, we will follow existing literature and focus on the multifaceted approach of the big eight which received the most attention in theory and business practice (Plummer 2003). This model comprises the following dimensions: race, gender, ethnicity/nationality, organizational role/function, age, sexual orientation, mental/physical ability, and religion. It has also been applied (in modified versions) in previous literature reviews, for example, Shen et al. (2009) concentrate in their review on race/ethnicity, gender, age, disability, sexual orientation, and cultural/national origin, while Shore et al. (2009) focus in their literature review on six dimensions of diversity (race, gender, age, disability, sexual orientation, and national origin). In the present study we concentrate on age, gender, race and ethnicity/nationality, as during a pre-study phase these dimensions were identified as those dimensions with the broadest empirical research basis.
While in everyday language it seems rather clear, what these categories mean, in the scientific literature there is a discussion regarding their precise definition. The present paper is not intended to add further insights into this discussion, instead we base our model on existing definitions of these categories as follows: We define race as a population with certain common social categories or observable physical characteristics (e.g., Gardenswartz and Rowe 2003;Kaplan 2011). As related to gender most works do not clearly distinguish between gender and sex, we treat the two terms as synonymous in this paper. It should be noted, however, that gender is often understood as socially constructed, i.e., feminine and masculine norms, which can constrain choices and behavior (West and Zimmerman 1987). Since at this point the literature often conflate the concepts of nationality and ethnicity, both concepts were examined. We define ethnicity as a population with homogenous culture, traditions and values and nationality as a population which refers to political state borders (e.g., Gardenswartz and Rowe 2003;Cox et al. 1991). Finally, we define age as the actual age of life of a person (e.g., Armstrong-Stassen and Templer 2005;Cleveland and Shore 1992;Shore et al. 2003).
In this context, also intersectionality, i.e., the impact of discrimination on persons who belong to several disadvantaged groups at the same time, has to be considered. It can lead to experiences of discrimination that differ significantly from people who belong to only one group (e.g., Crenshaw 1989). Beside an aggravation of discrimination experiences, intersectionality also can lead to a compensation between disadvantages and advantages resulting from the affiliation with different social groups (Crenshaw 1989). Accordingly, intersectionality plays a persistent and important role, when analyzing different diversity dimensions, and thus also in our model, even if it is not explicitly stated in the model.

HR Practices as the Connecting Piece Between Macro and Micro Levels
While workforce diversity and its manifestation within different dimensions, like gender or age groups, constitutes the organizational phenomena on the macro level, and particular mechanisms underlying discriminating practices are situated on the organizational micro level, a connection is needed to translate these mechanisms into the observed degree of workforce diversity. HR practices constitute these connecting pieces on the meso level. On the one hand, they are shaped by individual decision processes, which in turn are affected by discriminating mechanisms on the micro level. On the other hand, they determine through different processes, like hiring or retention, the particular manifestation of workforce diversity (or its absence).
In order to identify the relevant HR practices, which determine the composition of the workforce as a whole and within different hierarchical levels, both access and K treatment discrimination are of importance. While the discussion of access discrimination focuses on (un)equal possibilities of different minorities to get hired, the research on treatment discrimination examines differences regarding the opportunities members of different minority groups have when they become an employee of the firm (Dwertmann et al. 2016). Access discrimination determines the workforce diversity as such, i.e., whether a firm's workforce exhibits a high share of female employees. With respect to this discrimination type recruitment and retention activities are of importance. Treatment discrimination determines the composition of the workforce on the different hierarchical levels, e.g., while a firm exhibits an equal share of female and male employees, the share of female employees decreases from level to level. This kind of discrimination manifests in several HR practices. First, professional development provides opportunities to qualify for more sophisticated and managerial activities. Thus, the access to these trainings significantly influences the possibility to ascend within the organizational hierarchy. Second, the outcomes of performance appraisal often affect the possibility to get promoted and, thus, they are one further building block to shape the composition of workforces on different levels. Third, promotion activities directly determine access to higher positions through the process of selecting certain employees for higher hierarchical levels, and they are closely linked to the previous two issues.
While the components on the macro and the meso levels are directly derived from literature, the various mechanisms on the micro level are the subject of the systematic literature review. Consequently, they will not be further discussed at this point of the study. However, one further building block of the model has to be mentioned: As the existing workforce comprises those decision makers, who in turn hire and promote new employees, one has to assume that the configuration of the current workforce has an impact on the currently operating mechanisms on the micro level (e.g., Barney and Felin 2013;Felin et al. 2012;Minbaeva et al. 2012). Thus, we assume a self-stabilizing system.
The preceding discussion within the two previous sections provides the starting point for the systematic literature review by offering a framework for structuring the search process as discussed in the following section.

Literature Search and Inclusion Criteria
A theory-driven and systematic literature review was performed to complete the model described above (Cooper 1988;Tranfield et al. 2003;Webster and Watson 2002). Such a review is a form of concept-centric research to identify, describe, summarize, analyze, and critically evaluate the content of relevant studies regarding a clearly formulated research question with the aim of creating new determining factors, reconceptualizations and perspectives for the topic (Moher et al. 2009;Torraco 2005;Webster and Watson 2002). Basically, it comprises an inductive approach to collect and structure the relevant literature to bring more clarity and coherence into a field (Cooper 1988). The scope of such a review is to derive new ideas and research directions and thereby to prepare the ground for future research. In the context of the present review, a new conceptual model is built to highlight new relationships and perspectives at the micro level that have not yet been fully explored (Torraco 2005). However, we do not intend to identify the complete literature related to discrimination based on age, gender, race or ethnicity/nationality. Due to the scope of this literature, this would not be possible in a reasonable manner. In order to cope with the "tension between the statistical benefits of exhaustive inclusion and a large number of primary studies on the one hand, and high-quality reviews of fewer studies using more selective methodological criteria of inclusion and exclusion" (Davies 2000, p. 370), we aim at receiving an overview of a representative part of the enormous field, but not of the complete literature. To identify relevant empirical studies, we used the database ABI/Inform Collection, because it is one of the most comprehensive business databases. It contains about 4,000 journals with a high number of peer-reviewed journals in the fields of management, financial information, diversity, human resources, taxation, and economic conditions in organizations at all levels. Beside business-oriented journals, it also covers journals from other disciplines, particularly psychology and sociology, which have a close connection to the business context. Our research focus lies in HR Management, which is characterized by a high degree of interdisciplinary research fueled especially by business, psychological, and sociological theories and insights. As this broad perspective is reflected in the ABI/Inform Collection, it provides a reasonable ground for our literature search. However, one has to mention that the database covers journals predominantly from the Anglo-American language area supplemented mainly by journals located in European countries. This regional focus in terms of journals (not in terms of research topics) has to be judged as a limitation. Yet, in the business context exactly these journals are the main targets of scholars around the globe for reasons of prestige. Therefore, we are confident, that the database covers a wide range of highly impactful research, which we want to reflect in our overview. Moreover, as this database has a very broad coverage resulting in an already high number of peer-reviewed hits, we decided to use just one database to keep the literature search manageable. In sum, we are aware that we can only look at a sub-area in the large field of research here. Accordingly, our model should be seen as a starting point which needs to be extended in the light of results from other language areas and other databases.
