All in My Head? The Play of Exclusion and Discrimination in the Labor Market

Labor market discrimination is very difficult to pinpoint, even more difficult to measure and almost impossible to ?prove?. It has been studied in many disciplines of which economics and sociology are prime. The latter has focused more on the manner in which discrimination plays out and how it is related to different forms of social stratification. This paper reviews the literature and makes two main contributions: first, it builds a four-fold typology to think about discrimination?overt or covert; conscious or unconscious; legal or illegal and real or perceived. Second, it identifies screens and filters?devices through which discrimination plays out in the labor market. Unless more empirical studies identify the play of discrimination and exclusion, subordinate groups may well be told that discrimination is actually in their heads?that they are imagining it.


I. Introduction
It is almost impossible to "prove" discrimination in the labor market, yet most would acknowledge that it exists, and some would argue that the job you get and/or how much you are paid depends, in varying degrees, on who you are and the circles you move in. The insidiousness, the opaqueness, and the sophistication surrounding processes of discrimination are complex and difficult to separate out. Those who feel that they have been discriminated against frequently, risk being accused of being paranoid or indulging in conspiracy theories.
"It's all in your head", they seem to be told. Yet, the idea of differential labor market outcomes for different groups has long been the subject of theoretical and empirical work from across disciplines. It has for the most part focused on some social identity -race, ethnicity, gender, caste, religion and more recently, age, disability status and sexual orientation -as the major axis along which discrimination operates. This paper moves from the assumption that discrimination is complex because it is often hidden and plays out through processes based on social codes. Karl Polanyi (1944) famously argued that the labor market is a crucial site for the play of social relations. In that, it reflects the existing and historical inequalities in a society. So, placing the analysis of the labor market within the overall context of social stratification and social norms provides a more nuanced understanding of the processes through which discrimination plays out, and also helps in refining the very notion of discrimination. This paper is a step in the direction of unpacking these subtle processes by understanding mechanisms that abet and enable labor market discrimination. In doing so, it provides a typology of processes that underpin discrimination. It is based on a review of the literature on processes of discrimination and the norms and attitudes that accompany them. It takes forward previous work on the Indian labor market 1 .
1 See World Bank 2011a and 2011b.

4
The paper is divided into four sections. This first section is an introduction and anchors the context for this work. Section II and III offer, respectively, a brief discussion of the conceptual underpinnings and measures of labor market discrimination from a crossdisciplinary perspective. Section IV lays out a typology of processes of discrimination.
Section V identifies screens and filters -mechanisms through which discrimination plays out in the labor market. Section VI concludes.

II. Conceptual Underpinnings
The notion of discrimination in the labor market encompasses unfair restrictions (formal and informal) on entry, on wages and on upward mobility of subordinate groups. So, two groups with identical skills may have very different labor market outcomes based on mechanisms that exclude some groups while favoring others. Theories of discrimination in the labor market were developed mostly in the context of gender and race (see for example Arrow 1973;Becker 1957Becker , 1971. One strand of the literature that derives from the economics of discrimination focuses on the hiring stage and the fact that discrimination in hiring translates into lower earnings (Arrow 1973). Specifically, in the presence of imperfect information, employers may be unable to accurately assess the quality of workers who belong to historically excluded groups, and may offer higher wages or higher status jobs to workers from dominant groups (see also Phelps 1972). Thus, when information is incomplete, employers may screen candidates on the basis of perceived differences in productivity. For example in the US labor market, Black or Latino workers may have lower reservation wages in order to compete with otherwise observationally identical groups. Let us also be clear that only screening which is based on subjective criteria such as stereotypes of race, ethnicity or gender is classified as discrimination; that based on actual differences in skills or productivity is not so classified. This paper argues that ingrained understandings in the minds of both employers and prospective job applicants can affect hiring and wage, since they affect the manner in which both groups behave. In that sense, it may sometimes be difficult to separate out actual differences from perceived differences in productivity.
