Advertisement

Journal of Economics, Race, and Policy

, Volume 2, Issue 4, pp 257–268 | Cite as

Double Discrimination: Is Discrimination in Job Ads Accompanied by Discrimination in Callbacks?

  • Eva O. Arceo-GomezEmail author
  • Raymundo M. Campos-Vazquez
Original Article

Abstract

Audit studies have found employment discrimination in a variety of contexts. In Mexico, an overlooked aspect of this discrimination is that job advertisements usually include explicit criteria of gender, age, attractiveness, or require a photograph in the resume. These specifications, which we refer to as “explicit discrimination,” may affect which applicants receive responses. We pose two hypotheses. First, a reduction in matching costs may result in a higher callback rate from explicitly discriminating employers. Second, discrimination in the first stage of the hiring process could lead to other discrimination patterns in later stages. We test for such biases using a correspondence experiment, in which fictitious resumes with randomized applicant information were sent in response to job advertisements in Mexico City. Consistent with our first hypothesis, employers with explicitly discriminatory ads are 7.6 percentage points more likely to call at least one female candidate for an interview than those without explicit discrimination. With respect to our second hypothesis, the probability of a callback is close to 18 percentage points lower for married women responding to ads specifically targeted to women than for those responding to non-gendered ads. With respect to race, we find that only ads that include two or more discriminatory criteria resulted in a higher callback probability for white or mestizo women than for indigenous women or women with no photograph in their resumes. We find no economically significant results for men. Thus, although there is a higher callback rate from discriminatory ads, there is also exacerbated discriminatory behavior from employers, producing for some groups a type of double discrimination in the hiring process.

Keywords

Discrimination Gender Race Labor market Mexico Correspondence study 

JEL Classification

J10 J16 J70 O54 

Introduction

In the last two decades, economic research has provided clear evidence of discrimination in various markets.1 In many countries, discrimination is illegal, and researchers have used correspondence studies and other field experiments to uncover such practices.2 However, there are some countries in which open discrimination is to a certain degree tolerated. For instance, explicit discrimination in job ads has been found by Kuhn and Shen (2013) in China, by Delgado et al. (2016) in Mexico, and by Anand (2013) and Chowdhury et al. (2018) in India. These studies defined explicit discrimination to include criteria for age, gender, marital status, and other ascriptive characteristics. This type of job ad denies equality of opportunity in jobs to entire segments of the population.

In this paper, we analyze the callback rate behavior of employers that post discriminatory job ads. Explicit discrimination in a job ad does not rule out the existence of later discrimination within the groups that meet the initial discriminatory criteria. We have two working hypotheses in this regard. First, narrower searches may lead to a better qualified applicant pool, where the recruiter is able to call back a larger share of applicants for an interview. However, our second working hypothesis is that if employers begin discriminating so early and openly in the hiring process, then they may discriminate with a higher probability after the initial application, based on characteristics not specified in the explicitly discriminatory ad. There are two possible mechanisms behind this increased discrimination. First, the discriminatory ad could in itself be an expression of a discriminatory employer. Second, the discriminatory ad may trigger further discrimination by inhibiting an employer’s fairness in a sequential hiring process. This phenomenon would be similar to what is known in the dieting literature as the “what-the-hell” effect: once we violate a rule or norm, we may be more willing to commit further violations (Mazar and Ariely 2010). Another non-excludable hypothesis is that some employers are just adhering to the norm of explicit discrimination (Hoff and Walsh 2017). To test our hypotheses, we conducted a correspondence study in Mexico City.

Because discriminatory job advertisements are illegal in advanced economies, the literature on their effects is scant. Darity and Mason (1998) describe how in the era before the U.S. Civil Rights Act, job ads overtly discriminated against African-Americans and women. The Civil Rights Act of 1964 and the Equal Employment Opportunity Commission prohibited such explicit practices. Discrimination in job ads began to take less blatant forms and eventually disappeared over time. Research in more progressive countries thus uses indirect methods to uncover discrimination, examining such factors as wage gaps and occupational segregation through correspondence and audit tests. Countries in which explicit forms of discrimination persist also exhibit socioeconomic differentials across gender and racial categories. Lawler and Bae (1998) carried out one early study using job ads to examine discrimination in the labor market. They tested whether gender discrimination by multinational firms in Thailand depended on the firm’s country of origin, and found that multinational firms bring discriminatory practices in their cultures to their host countries. Anand (2013) collected close to 1000 ads from a newspaper in India and found that approximately 20% sought a specific gender. Chowdhury et al. (2018) found that a third of the ads in India did so.

Also recently, Kuhn and Shen (2013) found pervasive explicit discrimination in Chinese job ads: approximately 10% are directed exclusively to male or female applicants (roughly half and half), 25% have age requirements, and 10% include attractiveness or height requirements. The authors note that their ad-based approach to measuring discrimination uncovers different aspects of employer discrimination than correspondence studies; in particular, explicit discrimination “involves a conscious decision by the employer to invite only one group to apply,” whereas audit studies measure “both conscious choices and unconscious biases of employers” (p. 290). In this way, there could be discrimination in an audit study, even when there is no discrimination in job ads, or explicit discrimination in job ads where no discrimination was revealed in correspondence studies. Our paper presents evidence of a link between these two types of discrimination.

In their study of explicit discrimination in China and Mexico, Delgado et al. (2016) find that although job ads stating gender preferences are balanced between women and men, once age is introduced into the analysis younger women are preferred to younger men, and older men are preferred to older women. They refer to this finding as the “age-twist” in gender preferences, and they argue that it can be explained if employers are not completely revealing their preferences in job ads: that is, that there are unstated preferences regarding attractiveness, marital status, and leadership. For instance, job ads do not often discriminate explicitly by race.3

In this paper, we exploit a previously conducted correspondence study to test whether these unstated preferences produce biases in callback rates. In Mexico, European phenotypes are preferred to mestizo (light brown) or indigenous (dark brown) phenotypes,4 so if there is a large degree of unstated preference based on attractiveness,5 we should find that explicitly discriminatory ads tend to respond more to white than to mestizo or indigenous-looking female applicants. To our knowledge, ours is the first attempt to use a correspondence study to determine whether explicit discrimination in job ads is related to further discrimination in the hiring process.

This paper is an extension of the analysis presented in Arceo-Gómez and Campos-Vázquez (2014), where, in the spirit of the seminal work of Bertrand and Mullainathan (2004), we presented the results of a correspondence study examining employer discrimination along gender and racial lines in the Mexican labor market. The main findings were as follows: (i) surprisingly, women receive more callbacks than men; (ii) women experience racial discrimination, but men do not: white women receive more callbacks than either mestizo or indigenous-looking women; and (iii) women are discriminated against based on marital status, but men are not.6 However, in that paper, we did not consider the effect of explicit discrimination in callbacks.

