Вы находитесь на странице: 1из 45

Gender and Hiring patterns:

A Field Experiment on Gender Bias in the Swedish Labour


Market

Anna Pedersen & Adam Wrede

Master’s Thesis in Economics


Supervisor: Ali Ahmed
Examiner: Dominique Anxo
Spring term 2011
ACKNOWLEDGEMENTS

We would like to give our gratitude towards our supervisor Ali Ahmed for his invaluable
advices and opinions. We would also like to thank our co-supervisor Lina Andersson for her
advice in the statistical analysis. This master’s thesis was a part of the research project
“Discrimination against Homosexuals in the Recruitment Process” which was generously
funded by The Institute for Labour Market Policy Evaluation (IFAU).

2
TABLE OF CONTENTS

1. INTRODUCTION ........................................................................................... 5
2. LITERATURE REVIEW............................................................................... 7
2.1. STUDIES USING CORRESPONDENCE TESTING ................................................................................................. 8
2.2. STUDIES USING SITUATION TESTING ........................................................................................................... 12
2.3. SOME OTHER RELATED STUDIES ................................................................................................................. 16

3. THEORETICAL FRAMEWORK AND HYPOTHESIS ......................... 18


3.1. ECONOMIC DEFINITION OF DISCRIMINATION............................................................................................... 18
3.2. TASTE-BASED DISCRIMINATION.................................................................................................................. 18
3.3. STATISTICAL DISCRIMINATION ................................................................................................................... 21

4. EXPERIMENTAL METHODOLOGY ...................................................... 24


4.1. RANDOMIZED CORRESPONDENCE TESTING ................................................................................................. 24
4.2. IDENTITIES OF THE FICTITIOUS APPLICANTS ............................................................................................... 25
4.3. CHOICE OF OCCUPATIONS ........................................................................................................................... 25
4.4. APPLICATION PROCEDURE .......................................................................................................................... 26
4.5. RECORDED INFORMATION .......................................................................................................................... 26

5. RESULTS....................................................................................................... 28
5.1. DESCRIPTIVE STATISTICS............................................................................................................................ 28
5.2. REGRESSION ANALYSIS .............................................................................................................................. 31

6. DISCUSSION................................................................................................. 35
REFERENCES .................................................................................................. 38
APPENDIX ........................................................................................................ 40
A1. RESUME AND COVER LETTER FOR AN HIGH EDUCATED APPLICANT............................................................. 40
A2. RESUME AND COVER LETTER FOR AN LOW EDUCATED APPLICANT ............................................................. 42
A3. PROBIT ESTIMATES..................................................................................................................................... 44
A4. OCCUPATIONAL STRUCTURE ...................................................................................................................... 45

3
ABSTRACT

The object of this study was to examine the extent of gender discrimination in the Swedish
labour market by using a randomized correspondence testing procedure. To gather the data we
utilized a field experiment where fictitious job applications where conducted and sent to real
employers. The applications were carefully matched and differed only in gender which was
signaled by a traditional Swedish male name or female name. The responses were then
analyzed and were the basis in our tests for discrimination. We found only minor evidence of
discrimination. Statistically significant discrimination was only found against males in the
restaurant business and discrimination against females in full time positions.

Keywords: discrimination, gender, labour market, Sweden, experimental economics

4
1. Introduction
Gender inequalities have historically been large where the female and the male have been
placed in different roles and identities. In modern societies a paradigm shift has taken place
where traditional male and female stereotypes has faded out and promoted a more equal
society. As the females participation on the labour market has steady increased, the gender
earnings gap has narrowed and more females have entered leading positions. Sweden has
come far on this area with one of the highest female employment rates and also one of the
highest levels of gender equality in the world1. Laws and regulations and extended public
support might have contributed to this convergence between genders2.

Nevertheless considerable inequality remains between females and males. In some cases these
inequalities can be explained by factors such as: labor supply, the labor division of household
work or childcare, but it could also be due to discrimination. It is important to map the extent
of discrimination, as Blau and Ehrenberg (1997) puts it “social welfare is maximized if we
obtain the greatest possible productivity out of all our resources”. There are several possible
explanations why gender discrimination can prevail on the labour market where the
dominating explanations originates from Beckers (1957) theory of taste and Phelps (1972)
theory of incomplete information.

Extensive research has been done on gender inequalities and discrimination on the Swedish
labour market. Previous studies suggest that gender discrimination still occurring in Sweden.
Studies have for instance shown a remarkable preponderance of male supervisors and
managers in Sweden; see Bygren and Gähler (2007). There is also a great deal of segregation
where males tend to get hired in male dominated firms and in the private sectors while
females on the other hand in female dominated firms and in the public sector; see Hultin and
Szulkin (1999). Researchers have also found an unexplained gender wage gap where females
earn less than males. All these differences can in some extent be referred to discrimination;
see Meyersson Milgrom et al (2001).

1
In 2010 61.6 % of women 15-74 years old were employed compared to males 67.8 % (Labour Force Survey,
Statistics Sweden 2010)
2
The anti-discrimination act entered into force 2009, replacing the Equal Opportunities Act and six other anti-
discrimination laws.

5
Most of the previous research has focused on gender inequalities and discrimination on the
job and at wage bargaining level. However, discrimination can also emerge at the application
process and at the interview stage. Little attention has been paid to whether discrimination
occurs already in the hiring decision, especially not in Sweden. Internationally quite many
studies have been done on gender discrimination at the hiring level and the results are rather
incoherent, where discrimination has been found both against males and females, see for
instance Riach and Rich (1995, 2006), Weichselbaumer (2000). In Sweden on the other hand
there has only been one study by Carlsson (forthcoming) on this area, and he found some
evidence of discrimination. Due to the lack of research the aim of this study is to investigate
the existence of gender discrimination in the first step of the hiring process on the Swedish
labour market. Our main research question is: Do female workers have less probability of
receiving invitations to job interviews and thereby job offers?

We conducted a field experiment which is a convenient way to reveal and quantify


discrimination and have been widely used in previous studies, see for instance Riach & Rich
(1995, 2002). The advantage of utilizing a field experiment is contrary to survey based
studies, is that one get a clear and unbiased picture over how labour market differences due to
gender are caused by discrimination. To carry out the field experiment two identical resumes
and two cover letters was created, where each were assigned with either a female or male
sounding name. The applications were thereafter randomly distributed to real employers in 11
different occupations and the callbacks were then recorded and analyzed using a linear
probability model.

This study contributes to the existing literature on several ways. First, this is the only study
that employs a randomised correspondence test procedure, when previous studies have all
used a matched correspondence testing procedure. Second, we applied to jobs in 11 types of
occupations all with different sets of gender decomposition and educational requirements,
were most researchers have been limited to only a few occupations. Third, this is the first
nation wide field experiment in Sweden, where Carlsson (forthcoming) only studied the
metropolitan cities in Sweden.

6
2. Literature review

In this section, we review field experimental research on gender discrimination. The


following studies have all in common that they have utilized field experiments, a technique
that has been widely used during the last decades in discrimination research. When measuring
discrimination with the field experimental method, researchers make use of the real
environment where job applications or job applicants are directed to real employers.3 Field
experiments designed to measure discrimination can either be correspondence tests or
situation tests. In correspondence testing fictitious individuals send in written applications
without making visual contact to the employer. The individuals are carefully matched and
differ only in race, age, or gender. From a both methodological and practical point of view
this method is desirable as the researcher has substantial control over the experimental
treatment and control conditions. Practically, fictitious written applications ease the
application process since no visits and personal contact with the employer is necessary.
However, correspondence testing is also associated with drawbacks. In some cases visual
signs from personal contact can be needful as it might be difficult to signal some differences
in written applications. The method is also not optimal for all types of jobs, for instance when
personal applications and interviews are required. Another flaw is that it only can capture
discrimination in the first step of the hiring process despite the possibility that employers
discriminate after the interview stage.

The other common method to study discrimination is situation testing (or the audit method)
which is also used in a great deal of previous studies. In contrast to correspondence tests, real
matched individuals pretend to be job searchers and personally contact employers in situation
tests. The advantage of situation tests is that the researcher can capture strong evidence of
discrimination and catch the discriminatory employer in the act. It also generates information
about how applicants get treated by employers and thereby a delicate and high quality data
can be obtained. A drawback of this methodology is that it is time consuming since the
situation testers need careful training in their role as a job applicant. One prevalent and
serious difficulty is to match testers as there are always subtle differences between two
individuals.4

3
For further reading about the field experiment method see the excellent review by Riach and Rich (2002).
4
For a throughout discussion regarding the problems of situation tests, see Heckman (1998).

