The Sharing Economy and Discrimination Evidence from a Field Experiment in Sweden
Nima Farrahi
Supervised by: Arizo Karimi 2NE940 Master Thesis in Economics
Department of Economics Uppsala University
June 7, 2019
The Sharing Economy and Discrimination Evidence from a Field Experiment in Sweden
∗Nima Farrahi
†Abstract
To investigate whether there is unequal treatment for ethnic minorities in the sharing economy this paper conducts a field experiment on Airbnb in Sweden.
The key findings report that inquiries from guests with Arabic-sounding names are 17 percentage points less likely to receive a booking invitation compared to guests with Swedish-sounding names. The discrimination is robust across host and listing characteristics. Furthermore, the results show that being associated with a lower social class decreases the probability of receiving a booking invi- tation for guests with Arabic-sounding names but not for guests with Swedish- sounding names, suggesting that the signal of social class is stronger for guests with Arabic-sounding names.
JEL: J15, JC93
Keywords: The Sharing Economy, Field Experiment, Audit Study, Ethnic Discrimination, Social Class
∗I wish to thank my supervisor Arizo Karimi for her valuable feedback and helpful com- ments.
†E-mail address: nimafarrahi1@gmail.com. The data used in the thesis is available from the author upon request.
Contents
1 Introduction 3
2 The Sharing Economy and Airbnb 5
3 Theoretical Framework 7
3.1 Taste Based Discrimination . . . 7
3.1.1 Discrimination due to Inefficient Management . . . 8
3.2 Statistical Discrimination . . . 8
3.3 Hosts Decision . . . 9
4 Literature Review 10 4.1 Discrimination in The Sharing Economy . . . 10
4.2 Social Class and Discrimination . . . 11
4.3 Discrimination in Sweden . . . 12
5 Methodology and Data 13 5.1 Experimental Design . . . 13
5.2 Data Collection & Sample . . . 16
5.3 Limitations of the Experiment . . . 18
5.3.1 Signaling Ethnicity Through Names . . . 18
5.3.2 Paired vs Nonpaired Audits . . . 18
5.3.3 Ethical Issues . . . 18
6 Results 19
7 Conclusion 26
8 References 29
9 Appendix 32
1 Introduction
Housing market discrimination, in the form of unequal treatment, could lead to economic inefficiencies and segregation. Consequently, this issue is of central importance for policy. In contrast to the traditional housing market, the sharing economy provides an online market for housing rentals, allowing hosts to enter the market without any requirements. In recent years, the sharing economy has grown and therefore the interest of studying discrimination in the area has in- creased. Legislation designed to prevent discrimination in the housing market is unlikely to reach out to the smaller landlords using the online market for hous- ing rentals, and unlike the hotel market, the sharing economy allows landlords to choose to whom they want to rent. Studies investigating ethnic discrimina- tion in the sharing economy is mainly based on the platform of Airbnb. The topic of ethnic discrimination on Airbnb received attention after Edelman et al.
(2016) presented results from a field experiment which showed that individu- als with White sounding names have it easier to get accommodation through Airbnb than individuals with African American sounding names.1 Shortly after the revealing of discrimination, Airbnb made several changes in its terms and policies; they presented a new anti-discrimination policy and published a report on the problem of discrimination on the platform.2
Airbnb provides an online market place for rental housing where property own- ers, hosts, rent out their properties. Each host set a price and availability for their listing. The hosts’ profile pages provide information about themselves and their listings. Potential guests book properties at certain dates at the price set by the host. The design of the platform allows hosts to observe guests before an inquiry is accepted. However, the new anti-discrimination policy limits host availability to screen potential guests. For instance, the guest’s profile photo is not shown to the host until the booking is confirmed. The new settings make all guest profiles on Airbnb more similar to each other than before.
1The paper paved the way for other studies on Airbnb. For Swedish evidence see Bethen- court and Farrahi (2017).
2blog.atairbnb.com/wp-content/uploads/2016/09/REPORT_
Airbnbs-Work-to-Fight-Discrimination-and-Build-Inclusion_09292016.pdf
The aim of this study is to add knowledge to the existing discrimination research by investigating the effect ethnicity and social class has on the short-term rental market outcome. In order to test for discrimination, this paper presents a field experiment where 1006 inquiries were sent to hosts on Airbnb in Sweden from guest profiles combining Swedish-sounding names and Arabic-sounding names with higher and lower social class. There are several reasons for the focus on individuals with Arabic-sounding names in this paper. Previous studies on the Swedish housing market and labor market, such as Arai and Skogman Thoursie (2009) Carlsson and Rooth (2007) and Ahmed and Hammarstedt (2008), show that Swedish-sounding names get a higher call-back rate then Arabic-sounding names given equivalent attributes. Moreover, this paper also studies the inter- action between ethnicity and social class. Bertrand and Mullainathan (2004) argue that one of the main concerns of experiments that use names to signal eth- nicity is that the names can also be reflecting the individual’s social background.
In a field experiment in the U.S. housing market, Hanson and Hawley (2011) find that the discrimination against African American names decreases when the inquiry to a landlord signals a higher social class. Carlsson et al. (2018) iden- tify the neighborhood signaling effect on the hiring decision and find that the strength of the neighborhood signal, in the Swedish labor market, is stronger for applicants with Middle-Eastern names then applicants with Swedish-sounding names. To highlight the reason for discrimination, this paper takes advantage of the features of Airbnb by collecting variables for observable data for each listing and host.
This paper makes two contributions to the existing literature on ethnic discrim- ination. First, this experiment studies new data for a time when Airbnb has implemented a series of new anti-discrimination policies. In contrast to previ- ous experiments made on Airbnb, a host now has limited accessibility to screen a potential guest before the booking is confirmed. As the experimental guest profiles are relatively bare, the new settings potentially solve for one concern of the previous experiments made on Airbnb. Hence, guest profiles used in this experiment are relatively more common than guest profiles used in previous ex- periments. Second, by differing the guest’s profiles in terms of living area, either
signaling living in an affluent municipality or in a deprived municipality, and by the writing style, spelling and grammar of the inquiries, this paper introduces the effect of social class on discrimination. To the best of my knowledge, this paper is the first to test how interaction between race and social class affects the outcome in the Swedish housing market.
This paper finds widespread discrimination against guests with Arabic-sounding names. The overall results show a 17 percentage points penalty associated with having an Arabic-sounding name. These results are in line with previous studies and confirm lower call-back probabilities for individuals with Arabic names in the Swedish housing market. Moreover, the results suggest that more active hosts with experience and history on Airbnb discriminate less. The paper finds no effect from social class for guests with Swedish-sounding names. However, guests with Arabic-sounding names face an extra penalty of 15 percentage points when signaling a low social class as opposed to a high. These results indicate that the strength of the signal of social class is stronger for ethnic minorities.
