Master’s Thesis
The discrimination against
transgender in the rental housing market in Sweden
An experimental study performed on the Internet
Author: Sofia Fritzson Supervisor: Joakim Jansson Examiner: Tobias König Term: Spring 2021 Subject: Economics Level: Master
Course code: 4NA06E
Abstract
This paper investigated the discrimination against transgender people in the Swedish rental housing market, and is one of the first correspondence studies to examine this question at hand. A total of 800 applications were sent to various landlords advertising rental vacancies on Blocket.se. In total, the cisgender applicants attained a call back rate of 60.7 %, while the transgender applicants attained a rate of 59.0 %.
Furthermore, no unequal treatment was found in positive employer responses for being transgender, when compared to the cisgender group. In the case, where the cisgender and transgender groups were evaluated separately, a favoring of having a woman as a tenant over a man was prominent when comparing the testers within the cisgender group. A penalty of 12.6 % in positive landlord responses was found for the trans man, in terms of likelihood of getting an invitation to additional contact or to showing, when compared to the cis woman. Similarly, the trans woman had a 7.4
% lesser likelihood in getting an invitation to showing than the cis woman. With regards to these results, one can conclude that there exists both discrimination based on gender and gender identity in the rental housing market in Sweden.
Keywords
Discrimination; Field experiment; Transgender people; Housing market, Correspondence study
Acknowledgements
Firstly, I would like to thank my supervisor Joakim Jansson for his support and guidance throughout the writing process of the thesis, as well as Thomas Giebe, Tobias König and, Lars Behrenz for their comments and valuable discussions at the seminars. Secondly, I would like to give a special thanks to my interviewee for providing me with needed feedback regarding the application design of the study. At last, I would like to thank my opponent for giving me your thoughts and ideas of how to improve the study further.
Table of Contents
1 Introduction 1
2 Theoretical Framework 4
2.1 Taste-based discrimination 5
2.2 Statistical discrimination 5
3 Design of the experimental study 6
3.1 The rental housing market on the Internet 6
3.2 The experimental manipulation of gender identities of the applicants 7
3.3 The construction of the application letters 9
3.4 Application procedure 10
3.5 Power calculation 11
3.6 The limitations of the study 11
4 Results and Analysis 12
4.1 Summary statistics 12
4.2 Estimations of the likelihood of getting invited to further contacts and to showings 18
5 Summary and conclusion 22
Appendices
1
1 Introduction
The discrimination against transgender has been seen to be widespread according to a survey performed by the European union agency for fundamental rights in 2014. In the study, more than half of the transgender individuals had endured discrimination or some other form of harassment in the studied year (European Union Agency for Fundamental Rights, 2014). Similar results were found in the survey performed in the US in 2015, in which around one fourth of the respondents reported discrimination in hiring and within the workplace, and one fifth regarding housing, within the past year (James et al., 2016). The large fear of discrimination and mistreatment has further been noticed in health care avoidance (Kcompt et al. 2020) and could possibly be one of the factors to the large rates of suicidality and hospitalization that troubles the minority (Dhejne et al., 2011). Despite this, the studies that cover this topic are scarce and mainly consist of self-reported experiences to examine the discrimination (e.g. Schilt and Wiswall (2008); Grant et al. (2011); European Union Agency for Fundamental rights (2014); James et al., (2016)). This report presents a field experiment on the internet to examine the discrimination against transgender in the rental housing market in Sweden. In the experiment the four different testers (the cis man, the cis woman, the trans man, and the trans woman) applied for different rentals advertised on the Internet. By performing the experiment in Sweden, the study can provide a benchmark or a lower bound of discrimination present in the housing market. This is due to Sweden being seen as one of the most progressive countries in the world in terms of gender equality in the Global Gender Gap Report 2020 (World Economic Forum, 2020).
Even though the method of correspondence testing has become one of the common ways in revealing discrimination against many minorities in the Swedish housing and labour market (for example, see Carlsson and Rooth, 2007; Ahmed and Hammarstedt, 2008, 2009; Ahmed et al., 2008, 2010, 2013;
Ahmed and Lång, 2017, 2019; Carlsson and Eriksson 2019; Bengtsson et al.
2012), up until now only one study has made use of the method in studying transgender. This study was conducted by Granberg et al. (2020) to investigate the hiring discrimination against transgender in the Swedish labour market. The
2 study found that discrimination was highest in occupations largely dominated by a certain gender. In correspondence testing fictitious applications are sent to different job or rental housing vacancies. The amount of positive responses from landlords/employers are then measured and compared for the different testers to attain the treatment effect. The present study followed the methodology of Granberg et al. (2020) in randomizing the applications, such that only one application was sent to each vacancy, but with a focus on the rental housing market.
Another experimental approach was conducted by Langowski et al.
(2017), in which the discrimination against transgender was investigated in the housing market in Boston through matched paired discrimination tests, known as audit testing1. The results showed that the transgender tester were in 61 % of the cases treated discriminatory, and were 9 % more probable to receive a higher rental price than the cis gender tester. An issue with this technique, argued by Heckman and Siegelman (1993), as well as by Heckman (1998), is, however, that it can be hard for testers to be equal enough in all relevant aspects except for e.g. gender or ethnicity, for the results to not be biased, even if they are thoroughly trained. In addition, testers that are aware of the study might become motivated to provide data which are or are not in line with their beliefs regarding the discrimination (Heckman, 1998). By using the method of correspondence testing in the present study, and thereby sending out fictitious applications in which the testers are identical besides for gender identity, the possibility of achieving biased results becomes smaller. And thereby has a better potential in being robust to this critique. However, as highlighted by Heckman and Siegelman (1993) and Heckman (1998) is that, in correspondence testing the experimenter’s choice of how to set the other applicant characteristics, besides for the name, might also have an impact on the estimates of discrimination. This needs to be taken into consideration for the present study.