Human capital theories place the onus of differential labor market outcomes solely on human capital endowments. Hence, workers who are less qualified are expected to earn less than their better skilled counterparts. The story becomes more complicated when two groups of workers have observationally similar human capital endowments but dramatically different outcomes. As Glen Loury points out, "If the concern is economic inequality between the groups, then looking mainly through the lens of wage and price discrimination is unlikely to shed much light on the problem." (Loury 1999: 14). The literature on segmented markets offers a more nuanced understanding of differential labor market outcomes for different groups of workers. Becker (1957) famously predicted the segregation of workers into sub-economies as a result of discrimination in perfectly competitive product and labor markets. Within these sub-economies, however, the Beckerian model assumes that workers with identical characteristics would receive equal wages. However, wage differentials between workers that appear quite similar in terms of skills, could arise in imperfect product and/or labor markets. Such imperfection would mean that rents are shared with workers in the form of higher wages, or efficiency wages are paid to workers resulting in wage premiums across industries, occupations and firms. In addition, employers and firms may practice a "taste for discrimination" based on different market valuations of group-specific characteristics (Arrow 1973, Becker 1971. As a result, there would be wage differentials due to such discriminatory practices on the part of employers, and similar workers doing the same job would be paid differently depending on their social group. The sociological literature on segmented labor markets focuses on the importance of social relationships in labor market outcomes. Dual market theories think of labor markets as primary and secondary constructs -the first being high status with better wages and the second comprising workers who may be discriminated against and who receive lower pay and have worse conditions of work (Doeringer and Piore, 1971). Economic theories see differences in labor market outcomes between different groups as arising from the marginal commitment to the labor market that such groups might have -for instance, women's lower commitment compared to men's -while sociological theories attempt to get to the social dynamics that lead to primary and secondary markets. Or, more generally, economic theories focus on the outcomes and sociological theories also explain why those outcomes obtain.
At the core of labor market segmentation are social groups and institutions. The processes governing allocation and pricing within internal labor markets are social, opposed either to competitive processes or to instrumental calculations. The marginal labor force commitment of the groups which creates the potential for a viable secondary sector of a dual market is social. The structures which distinguish professional and managerial workers from other members of the labor force and provide their distinctive education and training are also social. -Piore, 1983:252 Taking forward Piore's (1983) conception of the labor market being a social entity, a strong body of literature focuses on the role of norms, values and culture in determining labor market outcomes and processes. For instance, most economic theories distinguish between demand for and supply of labor, but this paper argues that once you take into account the effect of macro level norms, the distinction between demand and supply of labor gets blurred. Take the case of women's market work. That norms operate at multiple levels, permeating the household as well as the public spheres is now well recognized (World Bank 2012). So, discrimination as well operates at, at least two levels -the macro level, when institutions discriminate, such as when unfair laws are passed and at the micro level when individuals are not fairly treated or compensated.
Take the case of Muslim women's employment in most countries with large proportions of Muslims. Despite considerable heterogeneity across the world, the practice of veiling and norms of physical segregation of sexes prevalent in Muslim societies, are often implicated in the low labor force participation rates of Muslim women. Hiring authorities in India for instance, sometimes bemoan the low numbers of female Muslim applicants to public sector jobs, but Das (2004) in an analysis of Muslim women's employment using national labor force data and qualitative interviews points out that cultural norms operate in a complex manner. In fact, hiring authorities in some countries have a strong role in reinforcing norms of seclusion, even when individual women would like to participate, leading to something of a self-fulfilling prophesy. This paper discusses how formal and informal institutions serve to reinforce the status quo for female workers. While it may be the case that norms of seclusion prevent Muslim women from entering the labor market, hiring authorities equally do not expect Muslim women to participate, because of their stereotype of the veiled Muslim woman. Finally, it is difficult to separate out whether it is the culture of seclusion or the perception of a hostile workplace that keeps Muslim women out of the labor force, because culture by itself is hardly ever the sole driver of labor supply decisions. And this apprehension of hostility may also keep other subordinate groups from applying to jobs, which they may be well qualified for. Clearly therefore, it is impossible to put a finger on "discrimination" and where or how it comes into play -whether at the level of the individual, the household, the community or the labor market, but it is easy to see in India for instance, that even after controlling for human capital endowments, Muslim women are less likely than Hindu women to be employed.