In this paper, we test whether there is an interaction between explicit discrimination in job ads and discrimination in callback rates. The experiment consisted of sending fictitious resumes in response to job advertisements directed at recent college graduates. Online job searches are more representative for recent graduates, and by focusing on them, as in Oreopoulos (2011), we avoid the problem of experience playing a more prominent role. We sent comparable resumes in response to approximately 1000 online job advertisements, approximately eight resumes to each, randomly varying the gender and the photograph (which in Mexico are commonly requested in resumes), along with other observable characteristics of the fictitious applicants. The photographs represent three distinct phenotypes plus a control without photograph: white, mestizo (light brown skin), and indigenous or dark mestizo.

In our sample of job ads, approximately 20% expressed a gender preference: among these, 40% specified only male applicants and 60% only female. Of the total, 16.7% requested a photograph in the resume, and 5.8% expressed a preference for attractiveness.7 We followed the criteria of individual advertisements, sending resumes for men or women where a preference was stated, and excluding the resumes without photographs for those that explicitly requested one. We present our results by gender. For men, we find no significant relationship between explicit discrimination and callback rates. For women, we find that explicitly discriminatory ads tend to result in more callbacks to the narrow applicant pool than non-discriminatory ads to the unbounded pool, thus providing evidence of our first hypothesis. We find that explicitly discriminatory ads are 7.6 percentage points more likely to result in callbacks to female applicants than ads without explicit discrimination. Supporting our second hypothesis, we also find that among women, ads that explicitly discriminate based on gender result in more discrimination in callbacks based on factors such as marital status and phenotype than neutral ads. The probability of a callback is close to 18 percentage points lower for married women with ads targeted specifically to women than with neutral ads.

Our results point to the existence of double discrimination in the labor market: explicit discrimination in job ads, and in the case of women, further discrimination based on marriage and phenotype. We thus provide evidence of a form of explicit discrimination, beyond the exclusion in the advertisement itself, that has not been previously addressed in the literature. Our intuition is that married women face a priori discrimination based on their childbearing potential. However, this intuition does not necessarily explain why women-only ads discriminate even more than neutral ads based on marital status. In this paper, we advance three possible mechanisms behind the double discrimination interaction, but none of them explain why the double discrimination is limited mostly to women. Further research is required to uncover the mechanisms behind our results.

Our paper also contributes to the literature on labor market gender gaps in Mexico. Our results show that married women of childbearing age have demand-side barriers to jobs from two different sources: the usual discrimination in hiring, and the double discrimination from explicitly discriminatory employers. Married women in Mexico have unusually low levels of female labor force participation (Arceo-Gómez and Campos-Vázquez 2010). Our evidence of married women’s double discrimination may be an additional barrier to female work.

The rest of this paper is organized as follows. Section 2 presents the empirical strategy, which includes a description of the experiment and the estimating equations. Section 3 presents descriptive statistics of the resumes we sent and the job ads they responded to. Section 4 presents our results, and Section 5 offers some conclusions.

Experimental Setup and Methodology

To test whether gender and phenotype affect the probability of callback for an interview, we constructed a bank of randomized resumes and a bank of job advertisements. A typical resume includes identity information (name, photograph, address, email, cell phone number, etc.), marital status, high school and university attended, professional experience, hobbies, and some additional information (like time availability and willingness to move to another city). We sent an average of eight resumes in response to each job advertisement. These were determined based on gender and phenotype (three phenotypes and a resume without photograph as control).8

We created resumes using information from samples available online such that the professional experience of our fictitious candidates was realistic. We then contacted recent college graduates and asked them to modify the resumes as if they were their own. We gave the fictitious candidates eight of the most common names and surnames in Mexico. We chose mainly surnames ending with -ez, because these are very common and are not associated with ethnic or social background.9 Following Lahey and Beasley (2009), we randomized all information across resumes10 and created ten sets of eight resumes each for six different academic majors and two experience levels; our bank of resumes thus has 960 different resumes. Each name was associated to an email account and a cell phone number. Because the characteristics of the resumes are randomized, each photograph corresponds on average to resumes of the same quality.

The photographs used were of three men and three women representing the different phenotypes. All of the photographs had a white background and the subjects wore similar attire.11 The photographs were taken and used with the written informed consent of the subjects, who were told about the nature of the experiment and the way in which their image would be used.12 For the purposes of this study, we defined three phenotypes: a European, or white, phenotype; a mestizo phenotype, with light brown skin; and an indigenous phenotype, with dark skin. These definitions are not necessarily related to particular eye or hair color or to “pure” ancestry; they are categories of race as socially perceived, based on skin color and facial features.

Our focus was on job advertisements seeking applicants with 0 to 3 years of experience in the following fields: business administration, accounting, economics, industrial engineering, electronic and telecommunications engineering, and computer and systems engineering. These fields were selected to maximize the number of available ads and to achieve gender balance: in 2008, 48% of the graduates in these majors were women (ANUIES 2009). Given a balanced gender distribution of graduates, in the absence of discrimination, we should expect a balanced callback rate.

Resumes were sent between October 2011 and May 2012. We collected advertisements on a weekly basis from internet websites commonly used for job searches in the Mexico City metropolitan area.13 We collected information on the job characteristics from each advertisement, but there was little specific information about the employers (such as firm size or revenue). However, from the text of the ad, we were able to classify job positions according to level, whether they involve customer contact, and a one-word description of the job.14 We also noted criteria specified in the advertisements regarding age, gender, marital status, attractiveness, and skills. If an advertisement specified women only, we sent only women’s resumes. If it specified language or programming skills, we added those qualifications to resumes responding to that ad.