We gathered articles of the database in two phases from October 2017 to Mai 2018. In addition, we carried out a third search-phase in February and March 2019. An overview of the different phases and an approximate time schedule of the literature search is shown in Fig. 1.
We included peer-reviewed and English written articles from scientific, academic journals in the title, the keywords, and the abstract. Furthermore, we just gathered articles after 31.12.1989. We searched each of the four dimensions (age, gender, race, ethnicity/nationality) in combination with each HR practice (hiring, retention, promotion, training, appraisal). With respect to the HR practices we used the keywords hire/hiring/recruit/recruitment/selection, retention/turnover/withdrawal, promote/promotion, professional/vocational development/training and development, and performance appraisal/evaluation. Moreover, we decided not to include the search term diversity in our search string, because it extremely limited our results. Since narrowing down the search by keyword to hits containing content only related to the micro-level proved to be difficult and error-prone, we refrained from using corresponding keywords. The selection of articles focusing on this level was performed manually when viewing the hits during the further process.
After the first entry into the database without any content restrictions, 10,279 hits were displayed. Therefore, we decided to further narrow our search and to do a preselection: Only content related to companies was considered and articles which dealt with topics in the nonprofit sector, the health sector/medicine (e.g., doctors, nurses), biology, the education sector (e.g., schools, universities) as well as public relations and services (e.g., government, police, military) were not included.
This preselection was made by the first research assistant, who logged the identified articles into an excel-list. This list consisted of four tabs, which are assigned to the four dimensions. Within these tabs the following information was recorded: Continuous numbering of articles of the dimension, used search terms of HR practices, total number of hits of each HR practice, title, author, year, journal, topic/ keywords and time of search. Difficult cases in terms of content-inappropriate articles or articles of another dimension were color-coded in the list. A second research assistant rechecked the created list with the database, i.e., the preselection, and if necessary, added newly published articles. After the preselection of the research assistants our list included in total 2,138 articles. 614 articles were related to the age dimension, 150 to the ethnicity/nationality dimension, 274 to the race dimension and 1,100 articles to the gender dimension.
Thereafter, the authors start with the main selection process by examining the 2,138 articles of the preselection list regarding their relevance to our topic on the basis of title, keywords, abstract and in difficult cases full text. We only included empirical articles that use quantitative and qualitative research methods. Case studies without profound methods like non-scientific interviews, reviews (including metaanalysis), simulations, and mathematical models were eliminated. Moreover, articles only were included if they contained explanations for micro-level mechanisms, as affecting individual decisions in the HR context focused on, i.e., hiring, retaining, promoting, appraising, and training. This means, that these articles had to provide insights into why discrimination was observed in a certain context related to one or more of the four diversity dimensions. Thus, not only non-empirical contributions were excluded in this step, but also articles that were purely descriptive regarding certain facts, such as the existence of the gender-pay gap in different industries. Some articles contained both perspectives and were then included with an eye toward the relevant parts of the text. It should be noted that this approach is characterized by a certain degree of subjectivity, which is a clear limitation. However, because we processed and included a very wide range of articles, we think that our overall sample of articles reflects the relevant literature on the mechanisms adequately. Additionally, only studies focusing on one or more of the four dimensions age, gender, race, and ethnicity/nationality, as well as on objective discrimination (in contrast to perception of discrimination) were included. Further, duplicates of articles within a dimension were identified and marked. In addition, we had a couple of articles that covered several dimensions (e.g., gender and age) but did not appear in both dimensions during the database search. They were copied into the other dimension with a note. Despite our intensive research, three full versions of articles could not be found. These were left out in the examination.
To ensure that a similar classification between the two authors in the main selection process was made and to set decision making norms, at the onset both authors reviewed the first 100 articles of the age dimension together. Afterwards, the articles were divided between the two authors and were revised separately and independently. However, during the process there was a continuous exchange between the authors to refine inclusion rules and to discuss special cases.
After the main selection process, we retained a list of 397 articles including empirical studies that only contained descriptive insights, like the degree of observed discrimination in a particular area. As these articles did not provide any evidence regarding causal impacts of mechanism on the micro level, the authors finally excluded them. This left us with 252 relevant articles including duplications across the dimensions to be part of the following analysis. In detail, we finally identified 44 articles on age, 21 on ethnicity/nationality, 145 on gender and 42 on race.