The labo(u)r market has internalized the so-called seclusion ethic of Muslim women.
This may translate into a situation where the prevailing notion … may prevent them from being hired -leading to something of a self-fulfilling prophecy. My conversations in Maharashtra with government personnel in positions where they recruit clerks, office assistants, teachers or nurses indicated that the stereotype of the veiled Muslim woman is firmly ingrained in the collective psyche of the governmental recruiting authorities. (Das 2004 page 203) One of the insidious ways in which exclusion plays out in the labor market is through systematic assignment of certain individuals or groups to certain occupations -or what the sociological literature terms "occupational segregation". There is a prolific body of empirical literature on "occupational sex segregation", as women are overrepresented in "female" occupations like teaching, the arts and care services, but underrepresented in the higher paid science, technology, engineering and mathematics (STEM) jobs. Reuben et al (2014) designed an experiment where hiring authorities were given no information about candidates other than their appearance, and found that men were twice as likely to be hired for a mathematical task as women were. If ability was self-reported, women still suffered and the authors attribute this to the fact that employers may not fully account for men's "tendency to boast about performance". When full information about candidates' past performance was provided, it reduced "discrimination" but did not eliminate it. The authors conclude that implicit stereotypes predict not only the initial bias in beliefs but also the "suboptimal updating of gender-related expectations when performance-related information comes from the subjects themselves" (Reuben 2014: 4403).
Occupation segregation based on caste represents an extreme form of institutionalized labor market discrimination. Caste is a system of social stratification prevalent in India and Nepal, as well as in other South Asian countries and is closely tied to norms (such as of purity and pollution) and in turn, to the life chances of individuals from lower castes. In short, the caste system is an occupational division of labor, canonized in an elaborate religious ideology. Sub-castes are assigned to specific occupations that also determine their social standing. Lower castes are thus restricted to menial, low paying and often socially stigmatized occupations while upper caste groups are concentrated in preferred occupations. This horizontal segregation results in workers from different social groups being ascribed into different occupations and jobs and being socially constrained from moving out. All these have an impact on the world-view of both privileged and underprivileged groups and spill into market interactions as well -most notably in labor market interactions. From an epistemological perspective, these characteristics set the caste system apart from other forms of stratification (such as gender or race) and their bearings on the labor market.
Using data from India, and comparing Scheduled Castes 2 and non-Scheduled Caste males, Das (2006) and Das and Dutta (2008) devise a notion of "glass walls" where individuals historically assigned to certain occupations based on their caste, also face barriers to leaving 9 these traditional occupations. Micro studies point to the possibility of small entrepreneurs belonging to the lowest castes, especially in India's rural areas, being prevented from moving out of caste based occupations into self-employed ventures through social pressure and ostracism (see for instance, Venkteswarlu, 1990, cited in Thorat, 2007) and in other ways being denied fair opportunities to participate in more lucrative trades. Iversen and Raghavendra (2006) also find similar patterns in small eating-places in Southern India where workers are typecast into their traditional caste occupations. While there are few studies that show the existence of such glass walls in other contexts, in fact subordinate groups may be more generally prevented from leaving assigned occupations through a set of processes that are described later.
The idea of "glass walls" draws on the more commonly used notion of "glass ceilings" which are invisible barriers to vertical advancement. The latter notion was originally used to explain gender differentials in earnings but can be applied to race or ethnicity as well. The existence of a glass ceiling implies that the gap between earnings of disadvantaged workers and those from privileged backgrounds would be higher at the top end of the earnings distribution rather than at the bottom or middle, conditional on worker characteristics.
Thus, glass ceilings manifest as wide differentials at the upper end of the conditional distribution (i.e., after taking productive characteristics into account) but there is less disparity at the low and middle portions of the conditional distribution.