In order not to raise suspicions about the experiment, we scheduled the deliveries of emailed applications at different times over 2 consecutive days using the Gmail Boomerang service.15 Each applicant name was associated with a cell phone number and an email account. In most cases we sent eight resumes per job advertisement, four men and four women. Within each gender, we randomized universities, marital status, and a photograph representing one of the three phenotypes or a control without photograph. If the employer contacted the applicant to schedule an interview, the callback was registered.16

Our first research question asks whether explicitly discriminatory employers have higher callback rates. We argue that narrower searches may lead to better matches than broad searches. To answer this question, we analyze the data at the job ad level. If there are multiple ads for the same firm, we drop the second ad from the sample. First, we create an indicator variable equal to one if there is a response to any of our applicants from a particular ad (and equal to zero if there was no response at all), denoted by AnyCallbackj. Using this as our dependent variable, we estimate the following linear probability model:
$$ \Pr \left({\mathrm{AnyCallback}}_j=1\right)=\alpha +\beta {D}_j+\delta {X}_j+{\varepsilon}_j, $$
(1)
where j denotes the ad/employer, Dj is an indicator equal to 1 if the ad explicitly discriminates, and Xj are ad-level control variables, such as academic major, level of experience, and type of job position as described by the ad. We also test whether different types of explicit discrimination elicit different callback responses. To this effect, we estimate equations similar to (1), but using as the explanatory variable of interest indicator variables that show whether the ad was gender-targeted (D1j), requested a photograph (D2j), specified a certain physical appearance (D3j), or requested two or more of these criteria simultaneously (D4j). Hence, \( {D}_j=\sum \limits_{k=1}^4{D}_{kj} \). We use a mutually exclusive grouping to explore whether there is heterogeneity across the different types of discriminatory ads.17 The coefficient of interest is β. Our hypothesis is that explicitly discriminating employers make more callbacks because the pool of applicants is better suited to their preferred profile. We thus expect that β > 0.18
Our second question is whether explicitly discriminatory employers discriminate further in subsequent stages of the hiring process based on other applicant characteristics. In this case, we analyze the data at the individual level. The dependent variable is an indicator variable equal to one if the applicant received a call from the employer, denoted by Callbackij. Our estimating equation is:
$$ \Pr \left({\mathrm{Callback}}_{ij}=1\right)=\alpha +\beta {D}_j+\gamma {M}_{ij}+\boldsymbol{\rho} {\boldsymbol{R}}_{ij}+\pi {M}_{ij}\times {D}_j+\boldsymbol{\theta} {\boldsymbol{R}}_{ij}\times {D}_j+\delta {X}_{ij}+{\varepsilon}_{ij} $$
(2)
where i denotes the applicant and j the job ad, Mij is an indicator for married applicants, Rij is a vector of indicators of phenotype (white, mestizo, and no photograph), Dj is an indicator of explicitly discriminating ads, and Xij are applicant or ad covariates, such as a public college indicator, type of academic major, a public high school indicator, command of English and other languages, time availability, leadership courses, and type of job position.19

Our working hypothesis for this second question is that employers who begin the hiring process with a discriminatory advertisement are more likely to continue discriminating at other stages in the hiring process. The coefficients of interest are thus π and θ. For instance, we would expect married women and indigenous applicants to be called back less by employers who discriminate. The possible mechanisms behind this continued discrimination in the hiring process are varied. First, perhaps the strongest signal of a discriminating employer is to post a discriminating job ad. It would thus be no surprise to see such an employer discriminate throughout the hiring process. Second, the hiring process is an extended one. Every stage of the hiring process presents an additional opportunity to discriminate. Hiring decisions may thus fall into what Mazar and Ariely (2010) term “sequential influences on dishonest behavior,” a phenomenon known as the “what-the-hell” effect in the dieting literature. According to this theory, individuals increasingly violate rules or norms when they are presented with sequential decisions in which they can make poor choices. Third, mental models are persistent and difficult to change (Hoff and Walsh 2017). Hence, employers may engage in explicit discrimination just because other employers in the past and present have behaved in the same way. It is unclear to us whether the discriminatory job ad is the trigger or the final consequence of such sequential decisions. Even when employers discriminate in job ads, they do not fully describe all the traits of the preferred applicant. They may reveal their preferences only partially, either because some of those preferences could be seen as unethical or because they are unconscious. Further research is necessary to disentangle this phenomenon.

Descriptive Statistics

A total of 8018 resumes were sent in response to the job ads selected.20 Table S1 in the supplementary materials shows the descriptive statistics. A little more than half of the resumes are from women. Our random assignment established 27% of the individuals as married, with an average age of 24.5 years, and 62% of them are identified as graduates of a public university. There are more job ads for business than engineering majors, which explains the large proportion of applicants with business degrees (71%). Of the whole sample, 12.7% were called back. However, there is a gender difference: 10.5% of the men were called back, but 14.9% of the women were. This difference may be the result of a greater explicit preference for women in job ads, or even a greater overall preference for women workers.

Table 1 presents descriptive statistics for explicit discrimination in the job ads, as well as other specified criteria. This table shows results like those of Kuhn and Shen (2013). It shows the degree of explicit discrimination in the job ads in our sample, but importantly, it also shows that explicit discrimination often occurs on more than one level simultaneously. Column (1) shows the proportion of job ads explicitly stating each requirement. A total of 34.5% of job ads have at least one type of discriminatory request: gender, a photograph, attractiveness, or marital status. Approximately 20% state an explicit gender preference, nearly 17% require a photograph, 5.8% request an attractive appearance, and 1.6% specify a marital status.21 The ads also make other requests that are more related to bona fide qualifications, like command of English (23.1%), and the ability to travel or to work extended hours (9.5%).
Table 1

Explicit Discrimination in Job Ads and Other Requirements

  

Gender-targeted ads

Any ad

Women

Men

Ratio[2]/[3]

p value

[1]

[2]

[3]

[4]

[5]

Panel A. Explicit discrimination

 Any discriminatory request

0.345

0.343

0.225

1.52

0.002

 Gender-targeted

0.197

0.604

0.396

1.53

0.001

 Photograph request

0.167

0.200

0.087

2.29

0.003

 Appearance-targeted

0.058

0.299

0.015

20.10

0.000

 Marital status request

0.016

0.316

0.105

3.01

0.163

Panel B. What the job requires

 Any career-related request

0.281

0.091

0.109

0.83

0.461

 English required

0.231

0.093

0.089

1.04

0.887

 Travel/move/time flexibility

0.095

0.071

0.196

0.36

0.010

 Travel

0.090

0.067

0.210

0.32

0.005

 Time flexibility

0.007

0.125

0.250

0.50

0.598

 Moving to another city

0.008

0.000

0.571

0.00

0.030

Panel C. Other job characteristics

 Higher level

0.164

0.119

0.083

1.44

0.263

 Lower level

0.189

0.225

0.086

2.63

0.000

 Customer contact

0.087

0.098

0.059

1.67

0.320

 No experience required

0.342

0.161

0.085

1.89

0.002

Notes: Authors’ estimations using the job ads dataset

Advertisements with explicit discrimination show important gender differences in the types of requests. Columns (2)–(5) in Table 1 focus on the ads that are gender-targeted. We find that among these ads, 60.4% are targeted towards women and the rest towards men, and that the difference is statistically different from zero. Among the ads that require a photograph, 20% explicitly solicit women, but only 8.7% solicit men (the rest do not specify a gender). These discriminatory ads are thus 2.3 times more likely to be directed explicitly to women than to men. The ads with appearance or marital status requirements were generally directed to women, whereas those that required time flexibility or the ability to travel were biased towards men. There was no statistically significant difference in gender in the ads that required English language ability.