Categorization and Coding
In order to code and categorize the identified articles, they were downloaded, analyzed, assigned to an inductively generated categorizing system with respect to the discrimination mechanisms at the micro level, and the used theory was noted. The generation of the categorizing system was organized in an iterative manner, i.e., in K a first step, from each article short summaries or original text parts were introduced into the master file. These text passages were further condensed to a manageable amount of categories by both authors. While we were categorizing the articles, we continually reviewed our previously established content restrictions to check for coherence. Through this process we identified the three following categories: economic determinants, behavioral determinants, and socio-structural determinants, where all three categories were further structured into subcategories. While all articles focus on discrimination regarding the four selected dimensions in corporations and thus have an inherent business focus, they are characterized by different perspectives, reflecting different disciplines, particularly economics and business administration, psychology and sociology. This results in a certain heterogeneity of articles that needs to be structured. In carving out the three main categories, we have tried to take these disciplinary boundaries into account. It turns out that an economic perspective can be delimited very well. On the other hand, there are overlaps in the contributions regarding (social) psychological and sociological perspectives, which is why we have sorted them into one category. Finally, there are articles dealing with structural aspects that do not fit into either of the two categories, resulting in a third main category. Thus, in detail, we have the following main categories: The category economic determinants contains articles that investigate mechanisms underlying discrimination which result from the rational choice perspective on human decision making: In a market economy with a special emphasis on investor rights firms mainly have to focus on activities that result in profit maximization. To reach this goal, firms implement management control systems that foster these activities by aligning employees' activities with profit maximization through appropriate incentive and control systems. Thus, under the assumption of rational choice and the implementation of effective management control systems, employees should act in a way that results in firm's maximal profits. From this perspective, the degree of workforce diversity depends on its compatibility with the economic goals of a company. Thus, if the inclusion of a specific minority group results in more (less) favorable financial results, the hiring, promotion, and retention of members of this group should be fostered (hindered). In this sense, HR practices should be designed to select, retain, develop, and promote the best fitting persons, and thus from a rational choice and profit maximization point of view organizations should be blind against diversity as such. Discrimination against certain groups is the result of an economic rational. If it is economically rational to foster the recruitment and promotion of members of a certain minority group, any discriminating activities are ceased. This perspective also comprises the consideration of legal aspects, like antidiscrimination laws, as firms have to weigh whether it is economically favorable to obey these laws or whether it is more favorable to disregard them, because this disregard results in a higher overall profit than compliance. Articles assigned to this category investigate aspects related to either the employees as such and are related to human capital theory and productivity or external pressure which firms follow for the above mentioned economic reasons.
The second category comprises processes rooted in behavioral foundations of human interactions, which are explained in literature by (social) psychological and sociological theories and that result in discrimination or inclusion of different groups.
In contrast to the economic perspective, papers assigned to this category apply mechanisms that explain discrimination or the lack thereof by inherent factors, i.e., discrimination occurs intrinsically and is not a means to achieve a particular target related to economic reasoning. This category can be further differentiated into the following subcategories: (1) stereotypes and prejudices and (2) relational demographics, similarity-attraction paradigm, interpersonal distance, and homosocial reproduction.
The third category focuses on socio-structural aspects, i.e., formal and informal structures that foster discrimination, like particular network structures and organizational routines and processes. This category differs from the previous two in the sense that here discrimination is explained by structural determinants. These determinants once were formed as outcome of economic rational or behavioral aspects, but now they are detached from this reasoning and form independent entities. This category comprises the following subcategories: (1) networks, (2) selection methods, (3) social construction, and (4) organizational practice (see supplementary material No. 1).

Descriptive Statistics of the Literature Sample
As indicated by Fig. 2, research streams related to the four investigated diversity dimensions show a varying distribution over time within the identified sample. Research on race has been carried out continuously over years in the literature sample, while research on ethnicity/nationality only became more popular in the business context since 2002. Age also was considered continuously since 1991, but it has been more intensively investigated in 2013-2015. The gender dimension has been     Fig. 3. With regard to HR practices since 1991 research has clearly focused on hiring followed by promotion as seen in Fig. 4. Finally, Fig. 5 shows a dominance of articles related to the behavioral perspective comprising 66% of the publications, while Fig. 6 illustrates that the subcategory stereotypes and prejudice with 128 articles is most strongly represented. The second behavioral subcategory, relational demographics etc. (46 articles) and the economic subcategory human capital and productivity (35 articles) are also intensively researched. It has to be noted, that these observations may result from the limited scope of the applied database. Thus, we do not see these statistics as representative for the complete research on the search topics, but they delimitate the boundaries of the theoretical nucleus, which we intend to derive in this paper and which accordingly has to be enriched in future research.

Economic Determinants
Human Capital Theory and Productivity Human capital, i.e., the disposition of psychological and physical capabilities, experience, and education, should constitute an objective and fair criterion, when deciding regarding hiring, promotion, etc. Yet, this is only the case, if every person has the same chances to gain human capital, but over decades scholars find ample evidence for an asymmetric distribution of relevant human capital variables across age, gender, race or ethnicity/nationality disadvantaging whole groups (e.g., Aubert et al. 2006;Fryer et al. 2013;Hargis et al. 2006;Harris et al. 2002;Heywood et al. 2010;Hirsch et al. 2000;Kirchmeyer 2002;Lahey 2008;Månsson et al. 2013;Tokunaga and Graham 1996;Wilkins and Gulati 1996). Particularly, older employees are disadvantaged as they are perceived of being less productive (Adams and Heywood 2007;Heyma et al. 2014;Hu 2003). Women have a disadvantage due to a lack of experience, education, as well as actual and expected future tenure (Blum 2015;Igbaria and Chidambaram 1997;Kaestner 1994;Kolev and Robles 2010;Lee and Wie 2017;Olsen and Sexton 1996;Pillai et al. 2011;Smith et al. 2013;Tokunaga and Graham 1996), due to a resulting performance difference (Smith, 2005), due to physical working conditions (Mohsen and Zouari 2014) or due to a possible motherhood (Dambrin and Lambert 2008). Racial minorities suffer from changing job profiles: For example, Moss and Tilly (1995) observe that changing skill profiles of jobs disproportionally affect black staff as they on average have lower skills. Additionally, particular groups have difficulties to present their skills in an adequate way. For example, Hiemstra et al. (2013) find evidence, that ethnic minorities in western countries have more difficulties to properly present their human capital in their resumes, which reduces their probability to get hired.
Access to human capital changes over time, e.g., at least in some parts of the world women get more access to academic education (Wood 2008), i.e., in Korea today's access to education for women and their legal protection have increased significantly (Patterson et al. 2013). Thus, the findings in the cited literature cannot be generalized over all time periods and for all regions in the world. But they show that minorities (like ethnical minorities) or marginalized groups (like women) systematically suffer from a lack of access to human capital. This circumstance means that meritocratic systems help cement discrimination, as their ostensibly fair selection mechanisms are blind to the fundamental disadvantage of certain groups and perpetuate their exclusion from certain jobs, as for example indicated by the findings by Fitzsimmons et al. (2014). Based on the life trajectories of male and female CEOs they point to less possibilities of women to accumulated career relevant experience throughout their life span. This in turn leads to the fact, that it is not those with the highest potential who get the job, but those who happen to have access to building certain human capital. Thus, from a strict rational choice perspective, the reduction of barriers to building human capital should be an aspiration in the interests of companies, which they should pursue consistently and group-specific. This is the only way to ensure long-term access to the most talented employees. For example, in this context several scholars find that education and training can have a stronger impact on promotion for women than for men (Cassidy et al. 2016;Garcia-Crespo 2001;Melero 2010). Thus, these measures are suitable for closing certain gaps in women's human capital that have stood in the way of promotion compared to men. However, this view is countered by a cost-benefit analysis: the removal of such barriers and the targeted development of human capital for disadvantaged groups initially incurs costs. If the employees supported in this way leave the company and move to a competitor, there is no possibility of recouping these costs. Accordingly, companies find themselves in a kind of prisoner's dilemma in this context, which they solve by only supporting measures for disadvantaged persons in building up human capital if there is a clear benefit to be gained from doing so and, thus, the discrimination-stabilizing mechanisms of meritocratic systems remain intact if this benefit does not exist.