7
2.1. Studies using correspondence testing

We start by discussing studies that employ the correspondence testing method. Firth (1982)
made a survey on gender discrimination in the labor market in the U.K. The purpose was an
in-depth study of a specific job area and the focus was only on the accountancy profession
which the author categorized into five different types of jobs. Different newspapers were used
and for each job advertisement 12 different applications were sent: half of them were from
male applicants and the other half were from female applicants. The author wanted to test
how gender, color, and marital status affect applicants’ chances of getting a job. The tests
were carried out during the first half of 1978. During this period a total of 146 jobs were
applied for. To avoid detection the 12 letters to each job were posted on separate days.

If an employer called any of the applicants for an interview or asked them for further
information on an application, the author defined it as “successful.” Firth (1982) compared the
success rates for different applicants and found that female applicants always had statistically
significant lower rates than male applicants. In order to examine the impact of the other
variables and their interactions on the success rates he also conducted some multivariate
analysis. He found that gender and skin color of the applicants were clearly significant in
explaining the success rates. Specifically, females and colored people were disadvantaged in
the job market. However, if employers were females it tended to lead to higher success rates
for female applicants.

Several months after the investigation Firth (1982) sent out a questionnaire to the participating
employers to ascertain their views on employing a female. Among 146 employers, 108
answered the questionnaire. Eighty-five percent of them said that they treated males and
females equally in job-hiring decision. However, when the author matched the employers’
responses to the questionnaire against the decision they made in the experiment, it proved that
the actual discriminatory behavior was more widespread than predicted by the questionnaire
responses. This implies that employers discriminated against females despite the fact that they
say that they do not.

Riach and Rich (1995) used a field experiment to investigate the extent of sexual
discrimination on the Australian labour market. All according to the correspondence testing

8
method, two standard resumes where sent out to potential employers. Essential job
characteristics such as age, qualification, and experience were carefully matched in both
resumes. Although, style and contents such as hobbies and interests were not completely
identical in order to avoid detection. Gender was thus the only distinguishing characteristic
and it was signalled by using common Anglo-Saxon/Celtic female and male names and
surnames. To avoid that a certain resume design would influence employers, the letter type
was alternated between applicants. If both of the applicants were invited to an interview it was
treated as a non-discrimination case (or equal treatment case). If neither was invited it was
treated as a non-observation since it then was impossible to know how gender affected the
employers hiring decision. The authors selected occupation using the criteria that vacancies
had to come up at a steady rate. The occupations were computer analyst programmer,
computer operator, computer programmer, gardener, industrial relations officer, management
accountant, and payroll clerk.

Riach and Rich (1995) found that out of 991 cases (or job vacancies), 144 cases or 28.4
percent involved differential treatment. Females were discriminated against in 16.6 percent of
the cases and males in 11.8 percent of the cases. This difference is statistically significant.
Across the occupations a variation was found in the results. Statistical significant
discrimination against females was found in the occupations computer analyst programmer
and gardener with a net discrimination of 11.9 percent 15.9 percent, respectively.
Discrimination against males seemed to be present in the occupations computer programmer
and industrial relations officer, but these results were not statistical significant.

Weichselbaumer (2000) applied the same technique (correspondence testing) as the authors
above but with another angle on her research question. Weichselbaumer argued that when
testing for discrimination it is not sufficient to only control for human capital, such as formal
education and work experience, as specific personality traits might also be important for the
firms’ hiring decisions. The aim was therefore to find out if stereotypical personality traits of
males and females could explain different hiring outcomes for males and females in the
Austrian the labour market.

Weichselbaumer (2000) argued that the male stereotype is regarded as ambitious and task-
oriented, while females on the other hand are perceived as emotional and affectionate. For
instance a manager is supposed to be ambitious and dominant which are according to the

9
stereotype male traits, and therefore males might be preferred by employers for manager
positions. In female-intensive occupations, on the other hand, female-stereotype
characteristics might be more appreciated.

The procedure of the study was as following: A male, a feminine female, and a masculine
female applicant were created to apply for different jobs. All of them had identical human
capital accumulation. Applications of all three applicants where sent out to all vacancies that
accepted written applications during the period of 1998 to 1999. The masculine female had
identical personality as the male and was therefore predicted to be treated like the male. If
masculine traits are beneficial in a typical male-intensive occupation, then the female with the
masculine features should be treated like the male applicant. If not, it is a case of
discrimination. But if the feminine female get unfavourable treatment, it is necessary not
discrimination as she does not have the masculine traits that are required. Consequently, if
personality matters and there is no discrimination the applicants to the masculine occupation
should be treated according to the following; callback rate of male = callback rate of
masculine female > callback rate of feminine female. For female occupations the opposite is
true (callback rate of male = callback rate of masculine female < callback rate of feminine
female). If all applicants are treated in the same way it implies that personality traits are not
relevant. If the callback rate of the male and the masculine female deviates it means that real
discrimination exist.

In the masculine occupation, network technicians, 117 enterprises were applied for. In 73
percent of the cases the employer contacted the male applicant while the masculine female
was contacted a little bit less, 63 percent. The feminine female only got contacted in 58
percent of the time. There is an observable difference between the contact rate of the
masculine female and the feminine female, however the net discrimination of 5.13 percent
between them is not significant. The applicants were in other words treated in the following
way: callback rate of male > callback rate of masculine female = callback rate of feminine
female. This implies that discrimination did exist and that personality traits did not matter. In
the second masculine occupation, computer programmers, there were no differences in
callback rates between any of the applicants.

In the feminine dominated occupation, secretaries, severe unequal treatment was found. Both
the masculine female and the feminine female got higher callback rates than the male. In the

10
second feminine occupation, accountants, no significant differential treatment was found.
The results of Weichselbaumer (2000) illustrate that different personality traits of males and
females are of little matter when explaining different labour market outcomes of males and
females. Instead, the results indicate instead a case of taste-based discrimination.

Riach and Rich (2006) conducted a field experiment replicating their study from 1995 but
instead this one was carried out in the British labour market. The authors selected four
occupations: one female dominated occupation, one male dominated occupation, and two
mixed occupations. The following occupations were included: computer analyst programmer,
engineer, secretary, and trainee chartered accountant.

Statistically significant discrimination was found in the two mixed occupations (computer
analyst programmer, trainee chartered accountant). The male and the female got equal
treatment in 41 percent of the cases for computer analyst programmers. Discrimination
against females was in 12 percent of the cases whereas the corresponding figures for males
were 47 percent. Thus the net discrimination against males was 35 percent which was
statistical significant. The second mixed occupation, the chartered accountant, also gave a
clear indication of discrimination against males with a net discrimination of 20 percent.

In the female-dominated occupation (secretary), the males were again discriminated against
with a statistical significant net discrimination of 43 percent. However males were favoured in
the male-dominated occupation where women got 23 percent of net discrimination against
them. The authors conclude that females are favoured in female dominated occupation and
males in male dominated occupations. The authors’ findings were consistent with previous
experimental studies conducted. More surprisingly was the result that males were
discriminated against in the mixed occupations: a result that had previously not been
documented.

Another correspondence testing study by Carlsson (forthcoming) investigated the


occupational segregation between males and females on the Swedish labour market. The aim
was to study if gender discrimination is the cause of gender segregation in different
occupations. If it proved to be true that would suggest that females have more difficulties in
getting an employment in a male dominated occupation, whereas the same is true for males in
female dominated occupations.

11
The author compared the probability of being invited to an interview in male-dominated,
mixed and female-dominated jobs. If females and males have different probabilities of being
invited depending on the share of females in that occupation, one can get an understanding
how sex segregation affects the employers’ hiring decision. The author analyzed 13 different
occupations where written applications from both a female and a male applicant were sent out
by e-mail to employers. Totally 3228 application was sent to 1614 job openings.