The remainder of the paper is organized as follows. The next section introduces a general overview of the sharing economy and Airbnb. The third section presents the relevant theoretical framework. The fourth section describes the existing literature on the topic. In the fifth section, I describe the experimental design and data, and the results are reported in the sixth section. The paper ends with a conclusion followed by an appendix.
2 The Sharing Economy and Airbnb
Throughout history, individuals have been sharing assets with each other. With the emergence of new technologies, such as the Internet, new platforms have made asset sharing between individuals easier. The sharing economy is based on the idea of an economic system where individuals share underused assets or services, for free or for a cost (Botsman, 2016). The concept has developed through digitalization which made it possible to reduce the role of a third-party and lowers the transaction cost. The sharing economy has led to improved mar-
ket efficiency as it unlocks hidden wealth and matches an individual’s needs and haves. Founded in 2008, Airbnb is a sharing economy marketplace and pro- vides an online market for home sharing. Property owners, hosts, set a price and availability for their listing in which potential guests can book the property. For this service, Airbnb charges a service fee, calculated from the booking price, of 3% for hosts and up to 20% for guests.3 The hosts’ profile pages, fully observed by potential guests, provide information about themselves and their listings.
Guests and hosts on Airbnb have the freedom to choose who they want to rent from and rent to. Hence, Airbnb creates opportunities for both hosts and guests to choose/reject potential hosts and guests on the basis of individual preferences.
Discrimination on Airbnb has recently been raised as an issue on the platform, and as a result, they introduced a new anti-discrimination policy.4 All users on Airbnb became required to sign and adopt the new policies which address discrimination based on ethnicity, gender, religion and geographic origin. To deal with discrimination, the new anti-discrimination policy limited the host’s availability to screen a potential guest. For instance, the guests profile photo is no longer shown to the hosts until the booking is confirmed.5 The platform design of Airbnb still allows hosts to observe a potential guest’s name, living municipality, level of verification and year of membership before the booking is confirmed. After the implementation of the new anti-discrimination policy, Airbnb encourages guests to report any concerns regarding discrimination, for instance, if a host cancels a booking after seeing the guest’s profile picture the host risks suspension or banning from the site. Furthermore, Airbnb encourages the use of “instant booking”, which allows guests to book listings immediately without any approval from the host. Consequently, it is of interest to investigate the existence of discrimination on Airbnb after the introduction of the new anti- discrimination policy.
3https://www.airbnb.com/help/article/1857/what-is-the-airbnb-service-fee
4https://www.airbnb.com/help/article/1405/airbnb-s-nondiscrimination-policy--\
\noindentour-commitment-to-inclusion-and-respect?locale=en
5https://press.airbnb.com/update-on-profile-photos/
3 Theoretical Framework
The standard definition of economic discrimination is that individuals of a mi- nority are treated less favorably than members of a majority group, with oth- erwise identical characteristics (Bertrand and Duflo, 2016). This section will present the two main theories of discrimination, taste-based discrimination and statistical discrimination, which can explain the possible discrimination in this experiment.6
3.1 Taste Based Discrimination
The first economic model of discrimination was introduced by Becker (1957).
The concept was based on taste discrimination and was initially applied on white and black workers in the U.S. labor market.7 The theory describes taste- based discrimination as prejudice against a particular group, in which employers get disutility from hiring a minority worker. To compensate employers for the disutility, the non-preferred workers must either be more productive at a given wage or accept a lower wage for equal productivity. According to the theory, if an employer is prejudiced against a non-preferred group, the employer will act as if hiring non-preferred workers costs wb+ de. Were wb denotes the wage rate for the minority worker belonging to group b and the additional cost, d, represents the so-called discrimination coefficient and its size depends on the strength of the employer’s taste for discrimination. Hence, the discriminating employer (de> 0) will only hire non-preferred, b-group members, over preferred, w-group members, if:
Wb− de> Ww
The differences in wages between workers of different groups will ultimately create an incentive for segregation as it will be more efficient for minority workers to work in their own business where the employers do not have a taste for discrimination. Over time, Becker’s model has been applied to other markets than the labor market. Correspondingly, the consequences of discrimination also
6For implicit discrimination see Bertrand et al. (2005).
7Taste-based discrimination can also occur from distaste against a minority group by em- ployees, customers or governments (Becker, 1957).
have an effect on racial segregation in the housing market. In this experiment, taste-based discrimination will be demonstrated if hosts on Airbnb feel that it occurs an extra cost from accepting a guest from the non-preferred group. For instance, if hosts on Airbnb get disutility from renting out their accommodation to guests with Arabic ethnicity the discrimination will be based on taste.
3.1.1 Discrimination due to Inefficient Management
In the terminology of Becker’s theory of discrimination, it is costly for indi- viduals to discriminate because it prioritizes characteristics that are irrelevant to productivity. In equilibrium, only non-discriminating employers will make profits, discriminating employers must fund the cost for their distaste and in a competitive market, they will be competed out from the market. A cen- tral prediction of Becker’s theory is that product market competition reduces inefficient management in general and discriminatory behavior in particular.
One common interpretation is that competition drives discrimination out of the market, another interpretation is that inefficient management is associated with discrimination. In this experiment, this theory can be tested by investigating the relationship between host efficiency and discrimination. If discrimination reduces among hosts with certain characteristics, associated with efficient man- agement, parts of the discrimination may be driven by inefficient management.
3.2 Statistical Discrimination
Discrimination might occur without any prejudice against a particular group, Arrow (1973) and Phelps (1972) introduced an information-based statistical theory for discrimination. The theory was first applied to the labor market and describes statistical discrimination as discrimination arising from limited infor- mation. With limited information, employers will use observable characteristics such as race, ethnicity or gender to assume workers expected productivity. The observable characteristics carry information about the average productivity and skills of a person belonging to a particular group. Hence, the theory predicts that discrimination depends not only on the individuals characteristic and pro- ductivity but also on the average of individuals within the same group.
In this experiment, hosts respond to guest inquiries with equal observable char- acteristics, but with names or social class signaling different group affiliation.
Under statistical discrimination, hosts have imperfect information about the potential guest and will make their decision based on accessible statistics. Both ethnicity and social class might give rise to statistical discrimination if hosts on Airbnb associate this information with the general behavior and performance of the group. Statistical discrimination will be demonstrated if hosts discriminate based on general negative stereotypes that the guest’s name, social class or the combination of them, are signaling.