Finally, Table 1 presents some statistics for the characteristics of the jobs. In our sample, 16.4% of jobs are high level and 18.9% are low level (the remaining jobs do not state a level within the firm). In addition, 8.7% of our jobs require customer contact and 34% do not require any prior experience. Among these job characteristics, low-level jobs and jobs that require no prior experience are more commonly ads targeted to women than to men, and the difference is statistically significant.

Results

Our two working hypotheses employ different data structures in our dataset. Equation (1) is estimated using ad-level data, whereas Eq. (2) uses data at the applicant level. Table 2 presents the characteristics of our sample for each of these two data structures. Each row of panel A presents different categories of explicit discrimination: no explicit discrimination, any form of discrimination, ads directed only at a specific gender (gender only), ads that only request a photograph (photo only), ads that have only an attractiveness or appearance requirement (appearance only), and ads which specify two or more of these three criteria. Given that we sent more resumes in response to non-discriminating advertisements,22,23 non-discriminating job ads have a greater representation in the applicant (resume) sample than in the job ad sample. Likewise, more resumes were sent in response to ads requiring photographs than to ads which were gender-targeted, whereas the reverse is true in the ad-level data. The gender and photograph requirements are the most common forms of discrimination in ads: 13.8% of ads discriminate solely based on gender and 10.3% solely based on a photograph. A total of 7.3% of the ads discriminate based on two or three criteria; among this latter group, the photograph and gender criteria are also the most common.
Table 2

Samples in groups

 

Resume sample

Ad sample

Observations

8018

1175

Panel A: explicit discrimination (%)

 

 None

75.2

65.4

 Any discriminatory request

24.0

33.8

 Gender only

8.3

13.8

 Photo only

9.0

10.3

 Appearance only

2.7

2.4

 2 or more

4.1

7.3

Notes: Authors’ estimations using the resume- and ad-level data. Explicit discrimination is defined only as those ads with a required gender, photo, or appearance

In the estimations below, we exploit this variation in resume- and ad-level data to answer our research questions. In one estimation, we make a distinction only between no discrimination and any form of discrimination. We also estimate regressions in which we disaggregate by type of discrimination using “none,” “gender only,” “photo only,” “appearance only,” and “two or more” to disentangle the source of the discrimination. The comparison group is always the ads that do not explicitly discriminate based on those characteristics.

In Table 3, we present the percentage of applicants who received a callback. The sample was divided between non-discriminatory and explicitly discriminatory ads. Within these categories, we present the results for all applicants and by gender. The first finding in this table is that explicitly discriminating ads resulted in a greater proportion of callbacks (14.9%) than non-discriminatory ads (12%). This finding holds across marital status and racial categories of the applicants. However, most of this gain in callbacks comes from callbacks to women, not men.24 When we compare columns (2) and (5), we find that the percentage of explicitly discriminatory ads that resulted in a callback to at least one woman is 5 percentage points larger than that of non-discriminatory ads. This is a sizable effect. It means that women responding to non-discriminatory ads need to send 36% more resumes to obtain the same number of callbacks as women responding to explicitly discriminatory ads. In the case of men, we do not find a large difference. We now turn to our regression results to find whether these differences hold after controlling for job characteristics.
Table 3

Percentage of callbacks by category

 

Neutral

Explicitly discriminatory

All

Women

Men

All

Women

Men

[1]

[2]

[3]

[4]

[5]

[6]

All

12.0

13.6

10.5

14.9

18.7

10.5

Single

12.1

13.9

10.2

15.8

20.1

10.3

Married

12.1

12.8

11.4

12.6

14.1

11.0

White

13.3

15.5

11.1

16.7

20.8

11.6

Mestizo

12.7

13.6

11.8

15.9

21.1

9.8

Indigenous

11.4

13.1

9.7

13.3

15.3

11.0

No photo

10.8

12.2

9.5

12.8

16.2

8.7

Ad-called back

28.2

23.3

19.8

28.2

29.4

17.5

Notes: Authors’ estimations using applicant-level and ad-level data. The last row presents the percentage of ads that called back within each category using ad-level data

Do Explicitly Discriminatory Employers Have Higher Callback Rates?

In Table 4 we present the estimates of Eq. (1) using a linear probability model at the ad level. Column (1) presents the results for the complete sample, and columns (2) and (3) separate women and men, respectively. First, we present the results without controls and find that overall, explicitly discriminatory employers do not call back more often than non-discriminatory ones. However, when we split the sample by gender, meaning at least one woman or one man responding to the ad receives a callback, we confirm that employers with discriminatory ads have a 6.1 percentage point (pp) greater probability of calling women back than those with non-discriminatory ads. We do not find any statistically significant effect for men. Rows B to E introduce controls. We first control for the academic major and the experience level to which the ad is directed. Controlling for these variables, we find an 8.6 pp. difference in the probability of a woman receiving a callback between discriminatory and non-discriminatory ads. This effect is substantial: a woman responding to an explicitly discriminatory ad has a 37% greater likelihood of receiving a callback than one responding to a neutral ad (the base rate of callbacks for neutral ads is 23.3%). We find no evidence of differences among men. In row C of Table 4, we control for other hiring criteria, such as English, ability to travel or move to another city, and time flexibility, and find still a greater likelihood of women receiving callbacks from discriminatory ads. Finally, rows C and E further control for job characteristics or type of job, and even though the magnitude of the effect decreases to 7.5 and 7.6 pp, the results are largely unchanged and still represent a more than 30% greater chance of being called back.25,26
Table 4

Probability of a callback

 

All

Women

Men

(1)

(2)

(3)

A. No control variables

 Any type of discrimination

0.001

0.061**

− 0.024

(0.028)

(0.030)

(0.028)

B. Controls: college major and level of experience

 Any type of discrimination

0.030

0.086***

0.000

(0.028)

(0.030)

(0.027)

C: Controls: + English, travel, and moving requirements

 Any type of discrimination

0.031

0.086***

0.001

(0.028)

(0.030)

(0.027)

D: Controls + job level and customer contact

 Any type of discrimination

0.022

0.075**

− 0.003

(0.029)

(0.031)

(0.028)

E. Controls: C + one-word job description

 Any type of discrimination

0.020

0.076**

− 0.013

(0.029)