External Pressure
Factors exerting pressure to make it economically reasonable to discriminate or curb discrimination are manifold. A general pressure from tense labor markets is not fully observed: According to Karpinska et al. (2013) a shortage at the labor market increases the probability of hiring early-retirees, while Goldberg et al. (2013) do not find hints of a relation between tight labor markets and hiring older job-seekers. In contrast, a number of stakeholders have a strong influence: Scholars find evidence that firms aim at equality to avoid litigation (Koivunen et al. 2015), to satisfy investor claims (Mun and Jung 2018), to implement an internationalization strategy (Nielsen and Nielsen 2010), or to match staff to customers along relevant dimensions, like age or ethnicity (Bendick et al. 2010;Lee 1998;Pärnänen 2012). Also, findings related to the influence of governmental policies and litigation are ambiguous: Mun and Jung (2018) observe hardly evidence of policy-induced discrimination against women because of generous family policies. Additionally, several authors provide evidence of positive effects of governmental intervention on hiring processes (Gregorič et al. 2017;Ilmakunnas and Ilmakunnas 2015;Pärnänen 2012), while others find only slight positive impacts, no effect at all or even negative effects (Agesa 2001 Garen et al. (1996) provide evidence that generously defined benefit pension plans attenuate the hiring of older workers in entrylevel positions and Scott et al. (1995) observe a negative effect of a combination between high insurance costs and age discrimination laws on the hiring chances of older employees. Literature also provides evidence of a change from overt to subtle discrimination in reaction to governmental policies (Whiting 2012). For example, the study by Colarelli et al. (2010) reveals a positive impact of affirmative action policies in increasing the hiring of minorities, but a decreasing "percentage of minority hires attributed to higher qualifications and increased perceptions that hires were due to AA more than was actually the case" (Colarelli et al. 2010, p. 166). Moreover, effects vary across groups: Hirsh and Cha (2017) find different effects of various court-mandated policy changes on gender-and race-based diversity, particularly: "Policies designed to reduce bias expand opportunities for white women but not for other demographic groups. By contrast, opportunities in management for all groups expand when policies are designed to increase organizational accountability by establishing specific recruitment, hiring, or promotion plans and monitoring arrangements. Policies designed to increase rights' awareness are associated with declines in managerial diversity" (Hirsh and Cha 2017, p. 42). In sum, the pressure on companies to satisfy stakeholders with very different goals leads to conflicting demands. For example, some stakeholders demand to stop discriminatory behavior (e.g., investors), while others have needs that encourage it implicitly (e.g., customers). At the same time, there are no clear results on which measures are effective in reducing discrimination and what unintended side effects there may be in each case, as these measures are embedded in overall complex social structures.
Finally, literature on external pressure indicates that corporate initiatives to curb discrimination need an incident providing the opportunity to do so: Findings based on a case study within a manufacturing firm by Dencker (2008) show that the restructuring process of this firm followed the opportunity structure for discrimination framework, i.e., the firm followed pressure to achieve gender equity when it had the opportunity to do so.

Socio-Sructural Determinants
Selection Methods Selection methods determine which person from a pool of interested candidates finally is hired and thereby can lay the ground for the observed configuration of a workforce along various diversity dimensions (e.g., Gardner and Deadrick 2008;Kokot 2014;McKinney and Collins 1991;Schmitt 2003). However, the impact of selection methods on discrimination depends on the method and the diversity dimension: In a British sample Ones and Anderson (2002) did not find large differences regarding three work-related personality inventories across gender and ethnic groups, while Guthrie et al. (2003) provide evidence that personality inventories based on the Big Five favor women in selection processes related to expat assignments. Moreover, Harris (2001Harris ( , 2002 identifies closed, informal selection methods as an important reason for lower rates of women in international assignments. Bernardin et al. (2012) observe discriminating effects of top-down appraisal, multisource appraisal, and an assessment center against particular race groups (nonwhites) but not on gender. Moreover, literature reveals several unintended effects of selection methods, i.e., Jacksch and Klehe (2016) find evidence that transparent selection methods not always lead to the intended effects: the communication of selection dimensions, which are related to negative stereotypes for particular members of the candidate field can mitigate their performance. Dickens and Kane (1999, p. 331) show that "race-blind screening processes normally will produce large gaps between the test scores of blacks and whites". In sum, although selection methods are designed to choose the most appropriate candidate, they also can result in overt or subtle discrimination.

Social Construction
Social construction processes lead to particularly subtle discrimination processes, as they stabilize the status quo (Holgersson et al. 2016;Lupu 2012;Riach 2007;Sharp et al. 2012). For example, Tienari et al. (2013) observe that practices of executive search stabilize the status quo by excluding women from these positions and Joyce and Walker (2015) find evidence that gender essentialist assumptions lead to re-segregation in the UK solvency profession.
Networking Networks are another socio-structural aspect that is extensively discussed in the analyzed literature, especially with respect to the discrimination against women in various industries (Beaman et al. 2018;Blum 2015;Booth et al. 2003;Mooney and Ryan 2009;Taber and Hendricks 2003). Additionally, networks can foster gender segregation, as e.g., female gendered occupations are filled with people from a network and women and men have different networks, i.e., they increase segregation (Fernandez and Sosa 2005). Finally, they also exert gender specific effects, i.e., it exists evidence that the exposure to powerful networks positively affects male managers' promotions, but not female managers' promotions (Eddleston et al. 2004).