3.3 Hosts Decision
To relate the presented theories with this experiment, I will provide a model illustrating the hosts decision.8 Hosts responds to guest inquiries with equal ob- servable characteristics, but with names or social class signaling different group affiliation. Assume that hosts, hn, evaluate guest’s, gn, inquiries and profiles and assign a score value, V . The score is based on observable characteristics, X, the expected value of unobservable group-related characteristics, Z, and guests group affiliation, G ∈S, A .9 Next, the hosts decision is based on the evalu- ated score value, V : X, G, E[Z]. Guests with a score high enough will get accepted by hosts. The observable characteristics includes name, level of verifi- cation, membership year and municipality of the potential guest. Except for the group affiliation, the hosts observe equal characteristics for the guests, hence, on average: X¯s = ¯Xa = ¯X. The differences in the hosts average evaluation gap between the groups, ∆ ¯V , is defined as the unequal treatment.10 Potential discrimination against guests with Arabic-sounding names in this experiment arise if:
∆ ¯V =n ¯V X, G = S, E[Z¯ s] − ¯V X, G = A, E[Z¯ a]o
> 0
A positive ∆ ¯V represents the scenario were Swedish-sounding guests are chosen
8This section is inspired by Arai et. al (2008), the paper provide an illustrative model on employers hiring decision.
9The reader can think of the groups as one group, S, representing guests with Swedish- sounding names and the other group, A, representing guests with Arabic-sounding names.
Similarly, this could be illustrated for groups affiliated with high and low social class.
10Where: ∆ ¯V = ( ¯V1, ¯V2, ..., ¯Vn) is the average difference in evaluated score value across n-number of hosts.
over Arabic-sounding guests, for this to be true at least one of the following two conditions must be fulfilled:
I. E[Zs] = E[Za] = ¯Ω and ∆ ¯V =n ¯V X, G = S, ¯¯ Ω − ¯V X, G = A, ¯¯ Ω]o
> 0
II. ∆ ¯V =n ¯V X, G = S = v, E[Z¯ s] − ¯V X, G = A = v, E[Z¯ a]o
> 0
In the first scenario, hosts value S higher than A, and hence prefers S to A, given equal observable characteristics, ¯X and expected unobservable characteristics, Ω. This is the case which is referred to as taste-based discrimination by Becker¯ (1957). The second condition refers to the case were the hosts discriminates based on statistics about group averages, Z.
An important question for policymakers is whether the discrimination is taste- based or statistical. While both taste-based and statistical discrimination yield disadvantage for the non-preferred group, the underlying reasons for discrimina- tion differ. However, it is difficult to specify the theory behind discrimination.
Along with most of the existing literature, presented in the next section, this paper is unable to make a definitive claim about the source of discrimination.
4 Literature Review
The following section presents the relevant literature within the field of discrim- ination. In particular, this section covers a literature review of ethnic discrimi- nation in audit or correspondence studies. The review, presented below, mainly focuses on three areas: studies regarding discrimination on Airbnb, studies in- volving discrimination and social background and studies regarding discrimina- tion against individuals with Arabic-sounding names in Sweden.
4.1 Discrimination in The Sharing Economy
Recently, a growing body of evidence has reported the persistence of racial discrimination in the sharing economy, particularly on Airbnb (Edelman et al.
(2016); Bethencourt and Farrahi (2017); Fisman and Luca (2016)).11 In a field
11For discrimination against hosts on Airbnb see Edelman and Luca (2014) and Kakar et al. (2017).
experiments on Airbnb, Edelman et al. (2016) find that guest profiles with African American-sounding names are 16 % less likely to be accepted by hosts compared to identical guest profiles with White-sounding names. The experi- ment inquires about the availability of around 6,400 listings on Airbnb across five cities in the U.S.. Furthermore, the paper suggests that both White and African American hosts discriminate against African American guests. Simi- larly, Bethencourt and Farrahi (2017) conduct a field experiment in the Swedish Airbnb-market and finds that the probability to get accepted is 21,5 % lower for guests with Arabic-sounding names compared to guests with Swedish-sounding names. The experiment was limited to three Swedish cities and the results showed that that the discrimination against guests with Arabic-sounding names decreases when the host has an Arabic-sounding name. Edelman et al. (2016) and Bethencourt and Farrahi (2017) studied discrimination on Airbnb in a time when it was required for a guest to upload a profile photo, hence, the hosts had more opportunities to screen a guest profile before accepting a request.
4.2 Social Class and Discrimination
Most studies that test for the effect of social background on the discrimination have used different neighborhoods to signal social class. In a field experiment in the U.S. labor market, Bertrand and Mullainathan (2004) use different names to measure ethnic discrimination and find that White-sounding names receive 50%
more callbacks for interviews than African American-sounding names. The au- thors argue that the experimental names, besides signaling race, may also signal social background. Consequently, the paper studies the effect of the neighbor- hood of residence on the likelihood of a callback. The paper finds that both White and African American job applicants have higher callback rates when living in a better neighborhood - associated with a higher fraction of white res- idents, level of education and average income. However, even though a better neighborhood increases the overall callback rates, the paper does not find that it decreases discrimination.
In a field experiment in the U.S. rental housing market, Hanson and Haw- ley (2011) test for discrimination by sending emails to landlords which differs
in both the home-seekers’ race and social class. The paper defines two social classes that differ in the e-mail sent to the landlord; The low-class e-mail contains spelling errors and informal or grammatically incorrect sentences. The paper finds discrimination against African Americans, however, when both groups are sending e-mails to landlords that signal a higher social class the discrimination decreases. Hence, the results indicate that social class can be one explanation for the discrimination against minorities.
To investigate the neighborhood signaling in the Swedish labor market, Carlsson et al. (2018) conduct a field experiment and send job applications to employ- ers with job vacancies. The paper categorizes an affluent neighborhood as an area were the relatively wealthy and residents with Swedish background tend to live, moreover, neighborhoods are associated with higher employment rates and incomes. The paper finds no evidence of a neighborhood signaling effect for job applicants with Swedish-sounding names. However, the results show that applicants with Arabic-sounding names have a 42% lower probability of receiv- ing callbacks if they live in a deprived neighborhood rather than in an affluent neighborhood.
4.3 Discrimination in Sweden
Regarding racial discrimination in Sweden, the existing literature shows that individuals with Arabic-sounding names face widespread discrimination in the Swedish housing and labor market. Several studies, such as Bursell (2007) Carls- son and Rooth (2007), use correspondence testing and finds that job applicants with Swedish-sounding names have a higher probability of receiving a call back for a job interview compared to job applicants with an Arabic-sounding name.