Furthermore, attempts in studying the discrimination has been done by using non-experimental methods. Geijtenbeek and Plug (2018) used registry data to study the earnings of transgender workers in the Netherlands. And found
1 Audit studies test for unequal treatment by having otherwise identical pairs of people (actors of the study) who vary on a single trait, such as race, sex, or gender identity, apply for the same apartment vacancies or job openings through personal approaches (Ghoshal, 2018).
3 a wage penalty of around 11 percent for male-to-female transgender workers, when registered as female; and only a slight effect when female-to-male transgender workers registered as male. The problem with using registry data is, however, that the effect discrimination has on wages can be hard to obtain, as other omitted variables may impact the wages as well (Neumark, 2018), and that it may only capture a subset of the transgender minority. This in terms of only attaining the individuals who have legally changed their gender and/or name. And thereby fail to include those who identify as the opposite gender, but has not yet began the process of change. In a study by Zeluf et al. (2016) only 34 % of the transgender population in Sweden had legally changed their gender, which is only about one third of the total transgender population in Sweden.
With regards to this, in the present study transgender is defined as an individual that socially identifies as the opposite gender it was assigned to at birth, and that has or has not legally changed their gender. This definition is used in aspiration to be able to include a more representative subset of the true transgender group.
As suggested above, the present study aims to contribute to the
existing literature by using a correspondence method in the Swedish rental housing market. This methodology makes is possible to only alter the gender identity of the applicant, while retaining all other relevant factors constant.
Furthermore, this methodology have not, as previously stated, been used when studying this particular question at hand. It will therefore be the first study to examine the discrimination against transgender in the rental housing market, through the method of correspondence testing. Regarding the minorities of the HBTQ this method has prior been used in studying discrimination against homosexual men (Ahmed and Hammarstedt, 2009) and women (Ahmed et al., 2008) in the Swedish rental housing market. In which the male homosexual couple were seen to be faced with unequal treatment, in terms of receiving far fewer call backs and amount of positive responses than the heterosexual couple.
However, no evidence of unequal treatment was found for the female homosexual couple when compared to the heterosexual couple. Since the perceived discrimination is found to be highest among transgender individuals when compared to the other minorities of the HBTQ community (European Union Agency for Fundamental Rights, 2014; James et al., (2016)), it is of
4 particular interest to see how the results of the present study will differ between earlier related studies conducting this method at hand. In addition, a number of extensions to the main analysis will be considered, in order to evaluate possible heterogeneous effects. Primarily, besides studying the general discrimination against transgender people, the present study will investigate if there is a difference in the discrimination level against transwomen and transmen. Given that there exist some prior evidence of discrimination against men in the rental housing market (see for example, Ahmed and Hammarstedt (2008), Bengtsson et al. (2012)), it is of certain interest to see to which degree this affects transwomen and transmen. Furthermore, the study will examine if there exist varying degrees of discrimination depending on the gender of the landlord, and if discrimination vary by rental price, size or region in the country.
The remainder of this paper will be presented in the following way:
Section 2 demonstrates the theoretical framework; Section 3 provides the design of the experimental study; Section 4 presents the results and analysis; and lastly, Section 5 gives a short summary and conclusion.
2 Theoretical Framework
Discrimination can be interpreted as: in any setting in which two applicants that are identical in all aspects besides for one, such as gender (man, woman), gender identity (cis, trans), age, or ethnicity, are not treated equally. According to the Swedish Discrimination Act, it is illegal to discriminate based on sex, transgender identity or expression, age, ethnicity, sexual orientation, religion or beliefs.
The theories regarding discrimination, is often divided into two main categories in the economic literature, such as the theory of taste-based discrimination (Becker, 1957) and the theory of statistical discrimination (Arrow, 1973; Phelps, 1972). However, since a more advanced experiment is needed in order to distinguish between which of the mentioned theories the discrimination falls in under, the present study is limited to solely find the existence of discrimination. Nonetheless, the above theories can provide this study with a better understanding of the driving forces behind unequal treatment
5 in the housing market. Therefore, a further explanation to these theories will be presented down below.
2.1 Taste-based discrimination
The theory of taste-based discrimination was first introduced by Gary Becker (1957), in his work “The Economics of Discrimination”, and refers to discrimination that is due to tastes of people. In the theory, Becker puts the notion of ethnic prejudice into an economic context. His theory suggests that a prejudiced employer, to a particular group of people, can act as if the perceived cost is higher for hiring someone within that group. Which leads to that fewer or no workers, who contain those traits will be hired (Borjas, 2020). This theory can be applied to the housing market as well. Instead of having an employer, the housing market has landlords. A prejudiced landlord, to a certain minority, could therefore, act as if there is an additional cost to the actual cost when renting out an apartment to someone within that minority. This could take the form of the landlord offering a higher rental price, avoids showing other rental vacancies in the area, and/or provides less information about the present vacancy, for the transgender individual than for the cisgender individual.
2.2 Statistical discrimination
The theory of statistical discrimination was introduced by Edmund Phelps (1972) and Kenneth Arrow (1973). This theory highlights how an employer conducts decisions under the scarcity of information. Due to asymmetric information, the employer will use observable characteristics as a proxy for unobservable traits when deciding who to hire. This is done when the cost for collecting information about the individual applicants’ productivity is excessive. Therefore, an employer whose goal is to maximize profits, but has a scarcity of individual information, will discriminate against individuals within certain groups if the employer believes that they are less trustworthy, qualified, hardworking etc. In other words, the employer will base the hiring decision on statistical information of the group. Therefore, if the group have a smaller productivity initially (which could be the cause of historic discrimination or having attained an unfavorable equilibrium), then each person that belongs to
6 this minority is going to be expected to be less productive, which causes discrimination to emerge (Phelps,1972). This theory is also applicable to the housing market, where landlords might decide to not rent out their vacancies, due to the potential tenants belonging to a group which the landlord has had prior negative experiences with, or in which statistical information regarding the group exists. Earlier surveys state a hardship for transgender individuals in attaining education, employment, and access to healthcare, due to the discrimination based on their gender identity (European Union Agency for Fundamental Rights, 2014; James et al., (2016)). In addition, a large share of transgender individuals in Sweden report facing poor mental health (Zeluf et al., 2016). These could all be factors to which the statistical discrimination could stem from.