Organizational Practice and Norms
Organizational practices and norms, like supporting diversity, diversity friendly organizational climate or anti-age discrimination policies, can attenuate discriminating tendencies (Chiu et al. 2001;Goldberg et al. 2013). Yet, corporate norms, partly affected by socio-cultural norms, also can foster discrimination (Björklund et al. 2012;Patterson et al. 2013). Moreover, scholars find impacts of organizational practices particularly in the context of genderbased discrimination. Several authors identify long-hours culture and old-boys networks as one major barrier for women to proceed in their career (Deschacht 2017;Mooney and Ryan 2009;Ogden et al. 2006). For example, Baker and French (2018) identify presentism, work practices as well as hiring and promotion practices as barriers for women in the Australian construction and property industry. But also less social support affects female careers negatively as well as practices of placing them in positions less suitable as starting point for a promotion: Germain et al. (2012) identify a lack of social support as major barrier during aviation training for female pilots. In a study by Cassirer and Reskin (2000) men attached greater importance to promotion than women because they were more likely to be located in organizational positions that encourage employees to hope for a promotion. Certain groups K are thus disadvantaged by particular organizational practices tailored to dominant groups.

Behavioral Determinants
Stereotypes and Prejudices A review of the identified literature suggests that stereotypes and prejudices are a fundamental cause of discrimination in the workplace context. The former, stereotypes, are "beliefs about the characteristics, attributes, and behaviors of members of certain groups" (Hilton and Hippel 1996, p. 240) that serve to reduce environmental complexity (DeArmond et al. 2006). Prejudices arise when stereotypes are charged with positive or negative emotions towards a group or person without using the available facts, which can result in an antipathy towards group members (Hilton and Hippel 1996). The identified literature stream in this area is very broad. In order to facilitate a structured discussion, an additional subcategorization is applied (see supplementary material No. 2).
The analysis of the identified articles reveals that these aspects determine discrimination related to age, gender, race, and ethnicity/nationality across different situations, like hiring and staffing (e.g., Kaplan et al. 2016), training, promotion, and retention (Chiu et al. 2001) or performance appraisal (Li et al. 2017). Yet, there are also sporadic observations which contradict large gender-or age-based differences: With respect to the impact of aggressive behavior (as one particular example of negative performance episodes) on performance appraisal, there seem to be no gender-based difference (Way 2017). Wren (2006) observed overall less genderbased differences regarding evaluation and rewarding by supervisors than expected. Liden et al. (1996) could not find a negative bias in performance ratings against older employees.
Moreover, findings in the analyzed literature indicate that stereotypical beliefs and their impact are affected by various characteristics of decision makers, like experiences working with elderly workers (Kim and Mo 2014), interethnic contacts and education (Blommaert et al. 2014), personal age norms (Karpinska et al. 2013), ambiguity intolerance (Johnson et al. 1998), attitudes towards women styles (Bednar and Gicheva 2018), and CEOs' status of being parent of daughters (Dasgupta et al. 2018). These factors are related to the outcome of experiences (e.g., interethnic contact, personal norms and attitudes) and the disposition to allow for such experiences (i.e., ambiguity tolerance). Thus, as stressed by the contact hypothesis (Allport 1954;Pettigrew and Tropp 2006) fostering the change of stereotypes and prejudices through allowing for positive experiences constitutes the first way to mitigate their negative impact. This strategy also is the only possibility to mitigate discrimination resulting from stereotypes related to appearance (e.g., Jawahar and Mattsson 2005;Johnson et al. 2010;Kaufmann et al. 2016;Leckcivilize and Straub 2018), to fitting processes, e.g., the positive effects of the fit of applicants or ratees to job-related, sex/gender or age stereotypes (e.g., Leung and Koppman 2018;Perry et al. 1996;Post et al. 2009;Rice and Barth 2017), to a candidate's fit with predefined masculinity norms (Flynn et al. 2015) as well as to the preference for candidates with previous gender-typical jobs (Hareli et al. 2008).
One important step to change stereotypes or at least to reduce their impact in the first instance is the mitigation of their activation. As indicated by the identified literature framing and wording activates and thereby strengthens the effect of stereotypes (e.g., Horvath and Sczesny 2016).
Yet, it should also be noted that even a positive change in stereotypes and prejudices does not necessarily lead to a reduction in discrimination, as scholars also provide evidence for ambiguous stereotypes and behavior: Loretto and White (2006) reveal both positive and negative biases against older workers affecting organizational practice. Rego et al. (2018) point to the rather complex interaction between manager attitudes in affecting hiring decisions and find an incongruence between the positive attitudes that managers express towards older candidates and their decision to hire a younger candidate. Krings et al. (2011) observe the discrimination of older candidates, even if the job required a trait in which they were judged high in, namely warmth-related qualities. This observation is further supported by a range of findings that point to complex relations between stereotypes and discrimination (e.g., Derous et al. 2015;Frazer and Wiersma 2001;McCausland et al. 2015;Powell and Butterfield 2002). Particularly, women's competence signals during hiring can result in less favorable future performance evaluations (Inesi and Cable 2015), as these signals threaten traditional gender hierarchy. Similarly, Snipes et al. (1998) find that while gender bias hardly played any role in the hiring processes analyzed by the authors, women's future performance was judged less favorable. Goldberg et al. (2004, p. 807) reveal a complex interaction between age and gender: "In addition, we observed a three-way interaction between gender, age, and age-type of industry indicating that younger men received more promotions in old-typed industries, while younger women received more promotions in young-typed ones." Ruggs et al. (2014) discover an interaction effect between age and gender which is explained by violating normative gender behavior in the sense of having a stable employment position: During the hiring process, older male applicants are more negatively evaluated than older women and younger men. Castilla and Benard (2010) show the pitfalls of promoting a meritocracy in organizations: they observe that "when an organization is explicitly presented as meritocratic, individuals in managerial positions favor a male employee over an equally qualified female employee by awarding him a larger monetary reward" (Castilla and Benard 2010, p. 543). Lee et al. (2015) find an effect of the probability that selecting persons will have to compete with a candidate on their propensity to discriminate due to age and race. According to the findings by Koeber and Wright (2006) employers tend to use signals derived from human capital characteristics, which show to which extend female applicants depart from negative stereotypes of female employees. Ferris and King (1992) find evidence that older nurses received lower performance ratings than their younger colleagues, despite no objective performance differences, due to less political behavior resulting in less liking. In sum, as the complex interaction between stereotypes related to several diversity dimensions illustrates, in order to sustainably reduce discrimination based on stereotypes and prejudices in the business context, stereotypes and prejudices must be tackled in parallel along all diversity dimensions concerned. At the same time, it should be noted that even in the case of positive stereotypes there is discrimination against minorities.