In a field experiment on the Swedish rental housing market, Ahmed and Ham- marstedt (2008) apply for vacant rental apartments from fictitious individuals different names. To test for discrimination, the paper conducts a field experi- ment on the Internet, more specifically, the paper use Blocket.se as a research platform. The result suggests that a male applicant with an Arabic-sounding name receives fewer call-backs and invitations when applying to apartment ads.
Carlsson and Eriksson (2014) find similar results regarding ethnic discrimination
in the Swedish rental market for apartments. The paper concludes that appli- cants with Arabic-sounding names face a penalty equally big as individuals that enjoy to party and smoke.
5 Methodology and Data
In this section, the main features of the experimental approach and the data collection is presented. Firstly, the experimental design and the model are presented followed by a description of the collected data and the limitations of the experiment.
5.1 Experimental Design
The structure and design of this experiment follow Edelman et al. (2016) and adopt a similar setting and methodology to investigate the existence of dis- crimination on Airbnb in Sweden after the implementation of the new anti- discrimination policy. The first step of the experimental design is to create twenty identical guest accounts on Airbnb which only differ by names.12 The names used in the experiment are obtained from Arai et al. (2008) and are pre- sented in table 1. The table lists four groups of guest names; Swedish sounding male names, Swedish sounding female names, Arabic sounding male names and Arabic sounding female names. For further validation, a survey was created in which people across Sweden categorized the names.13 All guest accounts used in the experiment are verified by email addresses and phone numbers.
As previously mentioned, one concern of experiments using names to signal eth- nicity is that the same names can also be associated with social class. This issue might lead to overestimation of the ethnic discrimination if the hosts discrimi- nate against the social class conveyed by the experimental names. To deal with this concern, the experiment introduces a notion of social class. To signal social class, this experiment uses two types of inquiries, one that signals high social
12Heckman (1998) critically discuss field experiments investigating discrimination and ar- gues that, to prevent bias, both types of group affiliation have to be identical in all terms except for the race.
13The platform used for the survey is Google Forms, table A1 in appendix present the survey results which confirms the signaling of different ethnicity of the names.
Table 1: List of names used in the experiment First name Last name First name Last name
Fateme Ahmed Kamal Ahmadi
Nasrin Hassan Abdallah Mohammed
Halima Mohammadi Islam Hashemi
Aïcha Abdallah Abdelaziz Hussein
Fatima Ahmad Abdelhakim Hassan
Sara Andersson Jonas Söderström
Marie Björkvist Erik Östberg
Johanna Gustafsson Johan Nyström
Karolina Svensson Mikael Andersson
Malin Wallin Martin Berggren
class and one that signals low social class. In Sweden, the relatively poor and those with foreign backgrounds tend to live in poorer areas that are associated with lower employment rates, educational level and average incomes (Andersson and Bråmå, 2004; Aldén and Hammarstedt, 2016). Inspired by Carlsson et al.
(2018), I define high-class types of inquiries as those signaling an affluent munic- ipality associated with high average income, education level, low unemployment rate and low level of foreign-born residents.14 To strengthen the signal of social class I take inspiration from Hanson and Hawley (2011) and also differ the types of inquiries in terms of message writing style, spelling and grammar. Accord- ingly, the high and low social class types of inquiries differ in two ways: one by writing-style which is signaled directly in the inquiry sent to the host, and another by living area which is signaled on the guest profile. Table 2 presents the types of inquiries that were randomly assigned to hosts.
Table 2: Types of inquiries
Swedish-sounding Arabic-sounding
High-class Type 1 Type 2
Low-class Type 3 Type 4
The two types of inquiries in this experiment have the same structure and are sent to hosts with a greeting phrase, a question of availability and ending with a
14The categorization is based on data from Statistic Sweden for Sweden’s municipalities.
Examples of a high-class municipality are Danderyd and Lidingö. Examples of a low-class municipality are Botkyrka and Södertälje.
signature with a name. All accounts were created on the same day and had the same level of verification. To get a deeper understanding of how the inquiries are observed from the host’s perspective, I created a host profile on Airbnb and listed one property. Figures A1 and A2 in the appendix demonstrate how inquiries and guest profiles are observed by hosts. The messages seen in the figures are the same ones as used in the experiment. The effect of ethnicity is isolated when comparing inquiries with the same social class which only differ by guest name.15
Inquiries were randomly sent from the guest accounts to hosts across Sweden asking for availability of their listings. To make sure that the inquiries were randomly assigned across host’s and listing characteristics a random number generator was used to match each listing with the guest profiles and the so- cial class types. The inquiries were sent five weeks in advance asking for the availability of the listing during a specific weekend, the search was limited to those listings that were scheduled as available during that particular weekend.
Furthermore, the experiment was limited to hosts that do not use the function
“Instant booking” on Airbnb.16
According to Airbnb the vast majority of hosts reply within 12 hours.17 In sim- ilarity to Edelman et al. (2016) the host’s responses were listed into different categories. “No response” (if the host did not respond within 3 weeks), “No” (if the host rejected the inquiry, no matter the reason), “Yes” (if the host invited guest with a booking invitation), “Request for more information” (if the host did not give a clear answer but instead asked for more information).
In the analyses, a positive response is restricted to “Yes”, as it is the only response that includes a booking invitation. To test for discrimination this paper is estimating the probability of getting a positive response from hosts for the different ethnicities and social classes. To do this, I run the following linear probability model:
15This is done by comparing Type 1 with Type 2 inquiries, or Type 3 with Type 4 inquiries.
16Instant Book listings don’t require approval from the host before they can be booked.
17https://www.airbnb.com/help/article/75/how-much-time-does-a-host-have-to-respond- to-my-reservation-request
P r(Respond = Y es | Ethnicity, Social Class, X) = β0+ β1Ethnicityg +β2Social Classg+ β3(Ethnicityg× SocialClassg) + β3Xhl+ εhl
where response equals 1 if yes (booking invitation); Ethnicity is a dummy vari- able taking the value one if guest, g, has an Arabic-sounding name; Social class is a dummy variable taking the value one if guest, g, signals lower social class; X is a vector of control variables (gender, city, area, price, Superhost status, number of reviews, number of listings, room type, cancellation policy and membership year), for host h and listing l.
5.2 Data Collection & Sample
In order to examine the existence of discrimination on Airbnb, a field experiment was conducted where 1006 number of inquiries were sent to hosts in Sweden from guest profiles combining Swedish-sounding names and Arabic-sounding names with higher and lower social class.18 The experiment was carried out between 1-12 April, 2019, and collected data for listed properties on Airbnb.