On aggregate level there was no significant difference between males and females in callback
rates, only a slightly higher callback for the female. On occupation level there was some
variations. In general in male dominated occupation there was no real disadvantage of being a
female as the call back rates was pretty much the same as for the male. In mixed and female
dominated occupations there was just a small advantage of being a female. Business sales
assistant which goes under mixed occupation the female has 6 percentage points higher
probability than the male compared to computer professionals. In the female dominated
occupations; restaurant workers, accountants and preschool teachers the female had a higher
probability of receiving a positive callback. Where restaurant workers had as high as 20
percentage points higher probability.

To sum up only a weak proof of discrimination linked to gender segregation in the Swedish
labour market was found and thereby this study could not find any statistical evidence that an
increasing share of females in the occupations can explain the differences in callback rates in
the Swedish labour market.

2.2. Studies using situation testing

A situation test by Levinson (1975) let the applicants make job inquires by phone to jobs in
Atlanta in Spring 1974. The employers were called by both one male and one female to jobs
that were defined as typical male or female occupations. First the female started to call an
employer to a sex-inappropriate job (i.e. a typical male occupation) which was followed half
an hour later with a call from the male which in his case would be a sex-appropriate job. To
be able to analyze the response, the applicants had to meet the employer in the same way with

12
respect to manner, choice of words, and qualifications. If one sex was treated differently it
would be a possible case of discrimination.

The employers’ responses were divided in four categories: “clear-cut” where the employer
clearly restricted or eliminated chances for employment due to gender, “ambiguous” where
the employer treated the male and female applicant differently but one could not ascertain that
it was due to the gender. When both applicants got the same treatment it was classified as
“none”, finally the last category was “not discern”.

Levinson detected a considerable amount of gender discrimination. Out of the 256 jobs, 35
percent of the responses fell in under the “clear-cut” category, 27 percent was “ambiguous”
responses and the remaining 31 percent was treated equally. When comparing responses to
inquiries of male and female, result showed that males faced discrimination in a larger extent
than females. When the female inquired for a typical male job, “clear cut” response was
produced in 28 percent of the cases whereas the corresponding figures for males was over 44
percent. Thus, when males applied for sex-inappropriate jobs they were discriminated against
in a greater extent than females applying for sex-inappropriate jobs.

A study by Neumark, Bank, and Van Nort (1995) also applied the situation testing technique
when gender discrimination in restaurant hiring in the US was examined. The aim was to
investigate if there were any differences in the hiring of waiters and waitress depending on the
restaurant status and if it can explain the sex gap in earnings between them. The authors
assumed that high-price restaurants offered higher earnings than lower and medium-price
restaurants. Further, previous studies had show that high status restaurants favoured waiters
instead of waitress. The authors, therefore, divided restaurants in “high-price”, “medium-
price,” and “low-price” to be able to investigate the role of discrimination in vertical
segregation.

Two males and two females were involved in the experiment and applied for jobs in 65
different restaurants in Philadelphia during three weeks in 1994. Three different resumes were
created with identical content that was rotated between the job applicants. One male and one
female applied to the same job where the applicant made visual contact with the employer.
The applicant went in to the restaurant and handed over the application to the first employer

13
encountered and requested that the resume would be handed over to the manager. The
decision of an interview was therefore solely based on the information on the CV and not the
personality or appearance of applicant. If there were differences in the outcome of job offers
(after the interview stage) then personality or appearance mattered.

Out of the 130 attempts, 54 of them resulted in a job interview and 39 of those interviews led
to a job offer. The job offer rates for the males were 0.48 in both high- and medium price
restaurants and 0.10 in low-price restaurants. The corresponding figures for females were
0.09, 0.29, and 0.38 respectively. The results show statistically significant evidence of that
women are discriminated against in high-price restaurants but favoured in low-price. At the
interview stage the figures are roughly in line with the job offer rates.

The probability of receiving an invitation to an interview and the probability of receiving a


job offer in high-price restaurants was lower for females than for males. This suggests that the
gender earnings gap might be explained by the existing hiring discrimination in high price
restaurants against females, as those restaurants are associated with higher wages and tips.

Table 1 present a summary of the above reviewed literature. To sum up, most studies find
evidence of discrimination suggesting that it is quite widespread phenomena across different
labour markets and occupations. One can also notice a tendency that in the early studies
females were generally the subjects of discrimination while later studies instead show a
tendency of males being discriminated against.

14
Table 1: Concluding summary of field experimental research on gender discrimination

Investigation Applications Nr. of


Study period Labourmarket Method (N) occupations Discr. Against males Discr. Against females
Firth (1982) 1978 U.K CT 1 No Yes

Riach & Rich (1995) 1983-1986 Australia CT 991 7 No Yes in 2 of 7 occupations

Weichselbaumer (2000) 1998-1999 Austria CT 477 4 Yes in 1 of 4 occupations Yes in 1 of 4 occupations

Riach & Rich (2006) U.K CT 873 4 Yes in 3 of 4 occupations Yes in 1 of 4 occupations

Carlsson (2011) Sweden CT 1614 13 Yes in 4 of 13 occupations No

Levinson (1975) 1974 USA Situation 256 no specific Yes Yes

Neumark et. al (1995) 1994 USA Situation 65 1 Yes, in low price restaurants Yes, in high price restaurants

15
2.3. Some other related studies

The following studies have not used traditional correspondence or situation testing
experiments but they are closely related field experiments that provide interesting information
about gender discrimination and are related to our study and therefore included.

Goldin and Rouse (2000) dealt with the issue of hiring in orchestras in the United States.
Males have been over-represented in hiring and until recently got handpicked by music
directors. To overcome the bias in the hiring of musicians, a policy change in hiring took
place in the 1970s and 1980s. Instead of handpicking musicians the orchestras now become
more opened and democratic in their hiring process and procedures like hiding the applicant’s
identity was adopted. The question the authors wanted to answer is whether the hiring process
became more impartial with “blind” auditions.

Orchestra auditions usually consist of three rounds, preliminary, semi-final and final. In blind
auditions a screen was used to hide the identity of the player for the committee. Almost all
preliminary rounds were “blind” and at the semi-finals most were as well, finals were,
however, rarely “blind.” Actual audition records of eight major symphony orchestras were
used from the years of late 1950s through 1995. The use of blind auditions increased female’s
probability to advance from the preliminary rounds by 50 percent and thereby the likelihood
that they would be selected for the final round. The authors also used roster data where blind
auditions explained 30 percent of the increase in the proportion of female hiring in symphony
orchestras. The study showed that blind auditions in the hiring process increased the
probability for female musicians to be employed.

Finally in a study by Edin and Lagerstöm (2006) a quasi-experimental technique was used to
reveal discrimination in the Swedish labour market. The aim of the study was to compare the
employers contact rate of individuals that fill out their gender in their resumes with
individuals that do not. That enabled the researchers to see how employers use gender as a
screening device in the hiring process. To carry out the experiment the authors pulled data
from the Swedish public employment office (Arbetsförmedlingen) where individuals have
been posting their resumes and thereby made themselves available to be contacted by
potential employers. The individuals can choose to either include or leave out personal

16
information such as name and gender. As the authors had no power of the applicants resumes
or the employers that contacted them the experiment was of a quasi-experimental nature.

Eight thousand and 43 individuals joined the experiment where 922 of them chose to conceal
their name, gender, or both. As information on gender was needed in study, the applicants had
to provide this to the researchers. By initially observing the full sample on employers contact
rates categorized by gender, the results showed that the females got 7 percentage points fewer
contacts than males (0.302 compared to 0.378). Even after controlling for observable
characteristics female received fewer job contacts than males regardless of whether the
authors examined estimates made on the full sample (-5.1 percentage points) or estimates
made on the non-blind sample (-4,7 percentage points). There seemed to be no effects of
concealing information on the contact rates as the main effect (blind-gender) were statistically
insignificant on the full sample. However, when estimating an interaction effect between
blind-gender and female a significant difference was obtained (0.057). This implies that
females, if they chose to conceal their name, increase their chances of getting contacted by 5
percentage points. These results suggest that employers used gender as a screening device in
the hiring decision. One should, however, be careful to interpret these results as clear cut
discrimination as the results may suffer from selection bias on observables such as cover
letter. Again this study provides evidence that females might benefit from hiding their gender
for potential employers.