(0.032)

(0.029)

Any callbacks from neutral firms

28.2%

23.3%

19.8%

Observations

1147

1062

1012

Notes: Estimation by the authors using a linear probability model. Standard errors in brackets are robust

***Significance at the 1% level

**Significance at the 5% level

*Significance at the 10% level

We now disaggregate the discriminatory ads by the type of criteria specified: gender only, photograph only, appearance only, and two or more criteria. Table 5 presents the estimations where we include dummies for each of these categories, with non-discriminatory ads as the baseline category. All the estimations include control variables for academic major, experience, other skills, job level, and customer contact.27 Column (1) shows the results for the entire sample of ads. Although the effects are larger than those in Table 4, none of the coefficients are statistically significant. However, when we separate the sample by gender, we find that ads which target women are 12.5 pp more likely to result in callbacks to women than neutral ads (first row of column 2). In contrast, ads which target men do not seem to have a higher probability of callbacks to men than neutral ads. We find no statistically significant effect on callback from other types of explicit discrimination, either for women or men.
Table 5

Callbacks at the ad level

 

All

Women

Men

[1]

[2]

[3]

Gender only

0.006

0.125**

− 0.032

(0.039)

(0.052)

(0.045)

Photo only

0.065

0.073

0.014

(0.047)

(0.045)

(0.039)

Appearance only

− 0.045

− 0.077

0.057

(0.081)

(0.068)

(0.081)

2 or more requests

0.009

0.071

− 0.041

(0.051)

(0.058)

(0.052)

Any callbacks from neutral firms

28.2%

23.3%

19.8%

Observations

1147

1062

1012

Notes: Estimation by the authors using a linear probability model. Standard errors in brackets are robust. Control variables include dummies for college major, whether the ad requires experience, additional job requests, job position hierarchy, and whether the job requires customer contact (controls in Table 4, model D)

***Significance at the 1% level

**Significance at the 5% level

*Significance at the 10% level

Consistent with lower matching costs, the evidence supports our first hypothesis that employers placing discriminatory ads have a higher probability of calling back an applicant. However, this seems to be the case only for women, and not for men. Further research is necessary to understand the cause of this difference in employer behavior.

Do Explicitly Discriminatory Employers Discriminate Against Applicants Based on Gender, Marital Status, or Race?

Our second hypothesis looks at differences in discrimination based on gender, marital status, and race between explicitly discriminatory and neutral employers. We estimate regression Eq. (2) at the individual level. In these regressions, we control for age, a business-degree dummy, a scholarship dummy, public college/high-school dummies, foreign language dummies, a leadership dummy, job position hierarchy, and whether the job requires customer contact. Table 6 presents the regression results. Column (1) shows the estimations using the full sample of female and male applicants. There are more callbacks for women than men (by 3.3 pp) and for white applicants than those with indigenous-looking phenotypes (by 2 pp). Explicitly discriminatory ads result in even more callbacks to women (an increase of 4.6 pp), but do not seem to result in discrimination based on other criteria.
Table 6

Callbacks at the applicant level

 

All

Women

Men

[1]

[2]

[3]

Women

0.033***

  

(0.008)

  

Married

− 0.003

− 0.015

0.009

(0.010)

(0.016)

(0.014)

White

0.021**

0.019

0.015

(0.008)

(0.013)

(0.012)

Mestizo

0.013

0.001

0.021*

(0.008)

(0.012)

(0.012)

No photo

− 0.004

− 0.012

0.002

(0.008)

(0.012)

(0.011)

Any discrimination

− 0.001

0.024

0.017

(0.021)

(0.026)

(0.024)

Any discrimination x Women

0.046**

  

(0.022)

  

Any discrimination x Married

− 0.029

− 0.050

− 0.008

(0.023)

(0.033)

(0.030)

Any discrimination x White

0.014

0.040

− 0.005

(0.016)

(0.025)

(0.024)

Any discrimination x Mestizo

0.012

0.056**

− 0.029

(0.016)

(0.023)

(0.022)

Any discrimination x No photo

− 0.003

0.018

− 0.026

(0.024)

(0.032)

(0.028)

Observations

7952

4069

3883

Notes: Estimation by the authors using a linear probability model. Standard errors in brackets are robust and clustered at the firm level. All regressions also control for age, business dummy, scholarship dummy, public college/high school dummy, dummies for foreign language, leadership dummy, job level, and customer contact (controls in Table 4, model D)

***Significance at the 1% level

**Significance at the 5% level

*Significance at the 10% level

The results are slightly different when we split the sample between men and women. Although the coefficients of the married dummy are statistically insignificant, the magnitude for women of the interaction of the married and any-discrimination dummies is economically significant. The magnitude of this coefficient for men is negligible.

Now, we turn to the coefficients of the phenotype dummies and their interactions with the any-discrimination dummy in the case of women, shown in column (2) of Table 6. All of the interactions with phenotypes are positive, but only the interaction with mestizo is statistically significant. We find that mestizo women are 5.6 pp more likely to receive a call from employers with explicitly discriminatory ads than from those with neutral ads. In the case of men, shown in column (3), we find no such effect from discriminatory ads. These coefficients are smaller in magnitude in most cases and have opposite signs, but none of them is statistically significant.

Given that different types of explicit discrimination may elicit (or be an expression of) different employer behaviors, we also estimate the effects for each type of discrimination: gender only, photograph only, appearance only, and two or more criteria. For ease of presentation, we show only graphical results, but the full estimations are presented in Table S2 of the supplementary material.28 As Eq. (2) shows, instead of including a dummy variable for explicitly discriminatory ads, we include the four mutually exclusive dummy variables \( {D}_j=\sum \limits_{k=1}^4{D}_{kj} \) representing each type of discrimination. A statistically significant result for any interaction means that the callback rate for that type of discrimination is different than the rate for the neutral ads. Figure 1 presents the coefficient for women and its interactions with the different types of discriminatory ads using all individuals in the sample. In this experiment, women have a higher probability of being called back than men, but women receive calls for gender-targeted ads much more than for neutral ones (the probability is 8.7 pp greater). The figure also shows that women have a lower probability of being called for ads specifying appearance (7 pp less) than for neutral ads. The estimates for ads that request only a photograph and those which include two or more criteria are noisy but positive. Overall, these results confirm the findings in Tables 3 and 6, which show that overall, women receive more calls from explicitly discriminatory ads than from neutral ones, and they demonstrate that the effect is driven mostly by gender-targeted ads.
Fig. 1

Do explicitly discriminating firms discriminate more based on gender?