Each host’s profile page provided a serial of variables of interest regarding the host and the listed property. To be able to analyze the hosts and listings char- acteristics in detail, I recorded data from each host’s profile page. The name and profile picture of the host is signaling a gender, which was categorized as either male, female, couple or unknown. To be able to measure host efficiency, I collected data for Superhost status. To get Superhost status hosts have to meet some requirements, such as maintaining an overall rating of at least 4.8 out of 5, a review rate of at least 50 % and a response rate of at least 90%.19 These variables all indicate how efficient hosts across Airbnb run their rental business, making Superhost status a good indicator of efficiency across hosts. Addition- ally, the profile page for each host provides data for number of reviews, year of membership, number of listings and answer frequency. This data indicate to
18Based on a power analysis, the initial goal was to send around 1800 inquires to hosts, however Airbnb made it impossible to continue to contact hosts as they started to block the guest accounts from sending inquiries to hosts.
19https://www.airbnb.ie/help/article/829/how-do-i-become-a-superhost
what extent the host uses the platform, allowing me to study the relationship between discrimination and host activity.
The profile page also provided information about the listing. Collection of data for each listings price per night for the weekend in question, number of rooms, room type and cancellation policy was made. To measure the relationship be- tween the level of interaction between the guest and host and discrimination, data for the room type was collected. All Airbnb listings are categorized into the following 3 different types of room: entire place, shared room or private room in a unit. In this paper, all listings that were not categorized as an entire unit was recorded as shared property as some of the living space was shared with the host. To analyze whether discrimination varies across geographical characteris- tics, data for the specific location of the listing was collected. Next, I used avail- able data provided by Swedish Association of Local Authorities and Regions, SALAR, to categorize each listing into either a metropolitan city, medium-sized town or rural area.20 Table 3 displays summary statistics for the sample includ- ing mean, standard deviation, minimum and maximum values for the recorded variables for hosts and listings.
Table 3: Summary Statistics of Sample
Variables Mean SD Min Max Observations
Host is female 0.57 0.49 0 1 1006
Host is male 0.38 0.49 0 1 1006
Host gender is unkown 0.02 0.13 0 1 1006
Host is a couple 0.03 0.17 0 1 1006
Property is shared 0.39 0.49 0 1 1006
Host has Superhost status 0.33 0.47 0 1 1006
Metropolitan cities 0.61 0.49 0 1 1006
Medium-sized town 0.18 0.38 0 1 1006
Rural municipalities 0.22 0.41 0 1 1006
Price per night (SEK) 765.03 490.51 100 4174 1006
Number of reviews 34.30 50.98 0 703 1006
Number of listings 1.31 0.90 1 19 1006
Number of bedrooms 1.36 0.78 1 7 1006
Note: This table reports summary statistics for hosts and listings characteristics. The first nine rows are dummy variables equal 1 for each observation if true.
20The classification of municipalities produced by SALAR divides the municipalities into three main groups A, B and C. Municipalities are categorized in A if it is, or are near to, one of the three metropolitan cities in Sweden (Stockholm, Goteborg or Malmö). The municipality is categorized as B if the municipality is, or are near to, a medium-sized town with at least 50 000 inhabitants. The classification categorizes rural municipalities in group C.
5.3 Limitations of the Experiment
5.3.1 Signaling Ethnicity Through Names
As discussed in Bertrand and Mullainathan (2004), a concern for all experiments using names is that the names do not directly report a race, ethnicity or gender.
Since the profiles in this experiment do not tell the actual ethnicity directly but instead signaling ethnicity through the names, it assumes that the hosts have the knowledge of the ethnic content of the names. Hosts might not notice the names or not recognize the ethnic content of the name. To minimize the risk of using names that are not signaling ethnicity this experiment is using a big variation of names, which have all been validated through a survey.
5.3.2 Paired vs Nonpaired Audits
This field experiment is constructed on the mechanism of non-paired audit de- sign, meaning that each host in the experiment received one inquiry. Paired audit design has statistical advantages as it can quantify the answer for each specific host. Although, paired audit designs are more often used by field exper- iments made on discrimination in Sweden it was not possible for this experiment due to two main reasons. Firstly, a paired design would force variation of in- quiries which might have influenced the findings of discrimination. Secondly, a problem regarding the time between the pair of inquiries would arise: too little time would give rise to suspicion, and too much time would give a first-mover advantage. Furthermore, research suggests that the paired audits are equally statistical efficient than non-paired audit design (Vuolo, Uggen, and Lageson, 2016).
5.3.3 Ethical Issues
Conducting a field experiment and observing people’s behavior, without them having the knowledge of it has its ethical considerations. To minimize the burden for each host I connected all the accounts to my personal cellphone, whereby I politely replied to each hosts response telling them that I am no longer interested in renting the property. Furthermore, each host only got contacted once no matter how many properties they had listed.
6 Results
In this section, the main experimental results are presented. Based on signaled ethnicity, social class and gender the analysis is using three forms of treatment of the guest profiles. If observed characteristics differ systematically by treat- ment status, the results might be biased. It is therefore useful to verify that the treatment assignment is not correlated with hosts and listings characteristics.
Table 4 examines this by reporting the results of regressions of the probability of being randomized into each group on host and listing characteristics. As ex- pected, given the randomization, the table reports that there is no correlation between any of the host and listing characteristics and treatment assignment.
Hence, the table is confirming that randomization has succeeded in generating balance which solves for possible selection bias.21
Table 5 reports the main result for the probability of receiving a booking invi- tation depending on the guest ethnicity, the table includes geographical char- acteristics for the listings and the interaction of those characteristics with the guests signaled ethnicity. Column 1 shows that guests with Swedish-sounding names are accepted 66 percent of the time, while guests with Arabic-sounding names are accepted 49 percent of the time. Thus, there is a 17 percentage point difference in the likelihood of receiving a booking invitation between guests with Arabic-sounding and Swedish-sounding names, the estimate is statically signifi- cant at less than 1% level. The overall discrimination coefficient lies within the same confidence interval as the one in Bethencourt and Farrahi (2017). Hence, there are no significant differences in discrimination when comparing the esti- mate from this experiment to a similar experiment made before the implemen- tation of the policy. This effect stays more or less constant at 16-20 percentage points when controlling for urban/rural region of residence (columns 2 and 3).
Hence, columns 2 and 3 show that the discrimination is robust across geograph- ical characteristics, suggesting that the gap in booking invitation exists in both metropolitan cities and rural areas of Sweden.
21For further validation, Table 2A in appendix reports the number of inquiries sent from each treatment group to hosts and listings characteristics in detail.