Chances are that both glass ceilings and glass walls co-exist, indicating respectively a vertical segregation (within the same employment type, workers from different social groups may be represented differently in the hierarchy of positions) and horizontal segregation (workers restricted to their occupations). In both cases, institutional barriers and social attitudes may hamper upward movement of workers through occupational categories and social hierarchies. Women in most developing countries, for instance, are concentrated in agricultural and allied industries and underrepresented in those occupations, which pay the highest (Das and Dutta, 2008).

III. Measurement Issues
As pointed out earlier, the complexity of the processes surrounding both the demand and the supply of labor make it almost impossible to accurately measure discrimination. That being so, the most commonly used techniques are decompositions that rely on traditional and easy to procure labor market surveys. They are based on measures of discrimination derived from between-group comparisons of market outcomes for "favored vs. disfavored groups", controlling for productivity-related individual characteristics (Salkever and Domino, 1997). Such analyses for the most part, look at earnings as the outcome of interest, while controlling for human capital endowments. Another way to get at discrimination is through audit studies, in which matched pairs of interviewees (actors) of different race or gender, respond to advertisements and attend interviews (see for example Yinger 1986, Neumark et al 1996. Well-known concerns with audit studies include imperfect matching (and training) of actors, the inability to double-blind (auditors may wish to please the researcher, and do not care about actually landing the job), expense (which limits both sample size and the scope of studies, which are typically focused on narrowlydefined occupations), and the need to deceive employers, which raises issues of informed consent (Kuhn and Shen, 2009).
Perhaps the most accurate way of measuring discrimination is through experimental studies first conducted by Daniel (1968) and soon after by Jowell and Prescott-Clarke (1970).
These type of studies simulate natural experiments to predict the behavior of individuals and groups in a particular setting. The spate of studies that followed have focused mainly on race and gender as the axes of discrimination, but there are increasing number of studies being conducted to test for discrimination based on ethnicity, immigrant or disability status and other axes. These studies do not attempt to provide any generalized picture of discrimination but have been instrumental in the widespread acceptance that discrimination exists. An increasing body of work suggests that in fact, people may not even be aware of the fact that they are discriminating or excluding and that a combination of social norms and practices "make people" behave in certain ways (World Bank 2014).
Since discrimination depends to a great extent on societal values and attitudes, many surveys are attempting to measure these aspects inter alia, in relation to labor market. The World Values Surveys, the Life in Transition Surveys, Latino Barometer and the Afro Barometer are examples of surveys that capture cross-country attitudes. Some governments have taken the initiative to field their own surveys showing their commitment to making a dent in discrimination. The Scottish Social Attitudes Survey conducted since 2002, examines views on whether different groups of people are equally suitable to a particular employment role (that of a primary school teacher) and compares responses to this question with views about these same groups marrying into a person's family. It also explores gender stereotyping in attitudes to parental leave, attitudes to older people working and perceived labour market competition from particular groups. The aim of this survey is specifically to provide inputs into policy to address discrimination. Other smaller studies have looked at perceptions of discrimination, and tried to "get into the minds" of hiring managers and disadvantaged workers. For instance, Jodhka and Newman (2007) interviewed human resource managers on the hiring practices of their firms, and found that caste-based stereotypes color the hiring process such that both very low caste and very high caste candidates are disadvantaged.
Ultimately, measurement depends on the outcome of interest. Where does discrimination play out? Studies of the labor market have historically looked at hiring practices, wages and promotion. In economies with large informal labor markets and concentration of workers in self-employment, these categories capture only some arenas of potential discrimination.
Others are mobility across occupations, access to successful entrepreneurship, to credit and other markets, including to timely and relevant information. For still others, notably women in countries with low female labor force participation, the barrier is not wages or occupation; it is the very entry into the labor market. In order to consider multiple labor market outcomes, some researchers have used a hierarchy of employment types and looked at group-based access to different types of employment (see Das, 2002;Lanjouw and Shariff 2002;Barooah 2010).