Figure 2 presents the coefficient for married status and its interaction with the types of discrimination for men and women. First, non-discriminatory ads do not differentiate between married and single applicants, and they treat men and women equally in this respect. The ads that specify only a photograph or only appearance do not appear to result in discrimination based on marital status either for men or women. However, we find that married women who respond to ads targeted specifically to women have a probability of receiving an interview call that is 18.3 pp lower than that for married women who respond to neutral ads.29 We do not find an analogous effect for men. Finally, we find that married men who responded to ads with two or more criteria have a probability of receiving a callback that is 8.8 pp lower than that for married men who respond to non-discriminatory ads. This is the first time we see such a result for men based on marital status.
Fig. 2

Do explicitly discriminating firms discriminate more based on marital status?

Figure 3 presents the coefficients for race and their interactions with the different types of ads specifying women. We find that only in the case of ads with two or more criteria are white and mestizo women more likely to be called back than indigenous-looking women. There are 218 such ads in our database, of which 74% are targeted specifically to women, 78% require a photograph, and 59% require “good appearance.” Some of the jobs in this category do not appear to involve customer contact (which may be the case for accountants, IT analysts, or product developers). Based on our analysis, employers using other types of discriminatory ads do not seem to discriminate based on race. Finally, Fig. 4 presents the race coefficients for men. None of the interaction coefficients are statistically different from zero: discriminatory ads do not seem to reflect different employer behavior than neutral ads based on race.
Fig. 3

Do explicitly discriminating firms discriminate more based on race?—women

Fig. 4

Do explicitly discriminating firms discriminate more based on race?—Men

Discussion and Conclusions

Our correspondence study of discrimination in the Mexican labor market found that some job ads explicitly exclude entire segments of the population by expressing their preferences for gender, age group, attractiveness, or marital status. About 20% of the job ads in our sample specified either male (40%) or female (60%) applicants; 5.8% specified appearance with a phrase such as “good presentation,” and 16.7% requested a resume with photograph. We tested two hypotheses: first, whether explicitly discriminatory ads result in calls for interviews more often than non-discriminatory ads, and second, whether explicit discrimination in advertisements is reflected in discrimination in callbacks. Kuhn and Shen (2013) argue that explicit discrimination will not necessarily be reflected in a correspondence study. However, we found that the two forms of discrimination coexist in Mexico.

We found evidence for our first hypothesis: employers with explicitly discriminatory ads do indeed have a higher probability of calling back than those with non-discriminatory ads. However, this effect is only for women: the probability that discriminatory employers will call a female candidate for an interview is close to 8 pp greater than that for non-discriminatory employers. Regarding our second hypothesis, we found that in general, explicit discrimination in ads does not imply greater discrimination in callbacks than that of neutral ads. However, there are some exceptions to this general observation. First, women responding to discriminatory ads have a higher probability of receiving callbacks (4.9 pp) than those responding to neutral ads, and this effect is driven mostly by gender-targeted ads (where the probability is 8.6 pp greater than that for neutral ads). Also, married women tend to experience more discrimination in response to ads directed specifically at women (the probability of a callback goes down 18 pp) than non-targeted ads. Married men responding to ads with two or more criteria are slightly less likely to receive callbacks (a drop in the probability of close to 9 pp) than those responding to neutral ads. Finally, white and mestizo women responding to ads with two or more requests are more likely to receive callbacks (probability increases of approximately 9 pp and 10 pp, respectively) than those responding to neutral ads. Hence, although the overall callback rate is greater in responses to discriminatory ads, the discrimination in callbacks is exacerbated for some groups, producing a type of double discrimination in the hiring process.

Our experiment was not originally designed to provide evidence for this problem and our point estimates of the effect of discriminatory job ads thus tend to be imprecise. We suggest that future research conduct this type of experiment on a larger scale to increase the power of the estimations and to provide external validity (using different occupations, age groups, and cities).30 It would also be interesting to know how employers respond to applicants who do not fit the criteria specified. Research on this last point would allow us to evaluate whether employers may be missing valuable candidates because of the discrimination in their ads.

It is hard to pin down the welfare effects of explicitly discriminatory ads. Our evidence shows that these ads have higher callback rates, at least for women, and may therefore reduce the search costs for employers and applicants. However, we find that in some cases employers end up discriminating more in the calls for interviews, pointing to the existence of double discrimination in the Mexico City labor market. This double discrimination is directed at married women and women with an indigenous phenotype. If the “what-the-hell effect” is operating and discriminatory ads trigger more discrimination, then eliminating those ads may eliminate some of the discrimination in callbacks. In contrast, if discriminatory ads are just a consequence of discriminatory behavior or stereotypes, the elimination of the ads will not necessarily reduce discrimination. There is also no research on whether the mere existence of discriminatory ads perpetuates negative stereotypes or perverse social norms in the labor market (Hoff and Walsh 2017). If such ads contribute in any way to a culture that tolerates discrimination, then their elimination may increase welfare in the long run. Further research is necessary to uncover the mechanisms behind double discrimination towards women.

Footnotes

  1. 1.

    See the literature reviews in Pager (2007) and Pager and Shepherd (2008).

  2. 2.

    Bertrand and Duflo (2016) and Neumark (2016) review the literature on field experiments of discrimination. Zschirnt and Ruedin (2016) performed a meta-analysis of field experiments in OECD countries and concluded that minority applicants need to send about 50% more resumes in order to obtain interviews.

  3. 3.

    Although one can find examples of explicit racial discrimination in job ads, this type of discrimination was not found in the subsample of ads we analyzed.

  4. 4.

    Aguilar (2011) finds that people are more inclined to vote for white than mestizo political candidates in Mexico. The word indio (Indian) is considered an insult in Mexico; see Oehmichen (2006) and Wade (2008). Mexican studies of racial discrimination include Béjar Navarro (1969) and Gall (2004), but neither of these examines the demand side of the labor market.

  5. 5.

    Ideals of attractiveness in Mexico are correlated with whiteness. There is literature in the USA linking skin tone to attractiveness among African-Americans (Hill 2002). In fact, white skin has become such an attractiveness ideal around the world that sociologists now speak of the “bleaching syndrome.” Many women with darker skin lighten their skin color using beauty products (Hunter 2008) sold by a substantial industry in Mexico. One study in Mexico City found that students with darker skin perceive themselves as being less attractive (Ortiz-Hernández et al. 2011).

  6. 6.