Relational Demographics, Similarity-Attraction Paradigm, Interpersonal Distance, and Homosocial Reproduction
The identified literature provides broad evidence that the similarity between the person applying an HR practice and the person affected by this practice exerts an important effect on the benevolence of the former. This effect can be observed with respect to age (Chiu et al. 2001;Davidson et al. 2006;Goldberg et al. 2013;Heyma et al. 2014), race and ethnicity (Breland et al. 2017;Elvira and Town 2001;Giuliano et al. 2011;Lin et al. 1992;Mount et al. 1997;Prewett-Livingston et al. 1996;Stoll et al. 2004) as well as gender/sex (Connerley et al. 2008;Furnham and Stringfield 2001;Gorman 2005;Holgersson 2013). In turn, scholars also show that nomination committees containing women and national minorities can increase board diversity with respect to gender and nationality (Hutchinson et al. 2015;Kaczmarek et al. 2012) and that decision makers' diversity increases the likelihood of women to get a top management position (Cook and Glass 2014;Elsaid and Ursel 2011). Moreover, according to McCarthy et al. (2010) highly structured interviews can mitigate demographic similarity affects regarding gender and race. Yet, it has to be noted that these findings are not unambiguous, as various scholars did not at all or only hardly find a similar-race, -gender or -age effect on performance evaluations (Liden et al. 1996;Sackett and DuBois 1991;Scherer et al. 1991;Van der Heijden et al. 2010;Waldman and Avolio 1991) and on the perception of job performance (Bertolino et al. 2012), or a gender effect in recruiting processes (Bednar and Gicheva 2014;Hardin et al. 2002). Moreover, Goldberg (2005) provides evidence of different effects of similarities in recruiting processes: While an impact of race similarities (with respect to white recruiters) was observed, in case of sex a dissimilarity effect was found (male recruiters preferred female applicants) and no age effect was identified. Similarity-effects also can attenuate discrimination: According to Cox and Beier (2014) raters' age has an effect on the attribution of older workers poor performance, i.e., compared to the poor performance of younger target persons, older raters attributed the poor performance of older workers to external and controllable factors while younger raters attributed it to stable factors.
Finally, literature also provides evidence regarding complex relations with regard to gender: Findings by Luxen and van de Vijver (2006) indicate that in selection processes individuals prefer attractive opposite-sex applicants, and additionally women prefer female applicants with low attractiveness over those with high attractiveness. Kunze and Miller (2017) observe complex relations between female superiors and peers: While they find "that higher shares of female workers at the next highest rank are associated with significantly smaller gender gaps in promotion" (Kunze and Miller 2017, p. 774), higher female shares at the coworker level reduce promotion rates for women. According to Lyon et al. (2004) in mentoring homogeneity of the mentoring dyad improves the outcomes for the mentee, whereas Ortiz-Walters (2009) only find a positive effect for male mentees in same-sex mentorships. In sum, literature provides mixed evidence with respect to the discriminating effects of similarity. Yet, overall, there exists broad evidence that discrimination related to similarity-effects at least under some circumstances exists and that it can be effectively mitigated by increasing diversity of the personnel applying the HR practices e.g., recruiting committees. Moreover, Taber and Hendricks (2003) point to a relation between networks and similar-gender and similar-race effects: They find evidence that "the more employees an organization has of the same gender and race as a newly hired employee, the more likely that employee was to have been hired through employee referral. The results support the idea that networking helps explain the use of employee referrals" (Taber and Hendricks 2003, p. 314). These findings again point to the importance of considering the identified parts of the microfoundation simultaneously to mitigate discrimination.

Discussion of the Finalized Model
Based on the previous content analysis in the present section we finalize the model as illustrated in Fig. 7 by elaborating on the interrelationships between the identified individual mechanisms. In the course of the previous discussion, three categories of the microlevel were identified which form the basis of discrimination in the workplace as conveyed by HR practices. Within these categories, individual subcategories that have a particular influence on this discrimination could be carved out. Isolated key findings of this analysis can be summed up as follows: Access to human capital constitutes an important determinant of discrimination along all analyzed diversity dimensions similarly to the existence of stereotypes and prejudices, while the effect of external pressure on discriminating practices within firms is ambiguous. Additionally, as demonstrated by Dencker (2008), particularly in times of upheaval firms are open for addressing issues related to discrimination. Thus, both governmental and corporate policy makers should use situations of upheaval to introduce measures to avoid discrimination. Moreover, selection methods can exert side-effects which lead to (unintended and undetected) discrimination against particular groups. Consequently, HR practitioners should search for such unintended side-effects and curb them by choosing more appropriate selection methods. Additionally, findings related to social construction indicate that in order to break self-stabilizing processes inducted by social construction, persons with atypical career paths have to become more salient, as they challenge the assumptions underlying social construction processes. Furthermore, findings related to networking indicate that it is recommended that disadvantaged groups be included in appropriate networks. In doing so, not only the networking of these groups should be promoted, but above all existing networks should be broken up. Related to the effects of organizational practices and norms one can derive that as the underlying behavioral patterns of these practices are deeply rooted and benefit the dominant group, external impetus, e.g., through consulting or coaches, is needed to break them up.
In the present section we take this discussion two steps further. First, we discuss the relations between the subcategories within each category and second, we thereafter look at the relations between the three main categories. Thereby we stepwise elaborate on the complex web of linkages between the individual mechanisms. Based on this discussion we than argue for a new theoretical perspective as a starting point for future research.