Table 4: Randomization Verification
(Ethnicity=1) (Social Class=1) (Gender=1) Arabic name Low social class Female name Geographical characteristics:
Metropolitan cities -0.028 -0.16 -0.051
(0.190) (0.223) (0.245)
Rural municipalities 0.056 -0.0026 -0.082
(0.195) (0.168) (0.195)
Hosts & listings characteristics:
Gender of host 0.016 0.030 -0.023
(0.054) (0.040) (0.050)
Price per night 0.048 0.12 -0.043
(0.060) (0.068) (0.065)
Room type -0.036 0.100 -0.029
(0.110) (0.092) (0.117)
Cancellation policy -0.039 -0.046 -0.075
(0.042) (0.058) (0.057)
Number of reviews -0.0002 -0.0007 -0.0018
(0.001) (0.001) (0.001)
Year of membership 0.0023 -0.0008 -0.0052
(0.019) (0.031) (0.049)
Number of listings 0.020 0.010 0.025
(0.047) (0.045) (0.037)
Number of bedrooms 0.066 -0.004 0.034
(0.049) (0.063) (0.056)
Observations 1006 1006 1006
Note: This table reports marginal effects from Probit regressions and checks the balance between the group affiliations. Regressions using OLS specification has been made and it produce essentially similar results.
Standard errors in parentheses, adjusted for clustering by county.
∗p < 0.10,∗∗p < 0.05,∗∗∗p < 0.01.
Table 5: Ethnic Discrimination & Geographical Characteristics Dependent variable: 1(host accepts)
(1) (2) (3)
Guest has arabic name -0.17∗∗∗ -0.20∗∗∗ -0.16∗∗∗
(0.026) (0.046) (0.026) Metropolitan cities -0.079
(0.054) Guest has arabic name 0.034
× Metropolitan cities (0.058)
Rural municipalities 0.14∗∗∗
(0.040)
Guest has arabic name -0.051
× Rural municipalities (0.062)
Constant 0.66∗∗∗ 0.71∗∗∗ 0.63∗∗∗
(0.033) (0.050) (0.028)
Observations 1006 1006 1006
R2 0.031 0.035 0.040
The table reports results from regressions of the probability of receiving a booking invitation on the guests ethnicity including variables for geographical characteristics and interaction of those characteristics with the guests ethnicity. Standard errors in parentheses, adjusted for clustering by county.∗p < 0.10,∗∗p < 0.05,∗∗∗p < 0.01.
Table 6 reports results from regressions of the probability of receiving a booking invitation from a host on the guests signaled ethnicity including variables for hosts and listings characteristics and the interaction of those characteristics with the guests signaled ethnicity. Just as the discrimination does not vary across geographical characteristics of the listing, table 6 shows that the discrimination is robust across hosts and listings characteristics. However, columns 5 and 6 reports significant positive interaction effects. Column 5 shows that the gap between guests with Swedish-sounding names and guests with Arabic-sounding names decreases if the host has more than 10 reviews. This result indicates that more active hosts with experience and history on Airbnb discriminate less.
There are different interpretations for this result: more active and experience hosts might be more interested in renting out their property, another view is that hosts have had prior experience of guests with Arabic-sounding names.
Similarly, Column 6 shows that discrimination decreases when the host has Su- perhost status. This result indicates that efficient hosts discriminate less. The result can be explained by Becker (1957), suggesting that discrimination comes at a cost for the host and that efficient, profit-maximizing, hosts would discrim- inate less.
Table 6: Are Ethnic Discrimination Driven by Hosts & Listings Characteristics?
Dependent variable: 1(host accepts)
(1) (2) (3) (4) (5) (6)
Guest has arabic name -0.19∗∗∗ -0.20∗∗∗ -0.17∗∗∗ -0.26∗∗∗ -0.22∗∗∗ -0.14∗∗∗
(0.044) (0.025) (0.029) (0.034) (0.021) (0.041)
Host is female -0.0036
(0.033) Guest has arabic name 0.035
× Host is female (0.058)
Shared property -0.060
(0.045)
Guest has arabic name 0.054
× Shared property (0.062)
Host has multiple listings 0.029
(0.027)
Guest has arabic name -0.0082
× Host has multiple listings (0.054)
Host has 10+ reviews 0.081∗∗
(0.032)
Guest has arabic name 0.13∗∗
× Host has 10+ reviews (0.052)
Host has Superhost status 0.091∗∗∗
(0.024)
Guest has arabic name 0.13∗∗
× Host has Superhost status (0.051)
Price per night > Mean -0.0037
(0.030)
Guest has arabic name -0.091
× Price per night > Mean (0.068)
Constant 0.66∗∗∗ 0.69∗∗∗ 0.65∗∗∗ 0.61∗∗∗ 0.63∗∗∗ 0.66∗∗∗
(0.036) (0.021) (0.033) (0.033) (0.030) (0.038)
Observations 1006 1006 1006 1006 1006 1006
R2 0.031 0.033 0.031 0.055 0.058 0.035
Note: The table reports results from regressions of the probability of receiving a booking invitation from a host on the guests ethnicity including variables for hosts and listings characteristics and the interaction of those characteristics with the guests ethnicity. Standard errors in parentheses, adjusted for clustering by county.
∗p < 0.10,∗∗p < 0.05,∗∗∗p < 0.01.
Table 7 presents the relationship between the guest’s ethnicity, social class and gender. Column 2 shows the result from the interaction term between the guest’s ethnicity and the guests signaled social class. Column 2 shows that guests that signal high social class with Swedish-sounding names are accepted 66 percent of the time which is the same probability as for the overall Swedish-sounding guest.
Hence, these results show no effect from social class for guests with Swedish-
sounding names. However, guests with Arabic-sounding names are accepted 56 percent of the time when signaling high social class, while they are accepted 41 percent of the time when signaling low social class. Hence, guests with Arabic- sounding names face an extra penalty of 15 percentage points when signaling a lower social class. This implies that guests with Arabic-sounding names have a 27 percent lower probability of receiving a booking invitation if their inquiry sig- nals low social class as opposed to a high. These results suggest that the strength of the signal of social class is stronger for guests with Arabic-sounding names compared to guests with Swedish-sounding names. These results are compatible with Carlsson et al. (2018), implying that living in a deprived neighborhood rather than in an affluent neighborhood has a negative effect on the probability of receiving a callback for job applicants with Arabic-sounding names but has no significant effect for job applicants with Swedish-sounding names. Column 3 reports that the female guest accounts have a higher probability of receiving a booking invitation, however, discrimination is robust across guests’ gender.