    Those results were not in line either with a preference-based discrimination model (Becker 1971) or a statistical discrimination model (Phelps 1972); our explanations thus appear to be ad hoc descriptions of particular coefficients that fall short of a comprehensive explanation of our findings. For instance, the fact that married women are discriminated against and that married men are not points to a statistical discrimination hypothesis. But when we examine the interaction of marriage with race, we find that mestizo women do not experience a marriage penalty (in fact, they experience a marriage premium); only white and indigenous women do. On the other hand, a preference-based discrimination model cannot explain why women experience racial discrimination, but men do not.

  7. 7.

    The attractiveness requirement is implied by phrases such as “excelente (buena) presentación” or “presencia,” which mean “excellent (good) presentation” or “presence.”

  8. 8.

    Where job advertisements specified the gender desired for the vacant position (only men or only women), asked for a resume with photograph, or a combination of both gender-targeting and photograph requests, we sent fewer than 8 resumes: 4 resumes in the first example, 6 in the second, and 3 in the third.

  9. 9.

    The following names were used: Alejandro Flores Álvarez, Antonio González Lara, Carlos Romero Gómez, Javier Rodríguez Mendoza, Claudia García Ramírez, Gabriela López Acosta, Mariana Hernández Silva, and Mónica Vázquez Rivera. According to the Instituto Federal Electoral (2012), Mateos (2010), and the Baby Center website (2011), these names are very common in Mexico.

  10. 10.

    The factors randomized were photographs, high schools, universities, professional experience, marital status, addresses, hobbies, and additional information.

  11. 11.

    Women wore a black blazer and a soft-toned blouse; men wore a dark suit, white shirt, and a discreetly-patterned tie. We used color photographs so that the physical characteristics would be better observed by the employers.

  12. 12.

    The photographs are shown in the supplementary materials.

  13. 13.

    The websites were OCC Mundial (http://www.occ.com.mx/) and CompuTrabajo (http://www.computrabajo.com.mx/). Future studies using different cities and occupations are warranted to confirm whether these results are externally valid.

  14. 14.

    The highest-level jobs are those where the ad asks for a director, manager or someone who has people under her charge. The lowest-level are job posts involving trainees or assistants.

  15. 15.
  16. 16.

    In order to avoid further interaction with prospective employers, our research assistants at this point thanked them for their interest and informed them that they had accepted another job.

  17. 17.

    We attempted to include a group that explicitly requested single marital status, but the proportion of ads specifying this characteristic was very small. Including or excluding these requests does not affect the estimates we show below.

  18. 18.

    There are multiple demand- and supply-side factors that determine whether the employer calls back or not, such as the amount of applications received, or whether the firm is experiencing an expansion or an increase in demand, among many others. We cannot control for these factors, but we expect that these would affect male and female applicants in the same way. The sector of employment is another issue, since there are sectors that are traditionally female or male. We try to control for these differences by adding a one-word-description of the job to our control variables. Our results are robust to these one-word descriptions. We are thus confident that the sector of employment is not biasing our results.

  19. 19.

    We also interacted the D variables with the Xs, but they were statistically insignificant.

  20. 20.

    We sent the resumes between October 2011 and May 2012. We received callbacks within a week of sending the resumes; hence, the callback period is roughly the same. We restrict the sample to one ad per firm.

  21. 21.

    Given the small fraction of ads requiring a certain marital status, we do not include this type of discrimination in our analysis. The exclusion of these ads does not change our results.

  22. 22.

    Recall that if, for instance, the ad was gender-targeted, we sent only resumes for the requested gender in response to that ad. If the ad required a photograph in the resume, we sent only resumes with a photograph (with some accidental exceptions).

  23. 23.

    An anonymous reviewer pointed out that our results may be due to this supply-side effect: discriminatory ads may receive fewer resumes and each would thus have a greater chance of being called back. We do not have any data to show that discriminatory ads indeed receive fewer resumes. However, Kuhn and Shen (2013) show that discriminatory ads are more common in looser labor markets such that they have a larger overall pool of applicants (that is, a large supply). It is unclear to us whether gender-targeting (or any other ad restriction or request) would reduce the number of applicants as compared with untargeted ads. Our argument is that the quality of the match increases in the bounded pool of applicants, but we do not have data to control for the size of the pool.

  24. 24.

    If the increase in the callback rates were purely a supply-side effect of gender-targeted ads (see the previous footnote), we would expect a similar increase in callbacks from ads targeted to men or women than from those that are not targeted. Our results indicate that this is not the case, and so there may be other mechanisms at play.

  25. 25.

    All our results are unchanged when a probit model is used, or when ads that specified a preferred marital status are excluded (not shown, but available from the authors upon request).

  26. 26.

    We will use specification D in Table 4 as our preferred specification in the rest of the paper, given that the controls of the type of job position are easier to interpret that the one-word description. However, all the following results remain unchanged when we control for one-word job descriptions.

  27. 27.

    The results are robust to the addition of English, travel/moving availability, and time flexibility controls (not shown, but available upon request).

  28. 28.

    The first three columns of this table include all the data; columns (4)–(6) exclude the ads that requested marital status. As can be seen, the results are robust to this sample exclusion. The regression includes the variables of marital status and three phenotypes. As there are four groups of mutually exclusive discriminatory ads, there is a total of 16 coefficients in the men’s or women’s regressions (columns [2], [3], [5] and [6]). If all individuals are included there is a total of 20 coefficients (columns [1] and [4]).

  29. 29.

    The callback rate for single women responding to ads targeted to women is 27.14%, whereas that for married women responding to the same ads is 6.9%. The probability of a married woman receiving a callback from an ad targeted to women is slim.

  30. 30.

    Our results are also only valid for internet job searches and may not be generalizable to other search channels. However, it is much more difficult, if not impossible, to implement a correspondence study using other job search methods.

Notes

Acknowledgment

We would like to thank Brisna Beltrán, Sarait Cárdenas, Tania Fernández, Andrés Hincapié, Lia Jazibi, Javier Parada, Carolina Rivas, Mauricio Sandoval, Sebastián Sandoval, and Luis Téllez for their excellent research assistance at various stages of this project. We are also grateful for comments from Nelly Aguilera, Lori Beaman, David Card, Gonzalo Castañeda, Carlos Chiapa, Peter Kuhn, Craig McIntosh, Enrico Moretti, Isidro Soloaga, Julio Vallejo, and seminar participants at El Colegio de México, the Centro de Investigación y Docencia Económicas, and the University of California at Berkeley. We thank the anonymous reviewers for their careful reading of our manuscript and their many insightful comments and suggestions. Any errors are our own.

Funding Information

This research was funded in part by Mrs. Cristina Sada Salinas, in memory of Joanna de la Cruz Sada.

Compliance with Ethical Standards

Conflict of Interest

On behalf of all authors, the corresponding author states that there are no conflicts of interest.