IV. Typology of Discrimination
Processes of discrimination are seldom well thought-out conspiracies. For the most part, the "code" is hidden even to those who are allegedly executing the act of discrimination, since it is so ingrained in social norms. While a limited body of literature deals with ways in which different practices play out, there is no single typology of these processes. This section lays out an organizational framework to think about processes of discrimination and proposes a typology (Figure 1). The typology rests on four types of discrimination  individual may be someone who has a particular accent in their speech, who comes from a certain socioeconomic background or one who has a certain height or weight.
It may equally be unconscious, and be based on stereotypes, prejudices and stigmas, which are socially constructed and influence a range of day to day social interactions 13 between groups and individuals. Stereotypes about groups can be so ingrained that hiring managers or peers may not even realize that they have them (see Deshpande and Newman, 2007;Loury 1999) or may not consider them stereotypes at all -but rather facts of the labor market. For instance, the stereotype of Roma or African American workers as being "lazy" or of women's low commitment to the labor market, are so internalized by the majority that they are often regarded as truisms.
Such stereotypes pervade hiring authorities in ways that have significant bearing on hiring processes. Cornell and Welch (1996) show that employers are able to judge job applicants' qualities better when candidates belong to the same group as them, and the top applicant is likely to have the same background as the employer. They moreover point to some "reinforcing effects" of such "screening discrimination". For instance, employers may infer the quality of applicants from the behavior of other employers amounting literally to herd behavior, under which preferences expressed by third parties drive decisions.
b. Discrimination may be overt and announced or it may be covert. In the case of the former, job ads can openly ask for females for certain jobs. Or there may be differential retirement ages for men and women. In other cases, it may be covert discrimination that is not easily identifiable. Covert discrimination is the manner in which occupational sex segregation sorts women into "female" occupations and men into "male" ones. There is no observable evidence of discrimination either empirically or in the policy realm. Sexual harassment is a form of discrimination against less powerful workers and is also for the most part covert and difficult to establish.
c. Here the distinction between legal and illegal discrimination becomes especially relevant. For instance, differential retirement ages for men and women in China are legal, but hiring authorities do not regard them as being discriminatory and in fact consider the difference, a concession to women. The fuzziness around whether something is discrimination or not, in legal terms, makes international conventions and national laws especially difficult to abide by, unless there is a strong lobby that Women were also relatively more likely to work outside the home, employ more educated workers, and in paid and full-time jobs where the reform had been enacted, after controlling for time and location effects.
d. Finally, discrimination may be perceived or real. Perceived discrimination may well be the result of historical inequalities that relegate certain groups to a lower place in the social hierarchy, because of which they collectively and individually perceive themselves as being discriminated. Even when laws seek to eliminate such historical discrimination, groups that have been historically excluded lack the trust and confidence that they will in fact be treated fairly. In such cases, the dominant group and the state need to do more to build the confidence and to enable subordinate groups to participate and not to opt out. The literature on race is replete with descriptions of perceived discrimination. The opposite of perceived discrimination is real discrimination, which is self-explanatory, whether or not it is legal and overt.
To conclude, it is important to recognize that there is both a strong cultural and a temporal dimension to whether a practice is regarded as "discrimination". The same practice which would not have been considered discriminatory some years ago may now be accepted as discrimination. Take the case of harassment in the workplace. Behaviors that are now considered to be harassing or bullying were once normative, and there was no recourse to reporting them. Basu (2003) points out that even after the passage of the Civil Rights Act of 1964, US federal courts typically refused to view sexual harassment as a form of employment discrimination under the meaning of the statute. He goes on to describe how sexual harassment, which emerged as a concept of discrimination in the US first, moved from being defined as overt threats to innuendo and now to the harassee's feelings of being harassed. Moreover, sexual harassment, which used to be considered specific to men's harassment of women, has now gained acceptance as women's harassment of men and same-sex harassment as well. He documents through case law how this came to happen.
With increasing awareness, lobbying and legislative backing, sexual harassment is acknowledged in many other countries, but still remains much more difficult to pin down than other forms of discrimination. Take also the case of nepotism. What is widely believed in countries like the US as an inappropriate use of kinship ties, or considered conflict of interest, is not regarded as being egregious in many other contexts.