Table 7: The Impact of Ethnicity, Social Class and Gender Dependent variable: 1(host accepts)
(1) (2) (3)
Guest has arabic name -0.17∗∗∗ -0.10∗∗ -0.16∗∗∗
(0.026) (0.043) (0.025) Guest signals low social class -0.0028
(0.034) Guest has arabic name × -0.15∗∗
Guest signals low social class (0.053)
Guest has female name 0.11∗∗∗
(0.023)
Guest has arabic name × -0.0074
Guest has female name (0.052)
Constant 0.66∗∗∗ 0.66∗∗∗ 0.61∗∗∗
(0.033) (0.044) (0.035)
Observations 1006 1006 1006
R2 0.031 0.041 0.042
Standard errors in parentheses, adjusted for clustering by county.
∗p < 0.10,∗∗ p < 0.05,∗∗∗p < 0.01
Table 8 presents the proportion of booking invitation by guest name. In the first row, the table reports that 57% of the inquires to hosts were responded with a booking invitation. Hence, the overall proportion of booking invitations is higher in this experiment than in previous experiments made on Airbnb.22 The new settings on Airbnb which makes the experimental guest profiles more similar to other profiles might explain the relatively high overall proportion of booking invitation. In line with the existing literature female guests have a higher proportion of booking invitations compared to male guests. However, the differences between the signaled ethnicities in the proportion of booking invita- tion occur for both genders. The table shows that the female Arabic guest with the highest proportion of booking invitation (Halima) has the same proportion as the female Swedish guest with the lowest proportion (Johanna). Similarly, the male Arabic guest with the highest proportion of booking invitation (Ab- delhakim) is lower than the Swedish guest with the lowest proportion (Jonas).
None of these results are statically significant but still indicates that discrimi- nation may be robust across names.
One concern is that the experimental profiles are new and lack strong verifi- cation which might give rise to suspicion by hosts. The profiles with Arabic- sounding names might be more likely to be classified as fake, none-serious or spam by hosts. This might lead to differences in non-responses between guests with Arabic-sounding names and guests with Swedish-sounding names, and an overestimation of the ethnic discrimination. However, figure 1 shows the share of host response by ethnicity suggesting that the discrimination occurs from differences in “Yes” (booking invitation) and “No” answers, and not because of differences in “No response”.23
22Edelman et al. (2016) has 43% share of booking invitations in their sample, while the same number in Bethencourt and Farrahi (2017) is 29%.
23Table A3 in the appendix strengthen this argument by showing the share of host response by inquiry type.
Table 8: Proportion of booking invitation by guest name
Entire sample: 0.57
Female swedish Female arabic
Sara Andersson 0.70 Fateme Ahmed 0.49
Marie Björkvist 0.74 Nasrin Hassan 0.51
Johanna Gustafsson 0.64 Halima Mohammadi 0.64
Karolina Svensson 0.79 Aïcha Abdallah 0.52
Malin Wallin 0.78 Fatima Ahmad 0.52
Overall 0.71 0.55
Male swedish Male arabic
Jonas Söderström 0.59 Kamal Ahmadi 0.53
Erik Östberg 0.62 Abdallah Mohammed 0.42
Johan Nyström 0.63 Islam Hashemi 0.31
Mikael Andersson 0.60 Abdelaziz Hussein 0.35
Martin Berggren 0.60 Abdelhakim Hassan 0.56
Overall 0.61 0.44
Note: The table reports the proportion of booking invitation by guest name.
Figure 1: Share of Host Responses by Ethnicity
00.20.60.4
No No response Request Yes No No response Request Yes Guest have arabic name Guest have swedish name
7 Conclusion
In a field experiment on Airbnb in Sweden, this paper finds that guests with Arabic-sounding names are 17 percentage points less likely to receive a booking invitation relative to guests with Swedish-sounding names. When comparing the overall discrimination from this paper to the previous experiment, I find no significant evidence of reduced discrimination from the anti-discrimination policy on Airbnb. This is probably explained by the fact that the hosts, al- though having limited screening opportunities, still observe the guests names before making a decision. Guests with Arabic-sounding names are found to have lower rates of booking invitations regardless of social class. Furthermore, the evidence suggests that being associated with a lower social class decreases the probability of receiving a booking invitation for guests with Arabic-sounding names but not for guests with Swedish-sounding names. These results are in line with previous studies reporting that the strength of the signal of social class is stronger for ethnic minorities. These results suggest that ethnic minorities associated low social class struggle the most to rent housing on Airbnb. As
the sharing economy is growing it has the potential to provide an option for closing the housing shortage, however, the results from this paper suggest that the online market for housing rentals does not provide equal opportunities for all.
While the experimental approach is an effective way to measure the effect of social class on discrimination, it has some weaknesses. One weakness in this study is that it cannot be shown if the effect from social class is driven by differ- ences in message-style or living area. However, to assure that the hosts observe differences in the inquiries, I was forced to differ the low and high social class types of inquiries in both message-style and living areas.
The presented discrimination in this paper can be explained from both taste- based and statistical discrimination. The results of this paper suggest that mem- ber of an ethnic minority is not always confronted by the same type of stereo- types related to unobserved characteristics. Hosts on Airbnb in Sweden seem to have stronger negative stereotypes, or statistically based views, for guests with Arabic-sounding names when using language and living in neighborhoods associated with low social class. The pre-information provided to hosts about the potential guest before a booking is confirmed are imperfect which could give arise for statistical discrimination. The gap in booking invitation between low and high-class Arabic-sounding guests may be explained by general stereotypes and statistical discrimination as hosts may associate immigrants from a low so- cial class with a certain behavior. However, the extra penalty that the ethnic minority is facing from having a low social class could also be explained by an adjusted taste-based discrimination model. Differences in the effect of social class between the ethnicities could be due to preference-based discrimination, with preferences varying by social class type. Simply put, the distaste for an ethnic minority may be larger when signaling a low social class.
The discrimination presented in this paper suggest that further policy initiatives may be needed to guarantee equal opportunities for guests on Airbnb. Drawing on the findings from Goldin and Rouse (2000), online platforms, such as Airbnb, should aim to create gender- and color-blind platform designs. By reducing the
signal of ethnicity, Airbnb could hopefully reduce discrimination. Already hav- ing limited the information for hosts by withholding the guest profile picture until after the booking is confirmed, Airbnb could follow the design of eBay, Craigslist and Tradera and implement guest usernames instead of real names.
Depending on the guest’s choice of username, an introduction of usernames on Airbnb could eliminate the signal of ethnicity.
8 References
Andersson, R., & Bråmå, Å. (2004). Selective migration in Swedish deprived neighborhoods: can area-based urban policies counteract segregation process?, Housing Studies, 19(4), 517-539.