17
3. Theoretical framework and hypothesis

3.1. Economic definition of discrimination

Economic models of discrimination can be divided into two main classes: competitive and
collective models. In collective models one group acts collectively against another whereas in
competitive models agents act individually. We will focus on competitive models as almost
all research has focused on these models.

There are several definitions of discrimination but we use a standard economic definition.
Labour market discrimination occurs when equally productive individuals get different
outcomes in terms of wages or job access due to observable characteristic such as gender, age,
and ethnicity. According to Cain (1986), discrimination can formally be defined as:

y = xβ + α z+ ε ,

where y represents the wage, x is a vector of productivity characteristics that determines


productivity (observable by firms), and z is a binary variable equal to 1 if the individual is a
member of a minority group. If α < 0 it implies that z have a negative impact on wage and
thus discrimination against the minority group exist.

Discrimination models have commonly been divided in two broad types: taste-based
discrimination models and statistical discrimination models. We will now briefly review these
two models and derive the main research hypothesis that we test in this paper: Female
workers have less probability of receiving invitations to a job interviews and thereby job
offers.

3.2. Taste-based discrimination

Becker (1957) developed the seminal economic theory of discrimination where he translated
employers’ prejudices into economic terms as a “taste” for discrimination. Becker (1957)
argued that an employer has a taste for discrimination when he prefers one individual to
another even though both individuals are equal in productivity. The employer believes that

18
some individuals are more costly than they actually are which is then reflected in the hiring
decision, wage setting, or both. In our case, it implies that employers with a taste for
discrimination against females will set the wage for females accordingly: wage of male =
wage of female – discrimination coefficient. To better understand taste-based discrimination it
is a good idea to formulate a model. In Becker (1957), the model of taste-based discrimination
prevails under the assumption of full competition and that workers are perfect substitutes
(equally productive). Assume that the utility of an individual employer is given by u. The
utility of an employer with a distaste of hiring females can then be expressed as5

u = Π − dL f .

The employer’s utility is a function of the profit and the number of women hired. The
variable Π is the profit, d is the discrimination coefficient which represents the size of the
distaste the employer experience by hiring a female, and Lf is the number of female workers
hired. Note that the employer’s utility increases with profit but decreases with d for each
additional female hired. Further, profit can be formulated as

Π = pq(Lm + L f ) − wmLm − w f L f ,

where p is price, q(Lm + Lf) is the production function, wf and wm represents the wage of
females and males which are given on a market characterized by full competition, and Lf and
Lm symbolise the number of female and male workers hired. By inserting the profit expression
into the utility function we obtain

u = pq(Lm + L f ) − wmLm − w f L f − dL f .

The employer’s aim is to maximize utility. By derivating the utility function with respect to
the number of workers the employer recruits, we can obtain an expression for the employment
decision of female and male workers. To make the mathematical derivation simpler we set the
price equal to one so that total revenue is simply given by the total production q(Lm + Lf).

5
Throughout this section we assume that female are the subject of discrimination, but of course, the models can
also be formulated for discrimination against males.

19
When derivating the utility function with respect to the number of female and male workers
employed, the following first order conditions are obtained:

∂u ∂q
= − wm = 0, and
∂Lm ∂Lm
∂u ∂q
= − w f − d = 0.
∂L f ∂L f

The employer will continue to hire male workers as long as the additional contribution from
that last worker exceeds the wage cost (i.e., as long as MP – wm > 0). Female workers will be
hired up to the point where the marginal product equals the wage cost, plus the employer’s
cost of the distaste from hiring a female (i.e., as long as MP – wf – d > 0). Depending on the
employers’ taste for discrimination, d, different combinations of female and male workers
will be hired in different firms. Three types of firms can be presumed on the market with
various discrimination coefficients. In the first case the firm will only hire males as the
employer’s taste for discrimination are such that

wm < wf + d.

Since the cost of a male worker is lower than the cost of a female worker, and since male and
female workers are perfect substitutes, the firm will choose to hire only males in order to
maximize the firm’s utility. Similarly, a firm with a taste of discrimination, such that

wf + d < wm,

will only hire females for the same reasons as above. The third type of firm will be indifferent
between hiring male or female workers since the cost of hiring any of them is equal. Hence, in
this case

w f + d = wm .

In this model, Becker (1957) assumes that the wages are flexible, that is, the wages of males
and females can be different. Accordingly, this will result in that female wages will be lower

20
than the wages of males. However, since wages in Sweden are protected by laws and
regulation and are often bargained collectively by unions, it is almost impossible for firms to
set different wages for same type of work. This would constitute too obvious and blatant form
of discrimination. From these premises it justifiable and reasonable to assume that employers
have to set equal wages for female and male workers (i.e., wf = wm = w). Given this
assumption, it follows from the optimization problem of employers that

MP − w − d < MP − w

for employers that have strictly positive discrimination coefficients. The result of a fixed
wage-level is that discriminatory firms with a positive discrimination coefficient cannot
discriminate against females with respect to wages. For this reason this model predicts that
employers’ will instead choose to discriminate already in the first step of the hiring process,
and consequently only hire male workers instead of female workers. Hence, with respect to
the present study, the hypothesis that results from this model is that female workers will have
a lower probability of receiving invitations to job interviews and thereby job offers.

3.3. Statistical discrimination

The pioneering papers by Phelps (1972) and Arrow (1973) document the theory of statistical
discrimination. Firms have limited information about the skills and productivity of the
applicants, as the marginal product is not easily observed. Hence, firms have incentives to use
easily observable characteristics such as race and gender to discriminate between workers.
Employers make conclusions about an individual applicant’s productivity based on the
productivity of the group he or she belongings to. Even though an employer has information
that indicates that two applicants have exactly the same productivity the employer will pay a
higher wage to the applicant that belongs to the group with the highest productivity. A formal
model can be employed to explain how statistical discrimination affects the wage of workers.

Assume that the employer make an assessment of the applicant’s productivity based only on
the applicant’s resume and interview. However, it is difficult to make a complete accurate
assessment of the productivity of an applicant, as it is only an approximate of the actual
productivity, say y. Formally, let the employer’s assessment be expressed by

21
b= y+ε .

Here, the employer’s assessment of the applicant’s productivity is given by b, which is equal
to the actual productivity y, plus an error term, ε. In addition to the assessment, the employer
also has information about the average productivity of the group the applicant belongs to.
Given the employer’s own assessment, b, and the average productivity of the group y*, the
employer makes a combined estimate of the worker’s productivity. Thus, the expected
productivity can be expressed as

ŷ = (1− g) y*+gb ,

where g measures the correlation between the employers assessment and the real productivity.
Note, if g is equal to one, the expected productivity is completely dependent on the
assessment of the employer. In this extreme case the employer ignore the average productivity
of the group. The other extreme case, where g is equal to zero, the employers base the
expected productivity solely on the average productivity of the group.

From the equation above we can now explain how statistical discrimination can arise between
females and males. Assume that the employer believe that the average productivity is lower
for females than for males (i.e., yf* < ym*). For instance females can be absent from the labour
market longer due to parental leave. We further assume that the correlation between the real
productivity and the employer assessment is equal for females and males, (that is, g = gm = gf).
Then, the expected productivity of males and females can be written as

ŷm = (1− g) y*m +gb, and


ŷf = (1− g) y* f +gb.

Now, if we subtract the expected productivity for females from the expected productivity for
males we can see that

ŷm − ŷ f = (1− g) ( ym* − y*f ) > 0 .

22
Hence, the male get a higher expected productivity than the female. Now, let the probability
that an employer hires an applicant be denoted by σ. It is reasonable to believe that σ is a
function of the expected productivity for a profit-maximizing firm. It is also reasonable to
believe that higher expected productivity increases the probability to be employed. Hence,
formally this means that

∂σ ( ŷ)
> 0.
∂ŷ

Since we have shown that the expected productivity of males is higher than the expected
productivity of females, the following result then immediately follows:

σ m ( ŷm ) > σ f ( ŷf ) .

Thus, the probability of being employed is higher for males, which in this model implies that
the employer will choose to hire males. Hence, as in the case of taste-based discrimination,
the hypothesis that results from this model is that female workers will have a lower
probability of receiving invitations to job interviews and thereby job offers.