3 Design of the experimental study
3.1 The rental housing market on the Internet
To investigate discrimination against transgender in the rental housing market on the Internet, one of the largest online advertisement sites in Sweden was used (www.blocket.se). On this particular site, private individuals and firms are able to put in advertisements to sell, purchase, and rent various objects, in which rental vacancies are a common category. There is a commission for putting an ad on the site, but to reply to an ad acquires no cost. The only obligatory moment that is needed is to create an account by using an email address and a password.
A phone number is then added in a verifying step, which is a new required demand in order to respond to the ads on the site. This number is, however, nonvisual to other participants. All contact is then made through messages on the site. Furthermore, to create a profile and thereby add additional information of the individual itself, such as previous work experiences, interests, education, and pets, is strictly voluntary.
Due to the fact that this study includes collection of information
regarding the responses of landlords without them being informed and able to give consent, the ethics of this study needs to be taken into consideration. The
7 present study was conducted accordingly to the regulations on the site, and the choice to not include consent from the landlords were done as to avoid potential influences of participants that could affect, and/or compromise the study.
Furthermore, offers were shortly replied to and denied, to not attain more time than necessary from the landlords. This in aspiration to do as little harm to the market as possible. Worth noting is, however, that this form of study is not covered by the Ethics Review Act according to the Swedish Ethical Review Authority. In addition, it was confirmed by the mentioned authority, that there were no objections against conducting the experiment from an ethical point of view. With regards to this, it is under the assumption that the present study has taken the ethical aspects into consideration, and that is has been conducted accordingly to these guidelines.
3.2 The experimental manipulation of gender identities of the applicants To create the fictitious applicants, a signaling tool for the gender identity was needed. After consulting researchers within the discriminatory field, as well as interviewing an individual within the minority, a name change was decided to be the identity signal used in the experiment. This treatment has been used by Granberg et al. (2020), in studying discrimination against transgender in the Swedish labor market. However, as opposed to writing the name change as conducted by Granberg et al. this study will attain a more subtle and authentic approach to how applications are formed in the rental housing market. The name change will be constructed, such that a parenthesis containing the individual’s former name will be included between the first and last name when signing the letter. From the interview it was suggested that this is a more common way of incorporating a name change when applying for rental vacancies in the housing market, as opposed to stating the current name and previous name, such as H Larsson (prev. L Larsson) which was used in Granberg et al. For the transgender applicants the name change was from a male to a female name (for the trans woman); and from a female to a male name (for the trans man). The cis gender applicants, on the other hand, included a name change from a male to a male name (cis man); and from a female to a female name (cis woman). Worth disclosing at this point, is that it is not uncommon in Sweden to change ones
8 name, regardless of gender change. Between the years of 2013 and 2018 95,090 individuals in Sweden changed their names (Granberg et al., 2020). Therefore, to have all the applicants including a name change in their application letters is assumed to not be portrayed as rare or particular in any sense for the landlords.
Thereby, two names were needed for each applicant of the study.
However, since names might signal more than just gender, such as socioeconomic character, the two names were randomly chosen out of five common Swedish names for each gender. This potential issue was pointed out in the study by Fryer and Levitt (2004), in which the authors argued that the results of the correspondence testing could be influenced by other socioeconomic signals when names were used to signal group belonging. The names used were drawn from the most common baby names in Sweden in the year of 2000, in which the female names were: Julia, Emma, Wilma, Hanna, Elin; and the male names were: Filip, Oscar, William, Viktor, Simon. These two names per applicant were then matched with a common last name in Sweden, such as Andersson, Johansson, Karlsson, Nilsson, and Eriksson. Where the last name was randomly assigned to each applicant.
The choice to use the most common baby names from the year of
2000, was to attain common names of people that were around 20 to 30 years old. These names have been in the top among the most common baby names since the late 1990’s (SCB, 2021) and were therefore a good fit to the experiment. Since they were randomly drawn there was a possibility that some names would be shared between applicants. However, since the identity involved a name change, there could only be one name, in terms of first names, that was shared with another applicant. Thus, since the names used in the experiment could not be identical, having a shared name with another applicant would have no impact. The final step was to construct email addresses for each of the testers. In this case one of the most common email providers was used (hotmail.com)2. At last, each applicant got an individual account on Blocket.se.
2 For detailed information regarding the applicants, see Appendices.
9
3.3 The construction of the application letters
The application letter firstly needed to consist of an opening fraise, in which the landlord was greeted and the applicant made an introduction of him/herself and expressed interest in the rental vacancy that was advertised. The message was then ended with curtesy, and signed by the applicant with both the current name, and the former name. In which the former name was present in the parenthesis.
The application letters were formulated in four different ways3, and then randomized for each application. The choice to randomize the formulation of the letter, was done as to avoid the possibility of having the landlords becoming suspicious and thereby not act truthfully, as this could compromise the study.
The construction of each message was written similarly as follows:
Hi!
My name is H., and I’m interested in the rental vacancy. Feel free to contact me if it is still available!
Best regards, H. (V.) Nilsson
v.nilsson@hotmail.com
The message in the application letter may appear to be short and not contain much content. Worth noting is, however, that this is a common way to express interest in rental vacancies in Sweden, and in particular on the Internet.