Aldén, L., & Hammarstedt, M. (2016). Boende med konsekvens: en ESO- rapport om etnisk bostadssegregation och arbetsmarknad, ESO Report No.
2016:1, Finansdepartementet, Stockholm.
Ahmed, A. M., & Hammarstedt, M. (2008). Discrimination in the rental hous- ing market: A field experiment on the internet. Journal of Urban Economics, 64(2), 362-372.
Arai, M., & Skogman Thoursie P. (2009). Renouncing Personal Names: An Emprical Examination of Surname Changes and Earnings. Journal of Labor Economics, 27 (1), 127-147.
Arai, M., Bursell, M., & Nekby, L. (2008). Between meritocracy and ethnic discrimination: The gender difference, Sociologiska institutionen, Stockholms universitet, Nationalekonomiska institutionen, & Samhällsvetenskapliga fakul- teten.
Arrow, K. J. (1973) The theory of discrimination. Discrimination in Labor Markets, 3(10) 3-33.
Becker, G. 1971 (1957). The Economics of Discrimination, 2d ed., Chicago:
University of Chicago Press.
Bethencourt, C., & Farrahi, N. (2017). Diskriminering i delningsekonomin - Ett fältexperiment på Airbnb. Unpublished manuscript, Stockholm University, Stockholm.
Bertrand, M., & Mullainathan, S. (2004). Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination.
American Economic Review, 94(4), 991–1013.
Bertrand, M., Dolly Chugh, & Mullainathan, S. (2005). Implicit Discrimina- tion. The American Economic Review, 95(2), 94-98.
Bertrand, M., Duflo, E., & Research, National Bureau of Economic. (2016).
Field experiments on discrimination. S.l.: National Bureau of Economic Re- search.
Botsman, R. (2016). RSA Journal, 162(5566), 48-48. Retrieved from http://www.jstor.org.ezproxy.its.uu.se/stable/26204509
Bursell, M. (2007). What’s in a name? A field experiment test for the existence of ethnic discrimination in the hiring process. SULCIS Working Paper 2007:7, Stockholm University, Linnaeus Center for Integration Studies, Stockholm.
Carlsson, M., Eriksson, S. (2014). Discrimination in the rental market for apart- ments.Journal of Housing Economics, 23, 41-54.
Carlsson, M., Reshid A. A. & Rooth, D.-O. (2018). Neighborhood signaling effects, commuting time, and employment: Evidence from a field experiment, International Journal of Manpower, 39(4), 534-549.
Carlsson, M., & D.-O Rooth. (2007). Evidence of Ethnic Discrimination in the Swedish Labor-Market Using Experimental Data, Labour Economics, 14(4), 716-729.
Edelman, B., Luca, M. (2014). Digital Discrimination: The Case of Airbnb.com, Working Paper, 14-054, Harvard Business School, Harvard University.
Edelman, B., Luca, M., & Svirsky, D. (2016). Racial discrimination in the shar- ing economy: Evidence from a field experiment. American Economic Journal:
Applied Economics, 9(2), 1-22. Also avaible as Working Paper 16-069, Harvard Business School, Harvard University.
Fisman, R., & Luca, M. (2016). Fixing discrimination in online marketplaces.
Harvard Business Review, 1.
Goldin, C., & Rouse, C. (2000). Orchestrating impartiality: The impact of
"blind" auditions on female musicians. The American Economic Review, 90(4), 715-741.
Hanson, A., & Hawley, Z. (2011). Do landlords discriminate in the rental hous- ing market? evidence from an internet field experiment in US cities. Journal of Urban Economics, 70(2), 99-114.
Heckman, J., J. (1998). Detecting discrimination. The Journal of Economic Perspectives, 12(2), 101-116.
Kakar, V., Voelz, J., Wu, J., & Franco, J. (2017) The Visible Host: Does Race guide Airbnb rental rates in San Francisco? Journal of Housing Economics, 40, 25-40
Phelps, E. (1972). The Statistical Theory of Racism and Sexism. American Economic Review, 62, 659-61.
Vuolo, M., Uggen, C., & Lageson, S. (2016). Statistical power in experimental audit studies: Cautions and calculations for matched tests with nominal out- comes. Sociological Methods & Research, 45(2), 260-303.
9 Appendix
Table A1: Results from survey: Validation of experimental names
Name Result Name Result
Fateme Ahmed 1.31 Kamal Ahmadi 1.34 Nasrin Hassan 1.36 Abdallah Mohammed 1.25 Halima Mohammadi 1.37 Islam Hashemi 1.27 Aïcha Abdallah 1.34 Abdelaziz Hussein 1.32 Fatima Ahmad 1.36 Abdelhakim Hassan 1.29 Sara Andersson 4.65 Jonas Söderström 4,70 Marie Björkvist 4.67 Erik Östberg 4.68 Johanna Gustafsson 4.69 Johan Nyström 4.65 Karolina Svensson 4.74 Mikael Andersson 4.73 Malin Wallin 4.73 Martin Berggren 4.70
Note: The respondents choose from a scale of 1-5, where 1 is Arabic- sounding and 5 is Swedish-sounding. The results is based on a sample size of 240 respondents.
Table A2: Number of Inquiries Send from Each Category
Arabic Name & Swedish Name & Arabic Name & Swedish Name & Total High Socal class High Socal class Low Social Class Low Social Class
Number of inquiries 261 255 243 247 1006
Host characteristics:
Female host 143 155 138 139 575
Male host 110 85 92 96 383
Host is a couple 5 10 8 8 31
Host sex is unknown 3 5 5 4 17
Shared property 91 110 99 93 393
Entire place 170 145 144 154 613
Host has multiple listings 46 63 57 54 220
Host has 10+ reviews 167 175 155 145 642
Host has Superhost status 92 81 85 75 418
Cancellation policy:
Moderate 89 82 82 76 329
Flexible 118 115 116 117 466
Strict 54 58 45 54 211
Geographical characteristics:
Rural municipalities 63 45 54 60 222
Metropolitan cities 148 180 154 134 616
Medium-sized town 50 30 35 53 168
Figure A1: Example of a Type 1 Inquiry
(a) Type 1 guest message
(b) Type 1 guest profile page observed by host
Figure A2: Example of a Type 4 Inquiry
(a) Type 4 guest message
(b) Type 4 guest profile page observed by host
TableA3:NumberofHostResponsebyInquiryType ArabicName&SwedishName&ArabicName&SwedishName& HighSocalClassHighSocalClassLowSocialClassLowSocialClass No57329142 Noresponse169913 Requestmoreinfo.42454329 Yes146169100163