As discussed in Sect. 4.2., one large stream in the economic category clusters around the topic of human capital and the discrimination that results from restricted access to possibilities of building this capital for certain groups. These mechanisms are linked to a rational choice perspective, i.e., companies select, train, promote, etc. only those individuals who are most suited, where most situated does not mean who has the initially most talent but who has access to build human capital. Firms have limited incentives to use their own resources to remove the barriers to human capital accumulation faced by certain groups, as this can result in costs without immediate benefits. The second component analyzed in the economic category, external pressure, should provide guidance on how to break this dilemma. For example, there are also some studies that fundamentally show positive effects for disadvantaged groups through regulatory measures (Gregorič et al. 2017;Ilmakunnas and Ilmakunnas 2015;Pärnänen 2012). Yet, other scholars observe detrimental effects of anti-discriminating policies (Lai and Sarkar 2017;Messe 2011;Ng and Wiesner 2007;Neumark and Button 2014). In addition, the literature classified in this area also shows that there are opposing claims from different stakeholders, like investors versus clients, which partly counteract antidiscrimination activities. These observations clearly indicate that neither the idea of meritocratic HR practices to select the best fitting person nor stakeholder pressure contain incentives which are strong enough to abolish discrimination. Consequently, to support firms to mitigate discrimination, regulation must ensure that companies operate under equal conditions so that no competitive advantage is gained from discrimination based on e.g., customer preferences. In addition, legislation and reforms so far are not always consistent leading to subtle discrimination or behavior opposite to the intended one. Thus, in the future, the legislator will have to design its regulations based on explanatory models that can depict such complex interactions. At the same time, findings bei Dencker (2008) indicate that companies act if they are in a state of transition anyway, i.e., antidiscrimination regulation should be considered every time there are major regulatory changes, as these lead to fundamental transformation processes and thus more openness on the part of companies.
The discussion of the behavioral determinants revealed two broad literature streams dealing with prejudices and stereotypes on the one hand and relational demographics and related concepts on the other hand. The effect of the latter is based on the fact that decision makers in HR processes perceive a similarity between themselves and the people affected by the processes, e.g., a hiring process. This is nothing more than a special form of (positive) stereotypes and (positive) prejudices towards people similar to oneself, which is rooted in a positive self-image, i.e., a person perceives herself as positive, thus she also perceives others who are similar to herself as positive. Thus, a close relationship can be identified between these two mechanisms which is based on similar cognitive processes. More precisely, existing stereotypes and prejudices activate the mechanism of relational demographics and similar phenomena, while the latter stabilize the former. Consequently, in order to curb discrimination induced by these mechanisms, the close relationship between both mechanisms has to be considered. Moreover, attempts to reduce discrimination rooted in these behavioral determinants have to consider both the deep entrenchment of existing stereotypes and prejudices and a lack of willingness to deal with one's own position in this context, as the deconstruction of stereotypes and prejudices also requires the critical engagement with oneself.
Within the possible range of socio-structural determinants we identified particularly four structures to which scholars have devoted considerable effort: selection methods, social construction, networks, as well as organizational practices and norms. As discussed in Sect. 4.2., each of these structures alone exerts a considerable impact on discriminating processes within companies. Beyond that, however, they can also work together. The choice of certain selection methods depends on organizational norms, as organizations tend to look for those employees who fit their existing structures. Social construction processes depend on and determine existing organizational norms. In addition, they take place increasingly in certain networks, as there is a strong exchange between network members here, and therefore draw on the archetypes found in these networks. Therefore, in summary, the aforementioned structures reinforce each other in perpetuating existing discrimination practices. Accordingly, breaking up only one structure while retaining the other will not effectively eliminate prevailing discrimination. If, for example, selection methods are adapted so that previously disadvantaged groups now have access to certain jobs, existing networks and organizational practices and norms may nevertheless hinder the advancement of these individuals in the company and, in extreme cases, lead to their voluntary withdrawal. Conversely, opening up networks is of little use if the people targeted do not gain access to the company at all due to selection methods. Both processes also lead to social construction processes, the result of which is the realization that the group of people whom one wanted to promote through the above-mentioned measures has no interest in doing so. In this context, the findings by Eddleston et al. (2004) are of particular interest. They observe that the exposure to powerful networks positively affects male managers' promotions, but not female managers' promotions. They show that the mere access to networks does not have a similar effect on individuals' success regardless of their affiliation with certain diversity groups. Thus, individual measures are not very promising, i.e., companies have to establish antidiscrimination programs which encompass all of these sociostructural mechanisms in a wholistic approach.
The previous discussion can be extended to another level by focusing on the relations between the three categories. The overview points to a possible relation between the behavioral and the socio-structural determinants, whereby the former determine the latter and the latter stabilize the former. For example, on the one hand, the results of Harris (2001Harris ( , 2002 and Bernardin et al. (2012) suggest an effect of the activation of stereotypes and of relational demographics on the results of selection methods. On the other hand, the transparency of selection methods can lead to an amplification of negative stereotypes, as this reduces the performance of the candidates and thus leads to a confirmation of these stereotypes. Stereotypes also represent antecedents of the above-mentioned social construction processes (e.g., Sharp et al. 2012), but they are also reinforced by these, since stereotypes cannot be broken, as certain groups are continuously excluded and do not get the chance to disprove them. Furthermore, the fulfilment of the associated stereotypes as well as the existence of relational demographics determine access to certain networks (e.g., Blum 2015).
Also, based on the literature a relationship between behavioral and economic determinants can be presumed. On the one hand, stereotypes make it difficult for certain groups to build up human capital (e.g., Fitzsimmons et al. 2014). On the other hand, the lack of human capital stabilizes existing stereotypes among certain minorities. For example, according to Koeber and Wright (2006) decision makers apply signals derived from human capital characteristics, which show to which extend female applicants depart from particular negative stereotypes of female employees. Thus, if these signals are not present due to the lack of a specific human capital, negative stereotypes may be perceived as confirmed.
Finally, there might also be a stabilizing relationship between economic and socio-structural determinants. For example, as analyzed by Tienari et al. (2013) certain practices of executive search, which are affected by social construction, exclude women. In doing so, they make it more difficult for women to access these areas of work and prevent them from gaining experience here and build up human capital. This in turn, can stabilize the social construction of executives as being male. Organizational practices and norms can exert a similar stabilizing effect. For example, Cassirer and Reskin (2000) observe that men are more likely to be located in organizational positions that encourage workers to hope for a promotion, which in turn lowers women's aspirations in this context. This can prevent them from building up human capital in higher positions.