Therefore, the construction has an appropriate approach in order to remain the authenticity of the experiment. Furthermore, the choice to not construct any online profiles for the applicants, was to be able to avoid having to stick to certain characteristics for all four testers. This problem is highlighted by Heckman and Siegelman (1993) and Heckman (1998), who discusses how the experimenter’s choice of how to set the other applicant characteristics, besides for the name, might also have an impact on the estimates of discrimination.
3 For detailed information regarding the formulations of the application letters, see Appendices.
10 Thus, the present study aims to have clean slate applications, in all other regards besides for the text message.
3.4 Application procedure
The correspondence test was conducted between April 17 and May 2 in 2021.
During this time interval, a total of 800 applications were sent to various landlords advertising rental vacancies on Blocket.se. By conducting the method of randomizing applications, only one application was sent to each landlord.
This method was chosen as to avoid raising suspicion with regards to the experiment among the landlords, which can be a concern when applying the method of matched applications4. The rental vacancies were chosen without regards to size and cost. Advertisements that required/requested contact through phone calls, postal letter, or to meet up in person were not included in the study, such that all contact was only performed by email format on the site.
Furthermore, each landlord was controlled to only be contacted once in the experiment. The time and date was recorded for each sent application, as well as information regarding the vacancy, such as location, rental cost, rental size, number of rooms, and landlord characteristics. These variables were collected in order to examine the variations in discrimination.
The observations were first measured in terms of receiving a response or not on the applications. The time frame for attaining a response was limited to one week. Past this limit, the observations were categorized as no response.
The attained responses were then divided into two categories of whether they were being positive, such that further contact and/or an invitation to showing was suggested, or negative, if the application was rejected for any potential reason. Information regarding time and date of attaining the email was recorded.
Lastly, the amount of positive responses which only included an invitation to a showing, without asking for further information, was noted. As earlier stated, each positive response which asked for further information or invited to a showing of the vacancy, was politely declined within short period of time. This was conducted as to not attain more time than necessary from the landlords5.
4 In the matched applications approach, each landlord receives applications from all testers of the experiment (Ahmed and Hammarstedt, 2008).
5 For detailed information of the response letter, see Appendices.
11
3.5 Power calculation
To calculate the needed sample size of the study, power calculation was performed using the statistical software Statas built in command power. In the calculation the anticipated incidence was put to 0.45 for group 1 (the cisgender), and to 0.39 for group 2 (for the transgender). The call back rate when performing correspondence testing in the housing market in previous studies usually lie around 40-50% for the control groups (see. Ahmed & Hammarstedt, 2008, 2009; Carlsson & Eriksson 2014). As for the anticipated effect size of 6 percentage points, it comes from the study by Granberg et al. (2020), in which transgender was seen to have a six percentage point penalty in positive employer response rate when compared to cisgender. Given that Ahmed and Hammarstedt (2009) obtain about twice as large effect in the rental housing market for homosexuals compared to the effect in the labor market (see Ahmed et al.
(2013)), this assumed effect size should be in the lower regions of the expected true effect. The standard deviation used was 0.3, based on the reported estimate by Bengtsson et al. (2012) in studying ethnic and gender discrimination in the rental housing market. Given these values and settling on a statistical power of 0.8 for 5 percent significance, a required sample size of 788 observations in total was obtained, evenly split between cisgender and transgender applications.
With regards to this lower limit, 800 applications were determined for the sample size of the experiment, which is around the amount of posted vacant apartments on Blocket.se over 48 hours.
3.6 The limitations of the study
Even though correspondence testing has been a common way of detecting and analyzing discrimination against minorities, this study needs to take some limitations into consideration. Firstly, due to the fact that gender identity was not directly written in the application, but signaled through a name change, there is a possibility that this signal would not be apparent enough. This in the way, that some landlords might not have noticed the name change at the end of the letter and thereby have missed the treatment. Nonetheless, as mentioned before, Granberg et al. (2020) used a similar design to signal the gender identity as the
12 present study, and were able to detect some effects in the labor market, thus decreasing this concern. Secondly, since the experiment is only conducted on the Internet, other channels in which rentals vacancies can be searched are not included. There is, however, no known reason to why the discrimination would differ between the channels. A third limitation that is worth mentioning is that the type of discrimination, earlier presented in Section 2 Theoretical Framework, will not be able to be distinguished in the present study. This is the case, since statistical discrimination is due to imperfect information of the applicant, which means that the experiment would need additional levels of information to the landlords in order to make this connection.
4 Results and Analysis
4.1 Summary statistics
Table 1 and Table 2 present the summary statistics of the landlords’ responses (in percent) for the two different groups and for the four testers of the experiment, respectively, in terms of mean call back rates and share of positive responses. Included in the parentheses are the actual number of cases out of the total for each group/tester. The share of positive responses is categorized in invitations to showing or further contact, and invitations to showing. The reason behind dividing the positive responses into the given categories, is to more clearly be able to distinguish between variations between the different groups and testers individually. Since the first category involves both invitation to showing and additional contact, there is a potential that this category will be too broad in order to find any variation between the groups. Therefore, by including a second category, in which unequal treatment will be more apparent, there is better chance in attaining a significant difference.
Firstly, by comparing the two different groups in Table 1, the cisgender group received the highest call back rates of 60.7 %, compared to the 59.0 % of the transgender group. Furthermore, similar results were found when looking at the different positive response rates. The cisgender group attained an invitation to showing or further contact in 60.2 % of their applications, in which
13 17.2 % led to invitation to showing. The transgender group received, on the other hand, an invitation to showing or further contact in 57.8 % of their applications, in which 15.3 % resulted in invitation to showing.
Similar patterns to the above results were found when investigating the variation in positive responses depending on landlord gender. The landlords showed a small preference for having cisgender applicants as tenants when compared to transgender; this in terms of the cis gender applicants receiving the highest share of positive responses, in terms of attaining an invitation to further contact or to showing, when compared to the transgender. However, among the female landlords, the transgender individuals had a slightly higher probability in getting an invitation to showing, than the cisgender counterparts.