Both models of discrimination, taste-based and statistical, suggest that discrimination will
emerge already in the first step of the hiring process. We use these predictions to motivate the
present study. Notice, that these theories are not mutually exclusive. Any discrimination that
will be found can therefore be a result of taste-based discrimination, statistical discrimination,
or both. It is, however, out of the scope of this study to discriminate between these theories.

23
4. Experimental methodology

4.1. Randomized correspondence testing

In the literature review section we dealt with issues concerning correspondence test and
situation test. In this study we applied a correspondence test since it allows us of rigorous
control over the treatment and control conditions, and also ease the logistics of the
experiment.

There are two ways to conduct a correspondence test, either using a randomized or matched
procedure. We are utilized the randomized procedure which implies that the employer receive
one randomly picked application either from the female or male applicant. This is contrary to
the matched procedure where the employer receives applications from both the female and
male applicant. The drawbacks of sending several applications to the same employer are that
it increases the chances of getting detected and that one cannot have completely identical
resumes. It requires a modification of the applications in a way that make employers believe
that they have received two different applications even though there is no difference in
quality, skills and characteristics (except for gender) between the applicants. On the other
hand, possible discrimination becomes clear as differential treatment of applicants
immediately appears. In a randomized procedure discrimination is not as apparent and
obvious, as we can not be assured if the employer rejects the applicants due to gender or for
other reasons for instance better qualified fellow applicants. One might then make false
conclusions over the existence of discrimination. Therefore this method requires collecting a
larger sample to out-rule outcomes that arise by chance.

All previous field experiments on gender discrimination have used the matched procedure
why we can further contribute to existing literature by using the randomized procedure.6 It is
also a good idea to use an alternative procedure when replicating previous studies in order to
make ourselves more or less convinced about a current paradigm.

6
Randomized design has been used more frequently in studies examining discrimination against ethnic
minorities and in studies dealing with housing market discrimination. See, for example, Ahmed, Andersson, and
Hammarstedt (2010) and Carpusor and Loges (2006).

24
4.2. Identities of the fictitious applicants

To carry out the experiment two identical resumes and two cover letters was created, where
each was assigned either with a female or a male sounding name. The names of the applicants
where picked from the name frequency data from Statistics Sweden where the most common
Swedish male and female name at the time was Erik and Maria. These names are optimal
choices as they unambiguously signal the gender of the applicant. We also used Johansson, a
typical Swedish last name, for both of the fictitious applicants.7

To uphold a decent callback rate the resumes and cover letters had to be attractive to the
employers. The fictitious applicants therefore were in an appealing age of 32 years at the time
of the experiment, married but with no children. They exhibited a considerable amount of
work-experience (11 or 15 years depending on occupation) within the specific occupation
they applied for. The cover letter also signalized that they were active individuals that enjoyed
jogging, walking, and outdoor life. Finally the applicants were engaged in charity work such
as, the Swedish Red Cross, in order to signalize a considerate and caring person.8

Next we created an email address for each applicant and also assigned them with a real phone
number with voice mail function.9 To enable receiving written response from employers a real
postal address was bought in Stockholm.10

4.3. Choice of occupations

Contrary to previous studies several occupations were selected to further contribute to


existing literature. The occupations were chosen using the following criterions. First, the
occupations had to have a large supply of vacancies in order to achieve a satisfactory size of
sample. Second, it had to be a balanced mix between male- and female dominated and mixed
occupations as it might be different outcomes depending on the gender decomposition.
Different results might also occur depending on educational attainment why we also chose a
mix of occupations were university diploma was required and not required.

7
The most common names in Sweden, current as of 31 December 2009 (see
http://www.scb.se/Pages/ProductTables____30919.aspx).
8
For an example of an resume and cover letter. Se appendix A1 & A2.
9
The e-mail accounts were registered at Gmail.com and two mobile telephone numbers were registered at Tele2
Sverige AB.
10
The real physical address was registered at the postal address provider Brevia Mail KB.

25
The following eleven occupations were chosen with respect to our criterions: nurses, cleaners,
child welfare workers, drivers, building workers, industry workers, restaurant workers,
teachers, shop assistants, salespersons and hairdresser.11

4.4. Application procedure

The experiment was carried out between August 16, 2010 and to the end of January 2011,
where all available jobs in the chosen occupations were applied for. To find the vacant jobs
we used the website of the Swedish Public Employment Service, which is the main channel
for job search in Sweden.12 Jobs that required applications through ordinary post or through
external web-forms were excluded since it is time-consuming and an unreasonable amount of
arduous work needed to be put into the application process. Also, we were usually unable to
provide some of the information that these web-based forms require (for example, personal
security numbers). Recruitment companies were also excluded as information about the real
hiring firm are hard to find. Finally, jobs that required application through telephones were of
course also excluded as the purpose was not to use a personal approach or investigate
discrimination at the interview stage. However, most of the employers expected applications
by email.

Each day vacant job announcement from the previous day were applied for, where each
employer received one random application from either the male or the female applicant. We
attached the resume and cover letter and wrote a short conventional message in the email to
the employer where the applicants declared interest in the employment.

4.5. Recorded information

Information about the employer and the job specifics was collected and registered. First, we
recorded the gender of the applicant and the type of occupation. We also recorded the location
and the date when the vacant job was available on the website, whether the job was full-time
or part-time, if it was one or several recruiters, and the gender of the recruiter (if only one). To
easily find the employers when they responded the email address and the telephone number

11
See, Appendix A4 for an overview of the occupational structure in Sweden.
12
See, the webpage of Swedish Public Employment Service (Arbetsförmedlingen) at http://www.ams.se/.

26
were also recorded. Further, detailed information was collected from a website that provides
information about firms and companies in Sweden.13 The following information was
collected: number of employees, percentage of females, if the company was publicly owned,
and the gender and age of the CEO. However this information was not available or complete
in some cases.

Finally, we recorded the responses from the employers. A positive response was considered
when the employer invited the applicant for an interview, offered a job or in another obvious
way showed interest in the applicant. Positive responses came in solely through telephone or
e-mail. They were promptly and politely declined, to minimize the inconvenience to the
employers.

13
We used the website http://www.121.nu/ to find complementary information about the employers. They
collect this information from well-known Swedish data bases.

27
5. Results

5.1. Descriptive statistics

Table 2 presents the percentage of positive callbacks both on aggregate level and for different
occupations for the male and female applicant. Recall that a callback is interpreted as a
positive response if the employer asked for an interview or requested complementary
information. Table 2 also contains the p-values to determine whether there is a statistically
significant difference in the positive response rate between the applicants.14

Table 2: Percentage of job applications that led to a callback decomposed


by occupations and p-values generated by a t-test (the total number of jobs
applied for in each category is given in the parenthesis)
Positive response p-values
Male (1) Female (2) Prob((1)=(2))

All occupations (2083) 29.63 (1036) 32.00 (1047) 0.243


Type of occupation
Shop sales assistant (266) 15.04 (133) 10.53 (133) 0.272
Construction worker (175) 30.00 (90) 30.59 (85) 0.930
Preschool teacher (329) 53.66 (164) 58.18 (165) 0.410
High school teacher (143) 47.89 (71) 51.14 (72) 0.678
Motor-vehicle driver (101) 15.01 (53) 20.83 (48) 0.457
Cleaner (261) 11.20 (125) 17.65 (136) 0.141
Restaurant worker (272) 25.19 (135) 35.77 (137) 0.058
Salesman (254) 32.92 (126) 29.69 (128) 0.374
Nurse (95) 36.36 (44) 45.10 (51) 0.394
Mechanic worker (133) 22.73 (66) 13.43 (67) 0.166
Hairdresser (54) 24.14 (29) 36.00 (25) 0.351

Note: The null hypothesis is that there is no statistically significant difference


in the response rate between the applicants.

In total, 2,083 jobs were applied for that was unequally distributed between the occupations.
In the nurse and hairdresser occupation only 95 and 54 jobs were applied for compared to, for
instance, preschool teachers and restaurant workers where 329 and 272 jobs were applied for,
respectively. The wide dispersion can be explained by three reasons; diverse demand, the
extent of which vacancies where handled by recruiting companies, and whether or not
applications were accepted by email. Most importantly, however the distribution of jobs

14
The null hypothesis is that there is no statistically significant difference in the response rate between the
applicants.

28
between the male and female applicant was evenly distributed, both on aggregate level as well
as for each occupation.