The previous discussion indicates that no single research perspective alone is capable to cope with the complex mechanisms underlying discrimination induced through HR practices. Misaligned economic incentives overlap with the effects of prejudice, stereotyping and relational demographics, while entrenched social struc-tures and practices perpetuate the resulting counterproductive effects. This can be illustrated for example by the effect of selection methods: As shown by Dickens and Kane (1999) and later Jacksch and Klehe (2016) structured selection methods, especially race-blind screening processes and transparency during personnel selection, which are seen as a way to attenuate discrimination by making selection decisions a structured and thus no longer arbitrary process (McCarthy et al. 2010), can foster (negative) stereotypes and prejudices. If they are designed in an inadequate way, they compromise the performance of candidates through activating negative stereotypes in their mind. As a consequence, the negative results regularly observed among members of the same group induced by these selection methods may lead to the consolidation of existing or the creation of new (negative) stereotypes and prejudices. This in turn, can further limit the access of these groups to possibilities to build up further human capital, while regulatory measures to support these groups might end up in more subtle ways of discrimination.
Another example goes as follows: The significant lack of human capital among individuals identifiable as a group due to gender, race, age or other similarities can be recognized as a pattern within this group related to a particular lack of relevant human capital, which in turn stabilizes (negative) stereotypes and prejudices and fosters existing social construction processes (e.g., Caputo 2002). In turn, decision makers' prejudices and stereotypes also can lead to the devaluation of an out-group members' human capital, which further stabilizes social construction, i.e., the required human capital is objectively present but is not considered by the decision makers in the HR process. In this case, even measures to increase human capital in a certain group of people are useless, because their result is not perceived by decision makers.
In sum, in order to reduce discrimination in companies effectively, one has to break up detrimental economic incentives, prejudices and stereotypes as well as counterproductive organizational structures and practices simultaneously. To do this successfully requires an interdisciplinary perspective. However, the adoption of such a perspective is made more difficult by the fact that some of the research takes incompatible perspectives. This is reflected in the very different theoretical starting points of the studies classified in the three categories. Particularly, the effects of behavioral determinants are related to a categorization process dependent on individuals' characteristics, either related to a comparison with the self or with predefined stereotypes which can be further linked to prejudices. This categorization is fundamentally different from the categorization process in an economic sense, where the decision maker follows a rational choice approach which is linked to profit maximizing or loss minimizing. These two perspectives do not have to be conflicting, but they can be. It is conceivable, for example, that despite an economic incentive to avoid discrimination, prejudice or relational demographics may lead to a discriminatory decision. Conversely, an economic incentive in the sense of external pressure or an applicant from a minority group with excellent human capital can also be so strong that it overrules the effect of prejudice or relational demographics.
The identified model proposes a structured way to cope with this complexity by identifying a problem hierarchy, which can be worked off step by step. At the first and the lowest level, the subcategories are examined individually, i.e., there are indi-vidual concrete questions that are analyzed using a specific theoretical perspective, for example the impact of selection methods on the chances to get a job for different groups, i.e., Bernardin et al. (2012) observe discriminatory effects of an assessment center against certain racial groups (non-whites). These questions are grouped into the main topics or subcategories presented in Sect. 4.2., such as networks, selection methods or human capital. From the literature identified, findings can be summarized for each component and recommendations for action can be derived. This represents a level of complexity to which, for example, regulation or companies can make good use because the recommendations for action derived are unambiguous and thus implementable. As the, albeit not comprehensive, literature indicates, on this level already a wide range of insights are gained.
However, as the previous discussion in this section indicates, the three determinants or components form a second level and the relations between the subcategories within each determinant are complex (economic determinants) or the subcategories reinforce each other (socio-structural and behavioral determinants). This shows that further insights could be generated by research that stays within the identified determinants and empirically analyzes the connections between the individual subcategories in more depth and identifies further subcategories by literature reviews with another scope in terms of time frame, databases, and diversity components to complete the picture. Finally, on a third level strong interactions can also be identified between the three determinants. As a result, the above-mentioned individual approaches to eliminating discrimination do not appear to be very effective in the long run, which makes a comprehensive perspective warranted. However, there are major differences between these three determinants with regard to the basic theoretical orientation, which can inhibit interdisciplinary collaboration. For example, a prejudice-based explanation of discrimination in the hiring process uses different explanatory mechanisms than a human capital-based approach: Prejudices lead to the unwillingness of a decision-maker to hire a certain person due to feelings of discomfort (e.g., Fernando et al. 2016;Jasper and Waldhart 2013), while a lack of human capital of a person results in a decision based on economical rational. The former reason for discrimination can be mitigate by measures to change prejudices in the decision-maker, while the latter needs actions which reduce barriers to building up human capital for the person affected by the decision makers decision. This points to the need of developing on common theoretical ground which unifies the different explanatory approaches. One first step into this direction could be to accept for these differences but to also elaborate on the most effective ways to attenuate discrimination by identifying a prioritization of the different causes of discrimination to target the most serious ones. However, this requires unprejudiced cooperation between different disciplines, i.e., in order to unravel the complex web of discrimination mechanisms at the micro level, research itself must act in a non-discriminatory manner.

Limitations and Conclusion
Like any research, the analysis carried out above is subject to certain limitations. First, the present model focuses on a selection of potentially relevant diversity dimensions. The categories analyzed were chosen because they are central dimensions in the prevailing anti-discrimination legislation and their consideration is therefore of central importance for companies. However, this selection is not exhaustive and can be expanded in the course of further research. Second, although the analysis of the prevailing literature is based on a systematic overview, this can only cover a small part of the field due to the very high number of publications. Accordingly, the model represents only a starting point for further research that complements it by additional factors at the micro level. Nevertheless, we consider the present model to be an important first starting point for identifying existing interdependencies between individual factors. Third, the contributions identified could not be discussed in detail. However, this was not the objective of the present study, as it focused on the basic identification of factors at the micro level and their relationships to one another. The implications and consequences for practice of the identified micro level factors and their relationships could be explored by researchers in the future. Finally, the regional focus in this study, which is particularly on journals from the Anglo-American language area supplemented mainly by journals located in European countries, must be judged as limitation. Also here, future research should add a broader perspective. In summary, the preceding discussion, the developed model and the derived paths for future research provide a comprehensive and thus new perspective on the background of stickiness of discriminatory HR practices, despite the limitations of the approach mentioned above.

Supplementary Information
The online version of this article (https://doi.org/10.1007/s41471-022-00136-w) contains supplementary material, which is available to authorized users.