V. Screens and Filters: Devices of Discrimination
This section outlines some of the ways in which status and privilege play out in assigning workers to different jobs or to the labor market in general. Since discrimination is often subtle and difficult to put a finger on, it is set up in a variety of ways, although the set-up itself may in some cases be unconscious. These could include the way job ads are framed, where they are posted, the way interviews are conducted and how applicant status is gleaned. A large number of studies have documented discrimination along lines of race, ethnicity and gender but increasingly, studies are also documenting the more "invisible" axes of discrimination -along lines of neighborhood or residence, good or plain looks, skin color, family background, elite or non-elite status, the way candidates speak English or the lingua franca. These mechanisms are screening or filtering devices that exclude some groups while including others. Figure 1 shows that discrimination can occur along a number of processes -some are real, overt, legal and conscious, while others are not. In some cases, the sheer invisibility and intangibility of the processes makes people wonder if they are real or perceived.
In periods when the labor market is tight and "good jobs" are few, potential employers may try and maximize their own utility by imposing additional filters. Homophily is the principle that a contact between similar people occurs at a higher rate than among dissimilar people. The pervasive fact of homophily means that cultural, behavioral, genetic, or material information that flows through networks will tend to be localized" (pg 416). Rivera (2008) makes the important point that homophily is not only an implicit cognitive bias that is an inevitable by-product of social interaction, but also an explicit criterion that firms use to screen candidates, but not necessarily for the purposes of resource control or opportunity hoarding. "Rather, in many instances, homophily appears to be more of a sense-making device in which individuals consciously and unconsciously seek points of commonality and make generalizations based on the presence or absence of common experience. Yet, regardless of the level of intentionality or "usefulness" in evaluation, selecting on similarity results in the reproduction of culturally homogenous organizations, even when not demographically so" (Rivera 2008, emphasis mine). Such "homophily" or gravitation towards candidates that most closely resemble the dominant workforce in an organization can be realized through the use of filters or screens used to make such selections; some of these screens are measurable and others are not. However, it is also important to add here that with increasing evidence on the value of heterogeneous work environments and the increasing evidence on discrimination, a large number of organizations are making efforts to seek out candidates that come from social strata different from the established culture of the firm.
Employers may often set up overt or covert filters through which candidates may have to pass. Those who expect to be discriminated may not reach the screening stage, largely because they "opt out of the game" 4 . When discrimination is overt, employers openly express their preference for a certain type of applicant. Kuhn and Shen's (2011) study of gender discrimination in a large sample of job advertisements taken from an internet job board in China, found that explicit, advertised gender discrimination is commonplace and over one-third of firms that advertised on the board during the authors' twenty-week sample period placed at least one advertisement stipulating a preferred gender. Thus, when it is legal to express such preferences, a large share of employers does so. The authors also found that large firms, state-owned firms, and firms seeking workers with high levels of experience tend to prefer men. Foreign-owned firms, and firms hiring for certain customercontact occupations tend to prefer women.
Status and "suitability" for a job can often be gleaned from an applicant's residence, as neighborhoods are often (unreliable) proxies for social class and implicitly, for human capital endowments. While clustering confers benefits of solidarity and networks for minorities, an address can also be an identifier and a stigmatizer. Clark and Drinkwater (2000) show that an individual's probability of being observed in a particular labor market state depends on the ethnic composition of the neighborhood in which s/he lives. In a study of England and Wales, they found that ethnic minorities who live in enclaves have a higher likelihood of being unemployed and a lower probability of self-employment than compatible individuals who live in less concentrated areas. They acknowledge that while individuals can sort non-randomly, the potential benefits of living with co-ethnics are trumped by the deprived nature of the urban areas in which minorities tend to cluster.