Turning our attention to the callback rates, there is no major difference between the male and
female applicant on aggregate level. The male applicant applied for 1,036 job openings and
received a positive response in about 30 percent of his applications. Correspondingly, the
female applicant applied for 1,047 job openings and received positive callbacks 32 percent of
the cases. The difference in the callbacks are not prominent and the p-value reveals that the
difference is not significant (p = 0.240).

Regarding callback rates for different occupations there are some differences between the
male and female applicant. The female receive quite surprisingly more callbacks than the
male in eight out of 11 occupations; however these differences are not statistically significant
at conventional levels. Only for restaurant workers a significant difference between the male
and female applicant is apparent (p = 0.058). The male received positive responses in 25
percent of the cases, whereas the corresponding figure for the female was 36 percent. These
figures suggest rather strong discrimination against the male. Only two other occupations,
cleaners and mechanic workers, exhibited a difference in callback rates but these differences
were only weakly significant. The figures for cleaners, again, indicate that the male applicant
is discriminated against, but for mechanic workers the female applicant is instead the subject
of discrimination.

In Table 3 we further continue with the descriptive statistics over the percentage of
applications that led to positive callbacks for the male and female applicant, but instead
decomposed by a set of various job characteristics.

29
Table 3: Percentage of job applications that led to a callback decomposed by job
characteristics and p-values generated by a t-test (the total number of jobs applied
for in each category is given in the parenthesis)
Positive response p-values
Male (1) Female (2) Prob((1)=(2))

Job characteristics
Job in metropolitan area (674) 28.36 (335) 32.74 (339) 0.217
Job in non-metropolitan area (1409) 30.24 (701) 31.64 (708) 0.571
Public sector employer (263) 41.30 (138) 44.00 (125) 0.660
Private sector employer (1799) 27.87 (890) 30.25 (909) 0.265
Full-time position (1538) 31.56 (770) 32.94 (768) 0.562
Part-time position (541) 23.95 (263) 29.49 (278) 0.146
Positon with conditional tenure (638) 33.77 (305) 38.14 (333) 0.252
Permanent position (1440) 27.92 (727) 29.17 (713) 0.600
Female recruiter (658) 35.58 (326) 42.17 (332) 0.083
Female share in company <20 % (572) 30.63 (160) 21.83 (142) 0.084
Female share in company >80 % (229) 41.75 (103) 42.06 (126) 0.962
Female share in company <40>60 % (1127) 29.34 (559) 30.46 (568) 0.682
Numbers of employees <10 (644) 27.91 (326) 27.04 (318) 0.805
Numbers of employees >10<50 (1360) 28.81 (663) 30.70 (697) 0.445
Numbers of employees >100 (125) 19.64 (56) 30.43 (69) 0.172

Note: The null hypothesis is that there is no statistically significant difference


in the response rate between the applicants.

Most of the jobs were full-time and permanent positions and from the private sector. As only
nurse and teachers are typical public sector occupations the dominance of the private sector is
not surprising.

In table 3 the overall picture of the difference between the male and female in callback rates is
that it is small, for most of the job characteristics. Furthermore, the differences are again
mostly insignificant. There are two job characteristics that reveal statistically significant
difference between the male and female applicant. A female recruiter tends to be an advantage
for the female applicant. The female applicant received a positive callback in 42 percent of
the cases when the male applicant received a positive callback in 36 percent of the cases when
the recruiter was a female. The female share in the company also has a significant effect on
the callback rates. A low share of females (less than 20 percent) in the hiring firm seems to be
a disadvantage for the female applicant as she only receives 22 percent positive callbacks
compared to the male’s 31 percent. There is also a weak tendency that the female applicant
has an advantage if it is a part-time position or a large firm.

30
5.2. Regression analysis

In this section we estimate a linear probability model to determine if the female and the male
have different probabilities of being invited to an interview.15 We are estimating four different
models with the callback dummy as the dependent variable. For each model we increase the
number of explained variables. Model 1 control only for the female dummy, which is the
variable of most interest to keep track off. Model 2 controls for the female dummy and
different occupations and Model 3 adds the job characteristics. Finally in Model 4,
interactions between female and type of occupation as well as female and job characteristics
are incorporated. The linear probability estimates of receiving a positive callback can be seen
in table 4, with the standard errors in parentheses.

15
To verify the credibility of the results we also estimated Probit models, which rendered similar results. These
estimates can be found in Appendix A3.

31
Table 4: Linear probability estimates of the probability of receiving a positive callback

Variable Model 1 Model 2 Model 3 Model 4


Female 0.024 (0.020) 0.024 (0.019) 0.011 (0.029) -0.022 (0.114)
Occupations
Shop sales assistant - Reference Reference Reference
Construction worker - 0.175*** (0.043) 0.164** (0.069) 0.067 (0.095)
Preschool teacher - 0.430*** (0.036) 0.501*** (0.065) 0.403*** (0.094)
High school teacher - 0.369*** (0.045) 0.495*** (0.086) 0.487*** (0.128)
Motor-vehicle driver - 0.051 (0.051) -0.026 (0.075) -0.103 (0.109)
Cleaner - 0.017 (0.038) 0.007 (0.059) -0.063 (0.088)
Restaurant worker - 0.177*** (0.038) 0.257*** (0.057) 0.127 (0.082)
Salesman - 0.195*** (0.038) 0.147*** (0.056) 0.114 (0.079)
Nurse - 0.282*** (0.052) 0.162 (0.102) 0.053 (0.153)
Mechanic worker - 0.053 (0.047) -0.016 (0.065) -0.035 (0.095)
Hairdresser - 0.169* (0.065) 0.441*** (0.131) 0.354** (0.178)
Job Characteristics
Jobs in metropolitan area - - -0.014 (0.031) -0.027 (0.045)
Public sector employer - - -0.344** (0.179) -0.276 (0.256)
Full-time position - - 0.050 (0.039) 0.094** (0.057)
Permanent position - - -0.038 (0.035) 0.018 (0.053)
Female recruiter - - 0.017 (0.034) -0.013 (0.048)
Female share (>80 %) - - 0.055 (0.046) 0.045 (0.069)
Female share (<20 %) - - 0.086 (0,043)** 0.125** (0.60)
Numbers of employees (<10) - - -0.13 (0.032) -0.30 (0.046)
Numbers of employees (>100) - - 0.010 (0.050) -0.073 (0.074)
Interactions
Shop sales assistant × female - - - Reference
Construction worker × female - - - 0.187 (0.139)
Preschool teacher × female - - - 0.192 (0.130)
High school teacher × female - - - 0.050 (0.173)
Motor-vehicle driver × female - - - 0.163 (0.151)
Cleaner × female - - - 0.139 (0.119)
Restaurant worker × female - - - 0.256** (0.114)
Salesman × female - - - 0.048 (0.113)
Nurse × female - - - 0.237 (0.207)
Mechanic worker × female - - - 0.066 (0.131)
Hairdresser × female - - - 0.209 (0.262)
Metropolitan area × female - - - 0.026 (0.062)
Public sector employer × female - - - -0.145 (0.361)
Full-time position × female - - - -0.083 (0.078)
Permanent position × female - - - -0.118* (0-071)
Female recruiter × female - - - 0.071 (0.068)
Female share (>80 %) × female - - - 0.007 (0.093)
Female share (<20 %) × female - - - -0.072 (0.086)
# of employees (<10) × female - - - 0.047 (0.064)
# of employees (>100) × female - - - 0.148 (0.101)
Constant 0.296 (0.014) 0.116*** (0.029) 0.108** (0.059) 0.120 (0.084)
R2 0.001 0.104 0.19 0.19
Note * indicates statistical significance at 10 percent, ** at 5 percent, and *** at 1 percent.
The standard errors are given in the parentheses.

32
From Table 4, insignificant estimates dominate the models, which reflect the general picture
from the descriptive statistics. In Model 1, we can see that the female dummy variable is
positive, stating that the female applicant has 2.4 percentage points higher probability of
receiving a positive callback. The estimate is, however, not statistically significant. We,
therefore, cannot reject the null hypothesis of no difference between the female and male
applicant in the likelihood of receiving a positive callback.