When investigating different locations, the result displayed a higher likelihood of attaining a positive response in non-metropolitan areas in comparison to metropolitan areas, for both groups. In 64.0 % of their applications the cisgender testers were invited to showing or additional contact in non-metropolitan areas, while only to 57.1 % in metropolitan areas. For the transgender testers 63.0 % of their applications resulted in a further contact or showing in non-metropolitan areas, and only to 53.8 %, in metropolitan areas.
In addition, both groups had higher shares of positive responses when the landlord was a company as compared to a private person. The cisgender testers attained an invitation to showing in 29.0 % of their applications when the landlord was a company, but only in 15.0 % when the landlords was a private person. The transgender testers’ applications resulted in showing in 19.6 % and 14.7 % of their applications, when the landlords was a company and private person, respectively. The largest difference of 9.4 percentage points, in share of invitation to showing, between the two groups were found in this given category, i.e. when the landlord was a company.
Even though the cisgender group attained slightly higher shares than the transgender group, the differences in call backs and positive responses are insignificant at this state. Therefore, no noticeable variations between the two groups can be found so far.
14 Table 1: Landlord responses (in percent) in terms of mean call back rate, and share of
positive responses, for the two different groups. The positive responses are divided into invitation to further contact and to showing, and invitation to showing. Number of actual cases are given in the parentheses.
Cisgender group (N=402) Transgender group (N=398)
Call back rate:
(per cent)
60.7 (244)
59.0 (235)
Inv to further contact/showing
Inv to showing Inv to further contact/showing
Inv to showing
Total:
(per cent)
60.2 (242)
17.2 (69)
57.8 (230)
15.3 (61) Metropolitan
area1 (per cent)
57.1 (128)
16.1 (36)
53.8 (121)
13.3 (30)
Non metro- politan area (per cent)
64.0 (114)
18.5 (33)
63.0 (109)
17.9 (31)
Company (per cent)
61.3 (38)
29.0 (18)
68.6 (35)
19.6 (10) Private person
(per cent)
60.0 (204)
15.0 (51)
56.2 (195)
14.7 (51) Landlord is a
male (per cent)
58.5 (114)
17.9 (32)
57.5 (103)
16.2 (29)
Landlord is a female (per cent)
62.1 (90)
11.0 (16)
54.8 (92)
13.1 (22)
1 Metropolitan area includes Stockholm, Gothenburg and Malmo
Moving forward to examining the different testers individually (Table 2), the cis woman and the trans woman had the largest call back rates of 65.3 % and 63.2 %, respectively, while the cis man and the trans man had the lowest rates of 56.2 % and 54.8 %.
Out of all four testers, the cis woman received the highest share of positive landlord responses. In 65.3 % of her applications she attained an invitation to showing or additional contact, and in 21.6 % of her applications she received an invitation to showing. This suggests that the cis woman is faced with least
15 difficulty in finding a rental apartment, when compared to the other testers. This is also in line with earlier findings suggesting a favoring of having a woman as a tenant over a man (see for example, Ahmed and Hammarstedt (2008), Bengtsson et al. (2012)). However, these findings were not directly applicable to the trans woman.
Even though the trans woman had the second largest call back rate of 63.2
%, close to the share of the cis woman, and was invited to additional contact or showing in 62.2 % of her applications, only 13.9 % of the given cases led to an invitation to showing. The trans man, on the other hand, had an invitation to showing rate of 16.8 % of his applications, and did thereby attain the second largest share of invitations to showing out of the four testers. Nonetheless, he was also the tester who received the smallest share of call backs, of 54.8 %, and positive responses, in the category: invitations to showing or further contact, of 53.3 %. By viewing the responses of the cis man, it is prominent that he was the tester attaining the least favorable result out of all four testers, with the second lowest call back rate of 56.2 %, as well as low response rates in both positive categories of 55.2 %, in invitation to additional contact or showing, and 12.8 % in invitation to showing. Out of all four testers the cis man was the tester whose applications led to the fewest showings.
Furthermore, all of the testers had a higher likelihood in receiving a positive response in non-metropolitan areas than metropolitan areas. The probability of attaining an invitation to showing was also higher when the landlord was a company compared to a private person. Out of all four testers, the trans woman had the largest share of positive responses of 78.3 %, in terms of attaining an invitation to additional contact or to showing, when the landlord was a company. When investigating the variation in positive responses depending on landlord gender, the share of applications that led to an invitation to showing was greater for the four testers if the landlord was male than if the landlord was female.
16 Table 2: Landlord responses (in percent) in terms of mean call back rate, and share of positive responses, for all four testers. The positive responses are divided into invitation to further contact and to showing, and invitation to showing. Number of actual cases are given in the parentheses.
Cisgender group (N=402) Transgender group (N=398)
Call back rate: 60.7 59.0 (per cent) (244) (235)
Cis woman (N=199)
Cis man (N=203)
Trans woman (N=201)
Trans man (N=197)
Call back rate: 65.3 56.2 63.2 54.8 (per cent) (130) (114) (127) (108)
Inv to
further contact/
showing
Inv to show- ing
Inv to further contact/
showing
Inv to show- ing
Inv to further contact/
showing
Inv to show- ing
Inv to further contact/
showing
Inv to show- ing
Total: 65.3 21.6 55.2 12.8 62.2 13.9 53.3 16.8 (per cent) (130) (43) (112) (26) (125) (28) (105) (33) Metro- 59.3 19.5 54.5 11.9 59.7 12.6 47.2 14.2 politan area1 (73) (24) (55) (12) (71) (15) (50) (15) (per cent)
Non metro- 75.0 25.0 55.9 13.7 65.9 15.9 60.4 19.8 politan area (57) (19) (57) (14) (54) (13) (55) (18) (per cent)
Company 70.0 30.0 53.1 28.1 78.3 17.4 60.7 21.4 (per cent) (21) (9) (17) (9) (18) (4) (17) (6) Private 64.5 20.1 55.6 9.9 60.1 13.5 52.1 16.0 person (109) (34) (95) (17) (107) (24) (88) (27) (per cent)
Landord 63.5 24.0 53.5 12.1 62.0 15.2 52.9 17.2 is a male (61) (23) (53) (12) (57) (14) (46) (15) (per cent)
Landlord 65.8 15.1 58.3 6.9 58.1 11.6 51.2 14.6 is a female (48) (11) (42) (5) (50) (10) (42) (12) (per cent)