Like neighborhoods, language, speech patterns, accents and non-verbal cues (such as a particular sound or a head bob) can also be identifiers -they give salient clues about an individual's social status and family background. It is perfectly acceptable to ask for language skills as part of eligibility criteria but when spoken language identifies the social status of an applicant and has no bearing on the needs of a job, but influences hiring decisions, it is an enabler of discrimination. Once such an applicant is hired, the lack of a sophisticated spoken language or existence of non-verbal cues like physical mannerisms, can lead to discriminatory behavior in the workplace in much the same way as it does in educational institutions. This in turn could affect earnings and/or promotion. That foreign accents or those that reflect stereotypes about a worker's social status affect hiring decisions is known anecdotally, but this is also borne out in some of the empirical literature. Bendick et al (2009)  Filters or screens such as language seem to matter more when rationing of "good jobs" is in evidence; that is, when there are fewer good jobs and a larger pool of applicants. In such markets, Desai et al (2008) find based on Indian data, that knowledge of English has a statistically significant association with getting "good jobs". Once in employment, nonpreferred workers may face discriminatory attitudes that prevent them from being integrated into the labor market on equal terms as the dominant group. ILO (2011) cites an example from Flanders, Belgium, where an internal rule of an automobile components company in which 70 per cent of the employees are of foreign origin, stated that workers using languages other than Dutch on three consecutive occasions were liable to be dismissed. The justifications of this rule were "security reasons" and "respect for other workers".
I do not for a moment imply that lack of language skills is a mechanism for discrimination.
Both language and non-cognitive skills may be essential for a job and for that reason may exclude a certain category of applicants. In general, unilingual applicants do worse than bilingual or trilingual applicants in many labor markets and this is the nature of their human capital endowments. But like other aspects of human capital, there may be a pattern in stronger or weaker language skills. Upadhyaya (2007) finds from a study of the Indian outsourcing industry, that English-speaking graduates who also have western social skills fare much better in the burgeoning market. These criteria manage to exclude Scheduled Castes from the pale of this growing opportunity. She shows that, for the most part, threequarters of the workers in jobs in information technology are from higher castes, while the remainder is from the Other Backward Class (OBC) category -individuals who are placed higher in the caste hierarchy than the Scheduled Castes. Deshpande and Newman (2007) interviewed postgraduate university students and found that Scheduled Castes in degree programs show lower expectations about jobs than their high caste counterparts and see themselves as disadvantaged because of their caste and family backgrounds. Because they 20 arrive at college with weaker skills, on average, they often do not succeed in pulling even with more advantaged students; hence, they enter the job market with weaker English language and computing skills. In addition, they lack the social awareness and other abilities that dominate in the private sector culture. However, these small samples are in no way representative of the entire information technology industry and Banerjee et al (2009) find no evidence of discrimination against non-upper-caste (i.e. Scheduled Caste, Scheduled Tribe, and Other Backward Class) applicants for software jobs. They find large and significant differences between callback rates for upper-castes and Other Backward Classes (and to a lesser extent Scheduled Castes) in the case of call-center jobs.
A new line of enquiry looks at the role of physical appearance and presentability as predictors of labor market outcomes. Hamermesh and Biddle (1993) find that plain people earn less than people of average looks, who in turn earn less than good-looking people. The penalty for plainness in their analysis was 5 to 10 percent, which was slightly larger than the premium for beauty. The effects were slightly larger for men than women; but unattractive women were less likely than others to participate in the labor force and were more likely to be married to men with unexpectedly low human capital. Better-looking people seem to sort into occupations where beauty is likely to be more productive; but the impact of individuals' looks on their earnings is mostly independent of occupation.
Villarreala's (2010) study in Mexico similarly found that skin color matters, and that individuals with darker skin tone have significantly lower levels of educational attainment and occupational status.
Similarly, an increasing number of studies are looking at height and weight as predictors of success, or lack thereof, in the labor market. Case et al (2008) for instance found that each inch of height is associated with a 1.5 percent increase in wages, for both men and women in the UK. Half of the premium in their analysis can be explained by the association between height and educational attainment; and of the remaining premium, half can be explained by taller individuals selecting into higher status occupations and industries. In a more stark experiment, O'Brien et al (2012) found that obese women are more likely to be discriminated against when applying for jobs and also more likely to receive lower starting salaries than their non-overweight counterparts. Based on a new measure of anti-obesity bias, the authors showed survey participants' resumes with pictures of female candidates before and after they had undergone bariatric surgery -so, the same woman's picture was viewed more than once, in a set of resumes. The obese women were found to be regarded as much less suitable for employment, for leadership, and regarded as being less deserving of higher remuneration.