When controlling for different occupations in Model 2, one can observe that the probability of
getting a positive callback depends greatly on the type of occupation. For pre- and high school
teachers there is a significant higher probability of receiving a positive response compared to
the reference occupation, shop sales assistant. Also, the probability of receiving a positive
callback is higher in the occupations construction worker, restaurant worker, salesmen, and
nurse than in the shop sales assistant occupation. The female variable is unchanged from
Model 1, i.e. positive but not statistically significant.

When adding job characteristics in Model 3 there are two variables that significantly affect
the probability of receiving a positive callback: public sector employer and the share of
females in the firm. A public sector employer drastically decreases the probability of getting a
positive response (-34.4 percentage points), while it increases when the female share in the
company is less than 20 percent (8.6 percentage points). All other job characteristics show no
significant effect. Again the female variable remains positive but insignificant although it has
decreased substantially from 2.4 percentage points to 1.1 percentage points.

In Model 2 and 3 there is no possibility to distinguish how the female alone affect the
probability of receiving a positive callback for different occupations and job characteristics.
Therefore, in model 4 interaction variables are included between the female and each
occupation and job characteristic. Generally there is no significant effect of being a female,
neither for different occupations or job characteristics. Only for one occupation the interaction
variable has a significant effect. In the restaurant worker occupation the female have 25.6
percentage points higher probability than the male of receiving a positive job response
compared to the reference occupation, shop sales assistant. There is also only one significant
interaction variable for job characteristics, permanent position. The female’s probability of
getting a positive response decrease if the position is permanent compared to a tenure
position.

33
The female dummy variable in Model 4 changed sign and is now negative but still statistically
insignificant. The models, on the whole, show no evidence of any gender discrimination in the
first step of the hiring process on the Swedish labor market. We only find some evidence of
discrimination against males in the restaurant occupation.

34
6. Discussion
In this study we examined the degree of gender discrimination on the Swedish labour market
during the autumn of 2010. We utilized a randomised correspondence testing procedure where
one female and one male applicant sent fictitious applications to real employers in eleven
different occupations. The aim was to test if female workers have less probability of being
invited to job interviews and thereby job offers.

We found almost no evidence that the female and the male applicant have different
probability of being invited for an interview. We found in only one occupation, restaurant
workers, that the male had lower probability than the female to receive a positive callback.
Based on the amount of femininity presupposed to be associated with the restaurant worker
occupation, the discrimination against males may not be surprising. None of the other female
dominated occupations exhibit any significant difference in callbacks even though several of
them must be considered as occupation associated with femininity. That most of the female
dominated occupations are from the public sector could be a possible explanation. Public
employers might be more obligated to follow regulations and recommendations and more
responsive to political pressure of employment equity than private employers and higher
propensity to treat applicants equal (see for instance Esping-Andersen 1990; Wharton 1989).

The probability for the female to be invited to an interview was not generally affected by
different job characteristics. We found weak evidence that a permanent position negatively
affected the female’s probability of getting a positive callback. Our applicants were childless
and in a relatively attractive age of potential entry into parenthood within the imminent future.
Of biological and traditional reasons it will affect the females working time (productivity)
more than the males which some employers might use as an argument to prefer males on
permanent positions.

Our results are in line with the comparable study by Carlsson (forthcoming) who employed a
matched correspondence testing procedure Carlsson (forthcoming) also found that males were
discriminated against in the restaurant worker occupation but in no other female dominating
occupation. He could neither find any evidence that male and female applicants have
different probability of being invited to an interview in male dominated occupations. Carlsson

35
(forthcoming) also included some job characteristics and he found evidence that males were
more likely to be invited to an interview if the employer was a public sector employer.
Carlsson (forthcoming) also found in line with our results that neither sex composition nor the
sex of the recruiter influenced the difference in the male and female applicant’s chance of
receiving a positive callback.

In comparison with international studies the picture is more ambiguous. In early studies
females have generally been the subjects of discrimination while later studies show a
tendency that males are being discriminated against. Generally, gender discrimination is more
widespread and apparent internationally compared to what Carlsson (forthcoming) found and
what we find in this study over the Swedish labour market.

Even though we cannot find much evidence of gender discrimination on the Swedish labour
market in our study one need to be careful before stating so. Our method does not reveal if
discrimination emerges in a later stage as it can only detect discrimination in the first step of
the hiring process. The discriminatory employers could postpone their discriminatory actions
after the interview stage. An employer aware of the non-discrimination regulations and laws
might be afraid of potential reprimands and therefore pick both genders to interview and
reject the female at a later point. Discrimination can also take other shapes than different
hiring outcomes, for instance gender wage differentials where the employers deliberately set
the wage lower for the female than her male counterpart. There are also examples of a glass
ceiling, which is a term used to reflect females difficulties to reach higher positions at the
firm. Furthermore, gender discrimination can prevail on other markets not only the labour
market. Therefore it is important to continue on with further investigation in this area.

Future studies can investigate if changes in the characteristics of the applicant can lead to
different outcomes, where for instance age and children can be modified. Having children can
affect the productivity and thereby the probability of being hired. Old stereotypes where the
female are home and taking care of the children might still be strong and one can presume that
females with children are more negatively affected in the hiring decision than males with
children. Moreover, it would also be interesting to alter the age of the applicant and see if
there is a difference in the probability of getting a job between different age brackets.

36
Another angle could be to apply for more demanding jobs since the current debate suggest
that gender discrimination are assumed to exist in positions higher up in the firm. For instance
managerial positions or jobs requiring a great deal of responsibility could be applied for.

As two studies using two different procedures has ended up with more or less analogous
results we can with precautious optimism say that there does not seem to be much gender
discrimination when it comes to hiring decisions on the Swedish labour market. From an
international perspective Sweden stand out as a front runner in strive for an equal society
which might be reflected in our findings of only modest discrimination.

37
References
Ahmed, A. M., L., Andersson, and M., Hammarstedt (2010): “Can discrimination in the
housing market be reduced by increasing information about the applicants?”, Land
Economics, 86.

Arrow, K (1973): “The theory of discrimination”, in Ashenfelter, O. and Rees, A. (red.),


Discrimination in Labour Markets, Princeton, New Jersey: Princeton University Press.

Becker, G. S (1957): The Economics of Discrimination, Chicago: University of Chicago


Press.

Blau, F. D., and R. C., Ehrenberg (1997): Gender and family issues in the workplace, Russell
Sage Foundation, New York.

Bygren, M., and M., Gähler (2007): “Kvinnors underrepresentation på chefspositioner -en
familjeangelägenhet?”, Institutet för arbetsmarknadspolitisk utvärdering (IFAU), Working
Paper 2007:25.

Cain, G. G (1986): “The economic analysis of labor market discrimination: a survey” in: O.
Ashenfelter and R. Layards, (eds.), Handbook of labor economics, Vol. 1 (North-Holland,
Amsterdam).

Carlsson, M (forthcoming): “An experimental study of sex segregation in the Swedish labour
market: is discrimination the explanation?”, Feminist Economics.

Carpusor, A. G., and W. E., Loges (2006): “Rental discrimination and ethnicity in names”,
Journal of Applied Social Psychology, 36.

Edin, P-A., and J., Lagerström (2006): “Blind dates: quasi-experimental evidence on
discrimination”, Institutet för arbetsmarknadspolitisk utvärdering (IFAU), Working Paper
2006:4.

Esping, A. G (1990): ”The three worlds of welfare capitalism”, in Pierson, C. and Castles, F.
G (red.), the Welfare State reader, Princeton, New Jersey: Princeton University Press.

Firth, M (1982): “Sex discrimination in job opportunities for women”, Sex Roles, 8.

Goldin, C., and C., Rouse (2000): “Orchestring impartiality: the impact of “blind” auditions
on female musicians”, The American Economic Review, 90.

Heckman, J (1998): “Detecting discrimination”, Journal of Economic Perspectives, 12.

Hultin, M., and R., Szulkin (1999): “Wages and unequal access to organizational power: An
empirical test of gender discrimination”, Administrative Science Quarterly, 44.

Levinson, R. M (1975): “Sex discrimination and employment practices: An experiment with


unconventional job inquiries”, Social Problems, 22.

38
Meyersson M., E. M. Trond Petersen, and V., Snartland (2001): “Equal pay for equal work?
Evidence from Sweden and a comparison with Norway and the U.S.”, Scandinavian Journal
of Ecomonics, 103.