1 Metropolitan area includes Stockholm, Gothenburg and Malmo.
17 Table 3 displays the joint p-values of the cis man, trans man, and trans woman.
By viewing the different p-values, no joint significance was found for the different testers, when examining the different control groups. This means that no evidence is found for collinearity between the different testers. Therefore, one can conclude that the randomization of the testers was conducted accordingly.
Table 3: Joint p-values of three of the testers
Variable Joint P-valuea
Apartment area 0.31
Rooms 0.49
Rent 0.17
Company 0.62
Metropolitan area 0.07
Landlord gender 0.51
a The p-value is for a F-test of equality between the cis man, the trans woman, and the trans man.
As the summary statistics presented in Table 1 and Table 2 showed, no larger variations between the two groups were found. Furthermore, when examining the different testers, the female testers and male testers, respectively, were seen to receive call backs very close in size. A discussion behind potential reasons to why the dataset took this form could be held.
Firstly, during the data collection it was apparent that some of the landlords needed clarification regarding the name change. This took the form of questioning what was ment by writing the two names at the end of the letter, addressing the tester in its former name as opposed to the current name, and asking if the tester was planning on living by him/herself. For further information regarding the share of applications which fell in under these forms of questions, see Appendices. Even if these responses only were a minority out of the total amount of responses for each tester, these are factors that could cause an understatement of the actual discrimination present in the housing market.
Therefore, as earlier stated, the category of getting an invitation to additional
18 contact or to showing could be too broad on its own in order to find signs of significant variation between the different testers and groups, respectively.
Therefore, by including the second category this potential problem was to some part accounted for. Nonetheless, an understatement of the actual discrimination could be caused if the identity signal was not clear enough.
Secondly, the discrimination the transgender people face might not be as apparent in correspondence studies, as discrimination against other minorities has proven to be (for example, see Carlsson and Rooth, 2007; Ahmed and Hammarstedt, 2008, 2009; Ahmed et al., 2010, 2013; Ahmed and Lång, 2017, 2019; Carlsson and Eriksson 2019; Bengtsson et al. 2012). Thereby, unequal treatment might be hard to come across when only the first stage of the renting process is examined.
These are factors that needs to be taken in to consideration as the analysis continues.
4.2 Estimations of the likelihood of getting invited to further contacts and to showings
In order to estimate how different explanatory variables, affect the likelihood of receiving a positive response, in terms of attaining an invitation to showing, or an invitation to additional contact or to a showing, a probit model was used. The model was constructed as follows:
Pr (yi=1|xi) = Φ(xiβ), (1)
The dependent variable, yi, was a dummy variable attaining a value of 1 if the applicant attained a positive response, and a value of 0 if the response was negative, such that it was declined by any potential reason. All applications who did not receive a response were coded as negative responses, as well. As earlier stated, the positive responses were divided into two categories, and the model was, therefore, run separately for each category to represent the positive responses. Furthermore, xi represented a vector of independent variables. The complete list of variables in the model, and their definitions, can be found in Table 4.
19 Table 4: Explanatory variables used in the estimations of the probability of getting an
invitation to additional contacts and being invited to showings
Variables Description
Dependent variables:
Getting an invitation to additional contacts or to showings
Takes the value of 1 if the person got an invitation to further contacts or to showings.
0 otherwise.
Getting an invitation to showings Takes the value of 1 if the person got invited to showing, 0 otherwise.
Independent variables:
Apartment area Apartment area, in square meters
Rooms Number of rooms in the apartment
Apartment rent Apartment rent, in SEK
Company Takes the value of 1 if the landlord is a
company, 0 otherwise.
Metropolitan area Takes the value of 1 if the apartment is situated in Stockholm, Gothenburg or Malmo, 0 otherwise.
Transgender Takes the value of 1 if the applicant changed
gender with the name change, 0 otherwise.
Cis man Takes the value if 1 if the gender identity is
cis man, 0 otherwise.
Trans woman Takes the value if 1 if the gender identity is
trans woman, 0 otherwise.
Trans man Takes the value if 1 if the gender identity is
trans man, 0 otherwise.
Table 5 and 6 display the probit estimates (marginal effects) of the likelihood of attaining a positive response from disclosing being transgender in the applications. Specification 1 shows the probit estimates when only the transgender variable and testers are included in the model, respectively.
Specification 2 displays the probit estimates when including control variables.
First, by examining the marginal effects in Table 5, the coefficients of the transgender variable were both negative in sign, which indicate a negative relationship between disclosing being transgender in the applications and receiving a positive response. Neither of these coefficients were, however,
20 significant. These results were prominent in both specification 1 and 2. By investigating the control variables, a positive relationship was found between apartment rent and receiving an invitation to further contact or to a showing. In addition, the likelihood of attaining a positive response was 11.8 % smaller in metropolitan areas than in non-metropolitan areas. These coefficients were, however, only significant (at the 1 % level of significance) when the positive response involved being invited to further contact or to showing. In the category invitation to showing, the probability of receiving a positive response was around 9.8 % larger, at the 5 % level of significance, if the landlord was a company than a private person. No significant relationship was found between apartment area and the likelihood of receiving a positive response. So far, no significant variation between the two groups are found, which is in line with the earlier results from the summary statistics in Table 1.