Employer preferences for a certain type of physical appearance sometimes translate into a preference for age; so, younger workers are considered more attractive and more suitable for some jobs. This is particularly the case for female workers. Lahey's (2005) labor market experiment to determine hiring conditions for older women in entry-level jobs in Boston, MA and St. Petersburg, FL found that a younger worker is more than 40% more likely to be offered an interview than an older worker. The hospitality industry is notorious for such screening for both men and women. The case of female crew in India's state airline, Air India shows that female crew members have only recently been allowed to assume inflight supervisory roles, after a long fought legal battle (Menon, 2011). Differential retirement ages are another case in point and many countries retire women earlier than they do men. In China for instance, women's retirement ages are lower than those of men in almost every category of state employment and are considered part of the state's role of protecting female workers.
In many countries, age based discrimination against women is often linked to the relationship of women with the state, more broadly. For instance, women workers may be treated as though they need protection and in a bid to curtail trafficking and exploitation of women, several countries have put in place age restrictions and other safeguards. The Indian Emigration Act was amended to prevent women under age 30 from migrating as housemaids and caregivers. NGOs expressed concern that this would lead to illegal and undocumented migration of younger women and prevent them from accessing any benefits they may be entitled to in the receiving country. Other countries also have age limits for women migrants, but whether this contributes to women's welfare or increases vulnerability is unclear (Manchanda 2007). There are three clear aspects to age based discrimination. First, it could be legal and justified on the basis of the state's paternalistic attitude towards women and their families, such as in China or in respect to migration policies. Second, it could be illegal and difficult to "prove" that it occurs systematically, such as in the case of Lahey's study. Third, it could be illegal but justified as a business case ("women are unable to cope physically with the demands of the job") such as in the Air India case.
Yet an important marker of status is family background. As noted earlier, employers unconsciously may select candidates that are from their own background, but in other cases, employers may have a clear vision of the entity they wish to create. Here, social background of their employees, as opposed to only their human capital endowments, plays an important part in building that vision. That names can be identifiers for social status is well known. Caste names in India and Nepal are often dead giveaways (unless they are deliberately changed). Deshpande and Newman (2007) found that prospective hiring managers almost universally ask questions about family backgrounds during employment interviews. Upper caste students can offer biographies that are much closer to the uppermiddle-class professional ideal. While Scheduled Caste students perceive a hidden agenda in questions on family background, the same questions appear to upper caste students to be innocuous and sensible inquiries from a human resources perspective.
Finally, names of applicants can be another giveaway about their origin and socioeconomic status. Experimental studies in the US and elsewhere show that first names (that conform to racial or national stereotypes) can also mar individuals' chances in the labor market (see for instance Bertrand and Mulianathan 2003). In another instance, while Bravo et al's (2008) experiment in Chile showed no statistically significant difference in callbacks to applicants by employers based on social class, gender or neighborhood, yet it is instructive that their proxy for social class was surname. Finally not merely lower class, but upper class lineage can also be identified by names. Ying Pan's (2011) analysis is illustrative here.
It finds that Chinese husbands and wives of the largest local lineage (identified by name) are more likely to become village administrators, and young men of the largest local lineage are more likely to have local non-agricultural jobs and higher wages.

VI. Summary
This paper shows that labor market discrimination is very difficult to pinpoint, even more difficult to measure and almost impossible to "prove". The literature on labor market discrimination comes from many disciplines of which economics and sociology are prime, where the latter has focused more on the manner in which discrimination plays out and how it is related to different forms of social stratification. This paper reviews the literature and makes two main contributions: first, it builds a four-fold typology to think about discrimination -overt or covert; conscious or unconscious; legal or illegal and real or perceived. Second, it identifies screens and filters -devices through which discrimination plays out in the labor market. Unless more empirical studies identify the play of discrimination and exclusion, subordinate groups may well be told that discrimination is actually in their heads -that they are imagining it.