Neumark, D., R. J., Bank, and K. D., Van Nort (1995): “Sex discrimination in restaurant
hiring: an audit study”, Working paper, 5024, National bureau of economic research.

Phelps, E. S (1972): “The statistical theory of racism and sexism”, American Economic
Review, 62.

Riach, P. A., and J., Rich (1995): “An investigation of gender discrimination in labor hiring”,
Eastern Economic Journal, 21.

Riach, P. A., and J., Rich (2002): “Field experiments of discrimination in the market place”,
Economic Journal, 112.

Riach, P. A., and J., Rich (2006): “An experimental investigation of sexual discrimination in
hiring in the English labor market”, The B.E. Journal of Economic Analysis and Policy, 2.

Weichselbaumer, D (2000): “Is it sex or personality?- The impact of sex-stereotypes on


discrimination in applicant selection”, Working paper, 0011, University of Linz.

Wharton, A. S (1989): “Gender segregation in private-sector, and self-employed occupations


1950-1981”, Social Science Quarterly, 70.

39
Appendix

A1. Resume and cover letter for an high educated applicant

Maria Johansson
Address: Frejgatan 13, 114 79 Stockholm
Phone number: 0720-17 50 67
E-mail: ms.jhnssn@gmail.com

Employment as a teacher

From your job advertise I see that you are looking to hire a teacher. I am very interested in the
position. With almost 10 years of work experience in teaching I believe I can be a valuable
asset at your school.

I am 32 years old, holding a teacher degree. I have almost 10 years of work experience in
teaching. I am cooperative, fond of my profession and hope I can continue working with it.

I am married but with no children. On my spare time I like to socialize with my husband.
Additionally I enjoy outdoor life. And I jog, take walks and shoot pictures. I am also active in
various organizations and societies. Due to my nature interest I am engaged in
Friluftsfrämjandet but also in the Swedish Red Cross where I had different missions
connected to different types of voluntary work.

I send over a resume with my work experience and educations; I have certifications from
them all.

I am able to speak and write both in Swedish and English.

Thank you for your consideration, I am looking forward to hear from you.

Sincerely,

Maria Johansson

40
Resume

Personal information:

Maria Johansson
Social security number: 1978-05-18
Address: Frejgatan 13, 114 79 Stockholm
Phone number: 0720-17 50 67
E-mail: ms.jhnssn@gmail.com

Work experience:

2005-2010 Employed as a teacher at Bromma High school, Stockholm


2001-2005 Different temporary positions at different schools in Stockholm

Education

1996-2001 The High school teacher program with orientation in Swedish and English,
University of Stockholm
1993-1996 Social Science program, orientation Social science, Bromma High school

Language:

Swedish (mother tongue), English

41
A2. Resume and cover letter for an low educated applicant

Maria Johansson
Address: Frejgatan 13, 114 79 Stockholm
Phone number: 0720-17 50 67
E-mail: ms.jhnssn@gmail.com

Employment as a waitress

From your job advertise I see that you are looking to hire a waitress. I am very interesting of
the position. With almost 15 years of work experience from the restaurant business I believe I
can be a valuable asset at your restaurant.

I am 32 years old and gathered 15 years of work experience in the restaurant business. I am
cooperative, fond of my profession and hope I can continue working with it. I have a
permanent position as a waitress at Quality Hotel Globe.

I am married but with no children. On my spare time I like to socialize with my husband.
Additionally I enjoy outdoor life. And I jog, take walks and shoot pictures. I am also active in
various organizations and societies. Due to my nature interest I am engaged in
Friluftsfrämjandet but also in the Swedish Red Cross where I had different missions
connected to different types of voluntary work.

I send a resume with my work experience and educations; I have certifications from them all.

I am able to speak and write both in Swedish and English.

Thank you for your consideration, I am looking forward to hear from you.

Sincerely,

Maria Johansson

42
Resume

Personal information:

Maria Johansson
Social security number: 1978-05-18
Address: Frejgatan 13, 114 79 Stockholm
Phone number: 0720-17 50 67
E-mail: ms.jhnssn@gmail.com

Work experience:

2005-2010 Employed as a waitress at Quality Hotel Globe, Stockholm


1996-2005 Employed as a waitress at Scandic Hotel, Bromma

Education:

1993-1996 The Hotel- and restaurantprogram, Stockholms School of hotel and restaurant.

Language:

Swedish (mother tongue), English

43
A3. Probit Estimates

Table 5: Probit regressions, reporting marginal effects of the probability of receiving a


positive callback. Standard errors are given in the parenthesis.

Variable Model 1 Model 2 Model 3 Model 4


Female 0.024 (0.020) 0.025 (0.020) 0.011 (0.032) -0.085 (0.140)
Occupations
Shop sales assistant - Reference Reference Reference
Construction worker - 0.234*** (0.055) 0.204 (0.088) 0.066 (0.110)
Preschool teacher - 0.476*** (0.040) 0.530*** (0.658) 0.426*** (0.110)
High school teacher - 0.426***(0.050) 0.529*** (0.076) 0.518*** (0.118)
Motor-vehicle driver - 0.047 (0.066) -0.024 (0.088) -0.117 (0.102)
Cleaner - 0.027 (0.048) 0.005 (0.076) -0.118 (0.094)
Restaurant worker - 0.234*** (0.049) 0.320*** (0.072) 0.151 (0.101)
Salesman - 0.254*** (0.049) 0.194*** (0.073) 0.128 (0.096)
Nurse - 0.347*** (0.061) 0.21** (0.129) 0.063 (0.190)
Mechanic worker - 0.080 (0.060) -0.009 (0.077) -0.034 (0.010)
Hairdresser - 0.229*** (0.080) 0.474*** (0.122) 0.373* (0.195)
Job Characteristics
Jobs in metropolitan area - - -0.014 (0.034) -0.027 (0.048)
-0.300***
Public sector employer - - -0.259** (0.062) (0.017)
Full-time position - - 0.060 (0.041) 0.107* (0.058)
Permanent position - - -0.046 (0.040) 0.013 (0.059)
Female recruiter - - 0.024 (0.037) -0.011 (0.053)
Female share (>80 %) - - 0.065 (0.054) 0.048 (0.082)
Female share (<20 %) - - 0.110** (0,051) 0.145** (0.071)
Numbers of employees (<10) - - -0.13 (0.036) -0.031 (0.049)
Numbers of employees (>100) - - 0.014 (0.058) -0.078 (0.077)
Interactions
Shop sales assistant × female - - - Reference
Construction worker × female - - - 0.297* (0.180)
Preschool teacher × female - - - 0.266 (0.174)
High school teacher × female - - - 0.098 (0.212)
Motor-vehicle driver × female - - - 0.276 (0.214)
Cleaner × female - - - 0.289** (0.181)
Restaurant worker × female - - - 0.347*** (0.150)
Salesman × female - - - 0.122 (0.150)
Nurse × female - - - 0.339 (0.261)
Mechanic worker × female - - - 0.108 (0.178)
Hairdresser × female - - - 0.300 (0.314)
Metropolitan area × female - - - 0.029 (0.071)
Public sector employer × female - - - 0.720 (0.016)
Full-time position × female - - - -0.090 (0.085)
Permanent position × female - - - -0.123 (0.074)
Female recruiter × female - - - 0.088 (0.080)
Female share (>80 %) × female - - - 0.014 (0.107)
Female share (<20 %) × female - - - -0.069 (0.088)
# of employees (<10) × female - - - 0.187 (0.135)
# of employees (>100) × female - - - 0.054 (0.075)
Constant - - - -
Pseudo R2 0.0006 0.086 0.135 0.158
Note * indicates statistical significance at 10 percent, ** at 5 percent, and *** at 1 percent.

44
A4. Occupational structure

Table 6: Share of females and males in different


occupations
Share of Share of
Type of occupation females males
Shop sales assistant 69 31
Construction worker 1 99
Preschool teacher 93 7
High school teacher 53 47
Motor-vehicle driver 4 96
Cleaner 79 21
Restaurant worker 74 26
Salesman 28 72
Nurse 91 9
Mechanic worker 2 98
Hairdresser - -
Soruce: Statistics Sweden, the occupational structure in
Sweden 2009, www.scb.se.

45

Вам также может понравиться