Table 5: Probit estimates (marginal effects) of the probability of being invited to further contacts and being invited to showings. Cisgender group works as reference group.
(1) (1) (2) (2)
VARIABLES Invitation to showing/
additional contact
Invitation to showing
Invitation to showing/
additional contact
Invitation to showing
Transgender -0.0241 -0.0184 -0.0217 -0.0163
(0.0348) (0.0261) (0.0347) (0.0260)
Apartment area - - -0.000651 -0.000268
(0.000796) (0.000598)
Rooms - - 0.00293** 0.00330***
(0.00115) (0.000893)
Apartment rent - - 0.0120** 0.00200
(0.00556) (0.00412)
Company - - 0.0356 0.0981**
(0.0520) (0.0453)
Metropolitan area - - -0.118***
(0.0430)
-0.0287 (0.0312)
Constant 0.602*** 0.172*** 0.579*** 0.175***
(0.0244) (0.0188) (0.0495) (0.0362)
Observations 800 800 800 800
R-squared 0.001 0.001 0.015 0.011
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
21 Moving forward to analyze the variations between the different testers of the study (Table 6). In specification 2 the cis man was found to have a penalty of 10.8 % and 9.0 % (at the 5 % level of significance), in the probability of receiving a positive response in the given categories, when compared to the cis woman. These results are in line with earlier findings presented in the summary statistics in Table 2, as well as previous literature (see for example, Ahmed and Hammarstedt (2008), Bengtsson et al. (2012)), that there exists discrimination based on gender in the rental housing market. A slightly smaller penalty (at the 5 % level of significance) was found in the given categories in specification 1, when no control variables were included in the model.
Furthermore, when investigating the other testers, the trans man was found to have a penalty of 12.6 % (at the 5 % level of significance) in the likelihood of getting invited to additional contact or to showing, when compared to the cis woman. However, no significant penalty was found in the likelihood of receiving an invitation to showing, when comparing the trans man and cis woman. Specification 1 provided similar results. The trans woman was, on the other hand, faced with an approximately 7.4 % smaller likelihood in receiving an invitation to showing (at the 10 % level of significance), when compared to the cis woman. No significant penalty in receiving an invitation to showing or additional contact was found when comparing these given testers. A penalty of 7.7 %, in terms of getting invited to showing, was found in specification 1 (at the 5 % level of significance), when only the tester variables were included in the model.
By examining the control variables, a positive relationship between apart- ment rent and receiving an invitation to additional contact or to showing was found at the 5 % level of significance. Furthermore, if the vacancy was located in a metropolitan area the probability of being invited to additional contact or to a showing was 12.5 % smaller, at the 1 % level of significance, than in a non- metropolitan area. In addition, if the landlord was a company the likelihood of receiving an invitation to a showing was found to be 9.6 % higher (at the 5 % level of significance) than if the landlord was a private person. At last, a positive relation between the number of rooms in the apartment and attaining a positive response, in both categories, was found at the 1 % level of significance.
22 Given these findings, one can conclude that the cis man, trans woman, and the trans man is faced with unequal treatment when compared to the cis woman;
and even though the results were the most apparent between the cis man and the cis woman, some indicators of having a preference for having the cis woman as a tenant over the trans woman and trans man were found.
Table 6: Probit estimates (marginal effects) of the probability of being invited to further contacts and being invited to showings. Cis woman works as reference group.
(1) (1) (2) (2)
VARIABLES Invitation to showing/
additional contact
Invitation to showing
Invitation to showing/
additional contact
Invitation to showing
Cis man -0.102** -0.0880** -0.108** -0.0904**
(0.0487) (0.0375) (0.0486) (0.0376)
Trans man -0.120** -0.0486 -0.126** -0.0491
(0.0491) (0.0396) (0.0488) (0.0396)
Trans woman -0.0314 -0.0768** -0.0279 -0.0744*
(0.0482) (0.0381) (0.0479) (0.0382)
Apartment area - - -0.000600 -0.000293
(0.000800) (0.000596)
Rooms - - 0.00346***
(0.00117)
-0.00286***
(0.000920)
Apartment rent - - 0.0116** 0.00156
(0.00557) (0.00409)
Company - - 0.0345 0.0961**
(0.0515) (0.0454)
Metropolitan area - - -0.125*** -0.0327
(0.0428) (0.0314)
Constant 0.653*** 0.216*** 0.638*** 0.227***
(0.0338) (0.0292) (0.0556) (0.0435)
Observations 800 800 800 800
R-squared 0.010 0.008 0.026 0.019
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
5 Summary and conclusion
This paper presents the first study to examine the discrimination against transgender in the rental housing market in Sweden, through the method of correspondence testing. In total 800 applications were sent out to various landlords advertising rental vacancies on Blocket.se in Sweden. To signal the gender identity of the four different testers (the cis man, the cis woman, the trans man, the trans woman) a name change was used in the experiment. The observations were categorized as receiving a response or not within the time
23 frame of one week. After this period of time the observations were classified as a no response. Furthermore, the responses were then noted as positive, if additional contact or an invitation to showing was suggested, or negative, if the application was declined for any potential reason. The primary results of the study showed no significant penalty in positive landlord responses for being transgender, when compared to the cisgender group. When examining the four different testers individually, the cis woman was found to be favored in terms of receiving less difficulty when applying for rental vacancies. This latter result is in line with earlier studies (see for example, Ahmed and Hammarstedt (2008), Bengtsson et al. (2012)), which have found that women are preferred as tenants.
Furthermore, some indicators of discrimination were found in terms of the transgender testers receiving penalties in positive landlord responses, when compared to the cis woman. However, as mentioned above, there is no clear evidence from this study that there is a favoring of having cisgender applicants than transgender applicants, when compared as groups. Nonetheless, one can reason that the cis man, trans man, and trans woman is faced with more difficulty when applying for apartments than the cis woman. Therefore, it is possible to conclude that there exists discrimination based on gender and gender identity in rental housing market.