Supplementary material

41996_2019_31_MOESM1_ESM.docx (139 kb)
ESM 1 (DOCX 138 kb)

References

  1. Aguilar, Rosario. Social and political consequences of stereotypes related to racial phenotypes in Mexico. Documento de Trabajo, No. 230, CIDE.2011.Google Scholar
  2. Anand R. Gender stereotyping in Indian recruitment advertisements: a content analysis. International Journal of Business, Governance and Ethics. 2013;8(4):306–22.CrossRefGoogle Scholar
  3. ANUIES. Anuario estadístico: población escolar y personal docente en la educación media superior y superior: ciclo escolar 2008–2009. Asociación Nacional de Universidades e Instituciones de Educación Superior, México. 2009.Google Scholar
  4. Arceo-Gómez EO, Campos-Vázquez RM. Labor supply of married women in Mexico: 1990-2000. 2010. El Colegio de México, Centro de Estudios Económicos, Documento de Trabajo Núm XVI-2010.Google Scholar
  5. Arceo-Gómez E, Campos-Vázquez R. Race and marriage in the labor market: a discrimination correspondence study in a developing country. Am Econ Rev. 2014;104(5):376–80.CrossRefGoogle Scholar
  6. Baby Center. Los nombres más frecuentes en México desde 1930. 2011. http://espanol.babycenter.com/pregnancy/nombres/frecuentes_1930/ (accessed March 23, 2012).
  7. Becker GS. The economics of discrimination. Chicago: The University of Chicago Press; 1971.CrossRefGoogle Scholar
  8. Béjar Navarro R. Prejuicio y discriminación racial en México. Rev Mex Sociol. 1969;31(2):417–33.CrossRefGoogle Scholar
  9. Bertrand M, Duflo E. Field experiments on discrimination. In: Banerjee A, Duflo E, editors. Handbook of Economic Field Experiments, Vol. 1. Amsterdam: Elsevier; 2017, p. 309–393.Google Scholar
  10. Bertrand M, Mullainathan S. Are Emily and Greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination. Am Econ Rev. 2004;94(4):991–1013.CrossRefGoogle Scholar
  11. Chowdhury AR, Areias AC, Imaizumi S, Nomura S, Yamauchi F. Reflections of employer’s gender preferences in job ads in India: an analysis of online job portal data. Policy Research Working Paper No. 8379, The World Bank. 2018.Google Scholar
  12. Darity WAJ, Mason PL. Evidence on discrimination in employment: codes of color, codes of gender. J Econ Perspect. 1998;12(2):63–90.CrossRefGoogle Scholar
  13. Delgado M, Kuhn P, Shen K. Age and gender profiling in the Chinese and Mexican job markets: evidence from four job boards. NBER Working Paper, No 22187. 2016.Google Scholar
  14. Gall O. Identidad, exclusión y racismo: Reflexiones teóricas y sobre México. Rev Mex Sociol. 2004;66(2):221–59.CrossRefGoogle Scholar
  15. Hill ME. Skin color and the perception of attractiveness among African Americans: does gender make a difference? Soc Psychol Q. 2002;65(1):77–91.CrossRefGoogle Scholar
  16. Hoff K, Walsh J. The whys of social exclusion: Insights from behavioral economics. Policy research working paper 8267, World Bank. 2017.Google Scholar
  17. Hunter M. The cost of color: what we pay for being black or brown. In: Hall RE, editor. Racism in the 21st century: an empirical analysis of skin color. New York: Springer; 2008.Google Scholar
  18. Instituto Federal Electoral. Estadísticas lista nominal y padrón electoral. 2012. http://www.ife.org.mx/portal/site/ifev2/Estadisticas_Lista_Nominal_y_Padron_Electoral/ (accessed March 23, 2012).
  19. Kuhn P, Shen K. Gender discrimination in job ads: evidence from China. Q J Econ. 2013;128(1):287–336.CrossRefGoogle Scholar
  20. Lahey JN, Beasley RA. Computerizing audit studies. J Econ Behav Organ. 2009;70(3):508–14.CrossRefGoogle Scholar
  21. Lawler JJ, Bae J. Overt employment discrimination by multinational firms: cultural and economic influences in a developing country. Ind Relat. 1998;37(2):126–52.CrossRefGoogle Scholar
  22. Mateos P. El análisis geodemográfico de apellidos en México. Papeles de Población. 2010;16(65):73–103.Google Scholar
  23. Mazar N, Ariely D. Sequential influences on dishonest behavior. In: Campbell MC, Inman J, Pieters R, editors. NA – advances in consumer research. Duluth: Association of Consumer Research; 2010. p. 143–5.Google Scholar
  24. Neumark D. Experimental research on labor market discrimination. NBER working paper 22022. 2016.Google Scholar
  25. Oehmichen C. Violencia interétnica y racismo en la Ciudad de México. Anales de Antropología. 2006;40(1):167–91.Google Scholar
  26. Oreopoulos P. Why do skilled immigrants struggle in the labor market? A field experiment with thirteen thousand resumes. Am Econ J Econ Pol. 2011;3(4):148–71.CrossRefGoogle Scholar
  27. Ortiz-Hernández L, Compeán-Dardón MS, Verde-Flota E, Flores-Martínez MN. Racism and mental health among university students in Mexico City. Salud Publica Mex. 2011;53(2):125–33.CrossRefGoogle Scholar
  28. Pager D. The use of field experiments for studies of employment discrimination: contributions, critiques, and directions for the future. Ann Am Acad Pol Soc Sci. 2007;609(1):104–33.CrossRefGoogle Scholar
  29. Pager D, Shepherd H. The sociology of discrimination: racial discrimination in employment, housing, credit, and consumer markets. Annu Rev Sociol. 2008;34:181–209.CrossRefGoogle Scholar
  30. Phelps ES. The statistical theory of racism and sexism. Am Econ Rev. 1972;62(4):659–61.Google Scholar
  31. Wade P. Race in Latin America. In: Poole D, editor. A companion to Latin American anthropology. Oxford: Blackwell Publishing Ltd; 2008.Google Scholar
  32. Zschirnt E, Ruedin D. Ethnic discrimination in hiring decisions: a meta-analysis of correspondence tests 1990-2015. J Ethn Migr Stud. 2016;42(7):1115–34.CrossRefGoogle Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  1. 1.Department of EconomicsCentro de Investigación y Docencia EconómicasMéxico D. F.Mexico
  2. 2.Department of EconomicsEl Colegio de MéxicoMéxico D. F.Mexico

Personalised recommendations