Since the experiment was conducted in Sweden, the results of this study can provide a benchmark or lower bound to the discrimination present in the rental housing market. This can be reasoned, as Sweden is seen as one of the most progressive countries in terms of gender equality according to the Global Gender Gap Report 2020 (World Economic Forum, 2020). In addition, there is a possibility that the discrimination was understated, if the cisgender applicants received fewer call backs and share of positive responses due to their name change, in comparison to applicants that had not gone through a name change.
This scenario is however not as likely, since both the call back rate and positive response rate among the cisgender applicants presented itself on the common level, as the control groups found in earlier studies (see, Ahmed &
Hammarstedt, 2008, 2009; Carlsson & Eriksson 2014). However, another reason to why the discrimination could be understated is if the signal of the gender identity was too subtle. As earlier mentioned in the summary statistics,
24 there were responses asking for clarification regarding the name change. If the name change would have been more apparent, some of these responses might have been coded differently. The balance between keeping the signal obvious and subtle is, however, an important factor when performing a field study. If the signal is too obvious, the landlords might see through the experiment and be influenced to give answers which are or are not in line with their beliefs regarding the discrimination; while if the signal is too subtle, the risk of understating the discrimination becomes higher. These given results of the present study has proven to appear more subtle than correspondence studies investigating discrimination against other minorities of the HBTQ, such as discrimination against homosexual men (Ahmed and Hammarstedt, 2009). The reason behind this could be, as earlier discussed, due to the signal being more subtle than the one used when signaling sexual orientation; or it is due to the fact that the discrimination against transgender people appears differently in the renting process than for this given minority.
Furthermore, the discrimination found in the rental housing market could be the result of tastes of the landlords (Becker, 1957), or on the landlords’
statistical information regarding the groups (Phelps, 1972; Arrow, 1937).
However, as earlier stated, this study was bound to solely find the existence of discrimination. Therefore, supplementary research to the present study can be conducted in order to evaluate the given results further. At this point, there is a necessity to emphasize that the method used in the present study is only able to obtain the discrimination present in the first stage of the renting process. It is therefore, well needed, to stress that the discrimination might look different further along the process. In addition, as previously noted, the experiment is conducted on the Internet, which means that other search channels, such as newspapers, social contacts, and queuing systems, are not included in the results. However, there is no known reason to why the discrimination would differ between the channels. Nonetheless, additional research is needed here.
One way to proceed in the direction of the present study would be to attain a larger sample size in order to investigate the variation between the different testers further. Since the present study only included 800 observations in the study, a larger sample size could help to improve and strengthen the results.
25 Secondly, by involving additional information regarding income status and degree it would be possible to investigate if the discrimination in the rental housing market stems from tastes of the landlords or statistical information regarding the minority. To conclude, this study provides some new evidence of gender and gender identity discrimination in the rental housing market in Sweden. However, since the experimental studies that investigate discrimination against transgender are only few, and are in most cases limited by survey data, the need for further experimental studies is critical in order to truly highlight the struggles of the minority.
26
References
Ahmed, A. & Lång, E. (2019) “Victimized Twice: A Field Experiment on the Employability of Victims”. Victims and Offenders. 14, pp 859-874.
https://doi.org/10.1080/15564886.2019.1641449
Ahmed, A. & Lång, E. (2017) “The employability of ex-offenders: a field experiment in the Swedish labor market”. IZA Journal of Labor Policy. 6, pp. 1-23.
https://doi.org/10.1186/s40173-017-0084-2
Ahmed, A., Andersson, L. & Hammarstedt, M. (2013) ”Are Gay Men and Lesbians Discriminated against in the Hiring Process?”. Southern Economic Journal. 79, pp.
565-585. https://doi.org/10.4284/0038-4038-2011.317
Ahmed, A., Andersson, L. & Hammarstedt, M. (2010) “Can Discrimination in the Housing Market Be Reduced by Increasing the Information about the Applicants?”, Land Economics, 86: 79-90. doi:10.3368/le.86.1.79
Ahmed, A. & Hammarstedt, M (2009) “Detecting Discrimination against
Homosexuals: Evidence from a Field Experiment on the Internet”. Economica. 76, pp. 588-597 doi:10.1111/j.1468-0335.2008.00692.x
Ahmed, A., Andersson, L. & Hammarstedt, M. (2008) ”Are lesbians discriminated against in the rental housing market? Evidence from a correspondence testing experiment”. Journal of Housing Economics. 17, pp.234-238.
https://doi.org/10.1016/j.jhe.2008.06.003
Ahmed, A. & Hammarstedt, M (2008) “Discrimination in the rental housing market:
A field experiment on the Internet”. Journal of Urban Economics. 64, pp 362-372.
https://doi.org/10.1016/j.jue.2008.02.004
Arrow, K. (1973) “The theory of discrimination”. O. Ashenfelter, A. Rees (Eds.), Discrimination in Labor Markets, Princeton University Press.
Bengtsson, R., Iverman, E. & Tyrefors, B. (2012) “Gender and Ethnic Discrimination in the Rental Housing Market”. Applied Economics Letters. 19, pp. 1-5.
https://doi.org/10.1080/13504851.2011.564125
Carlsson, M. & Rooth, D.O. (2017). “Evidence of ethnic discrimination in the Swedish labor market using experimental data”. Labour Economics. 14, pp 716-729.
https://doi.org/10.1016/j.labeco.2007.05.001
Carlsson, M. & Eriksson, S. (2019) “Age discrimination in hiring decisions: Evidence from a field experiment in the labor market”. Labour Economics. 59, pp 173-183.
https://doi.org/10.1016/j.labeco.2019.03.002
