Identifying rental discrimination on the Flemish housing market: an intersectional approach

ABSTRACT

This study aims to identify rental discrimination on the Flemish rental housing market in Belgium, taking the intersectional nature of discrimination into account. Most discrimination studies focus on unequal access based on one individual discrimination ground. This practice eludes the context of the other discrimination grounds in which rental discrimination occurs and neglects the intersectional nature of discrimination. We therefore conducted 8.245 correspondence tests in almost all municipalities in Flanders, the Dutch-speaking part of Belgium. We apply an intersectional lens, by considering the relation and interaction between gender (male/female), ethnic origin (Moroccan/Polish) and the homogeneity of names (homogenous/mixed). We find four strata of rental discrimination, in which Moroccan female rental candidates with a homogenous name experience most discrimination, indicating that multiple categorical identities enforce each other. Without a full intersectional approach, these layered patterns of exclusion would have been hidden behind the bold boundaries of ethnic or gender categorizations.

KEYWORDS:

• Ethnic discrimination
• gender discrimination
• intersectionality
• housing market
• correspondence tests
• homogeneity of names

Introduction

Discrimination and unequal access to housing remain major sources of inequality in our modern-day society. Although the level of discrimination varies strongly between countries and contexts (Auspurg, Hinz, and Schmid 2017), the general pattern is still one of severe disadvantage for more vulnerable groups (Auspurg, Schneck, and Hinz 2019; Flage 2018). Besides individual consequences of rental discrimination – like poorer access to employment, education or limited labour mobility (Squires 2007) – there are larger societal and residential inequalities at stake too (MacDonald, Galster, and Dufty-Jones 2018).

Most discrimination studies focus on unequal access on the basis of one individual discrimination ground, like ethnic origin (Quillian, Lee, and Honoré 2020), sexual orientation (Flage 2021) or disability (e.g. Fumarco 2017; Verhaeghe, Van der Bracht, and Van de Putte 2016), sometime, but only rarely in combination with gender (Flage 2018). However, considering merely one ground of discrimination when measuring rental discrimination is to artificially break a candidates’ characteristics down and reduce it to a single feature. This eludes the context of the other discrimination grounds in which discrimination on the housing market is occurring and neglects the intersectional nature of discrimination. This might lead to an underestimation of overall levels of rental discrimination, as multiple personal features may coexist and even reinforce the level of discrimination (Collins 2015; Crenshaw 1989). It is becoming increasingly apparent that dynamics of discrimination are only to a limited extent understood by uncovering simple and dualistic links between two social groups (Ruwanpura 2008).

Several theoretical perspectives have been developed on the intersection of discrimination grounds across multiple domains (Cho, Crenshaw, and McCall 2013), but these have hardly been linked to experimental evidence of discrimination (Di Stasio and Larsen 2020). In this study, we present the results of a field experiment on housing market discrimination, considering intersections between three major discrimination grounds: ethnic origin, gender and the homogeneity of names (first and last name signalling the same or a different ethnic origin). We aim to make several contributions to the literature. First, we consider multiple forms of intersectionality in our data collection and analysis. Building on previous studies, we integrate a full intersectional model in design and analysis. Analysing two-, and three-way interactions between ethnic origin, gender and the homogeneity of names with multilevel analysis on invitation rates, we provide an overview of all (90) possible interactions between three major candidate characteristics.

Second, we aim to add to scientific knowledge by measuring discrimination towards fictious rental candidates with homogenous names or mixed names. Whereas previous experimental research predominantly used homogenous names as a proxy for ethnic origin (first and last name signal the same ethnic origin), we assess whether candidates with a mixed name (first and last name proxy for a different ethnic origin) experience differential treatment on the rental housing market. The combination and interaction with ethnic origin and gender provide us with a clearer understanding why treatment would differ between candidates with homogenous names and candidates with a mixed name. With societies becoming increasingly superdiverse (Meissner 2015), it is worth considering whether there is a different treatment towards people with mixed names as compared to people carrying ethnic homogenous names, as well as how this depends on their specific ethnic origin and gender.

Third, we examine rental discrimination in almost all municipalities of Flanders, the northern region of Belgium. In total 16.490 enquiries for rental, dwellings were made. This allows us to establish a comprehensive view of rental discrimination within a larger context and enables us to extrapolate results to a broader framework. This is the first paper to measure discrimination in such a large area in Belgium with this high number of tests. Although Belgium is a country that has already been subject to research in multiple studies on housing market discrimination (Ghekiere and Verhaeghe 2022; Heylen and Van den Broeck 2016; Verhaeghe and De Coninck 2021), both the amount of conducted tests as the selected area are unique. In this study, we apply multilevel logistic regression analysis on the odds to be invited for a viewing to 16.490 enquiries to test the relation and interaction between ethnic origin, gender and the homogeneity of a name in the search for a rental property.

Theory and hypothesis

Compared to the vast amount of research on ethnic origin, very little attention has been paid in previous research to the intersection of discrimination grounds when applying for housing. Some exceptions exist, such as gender and ethnic origin (e.g. Andersson, Jakobsson, and Kotsadam 2012; Björnsson, Kopsch, and Zoega 2018), ethnic origin and sexual orientation (e.g. Murchie and Pang 2018; Schwegman 2019) and gender and wealth (Faber and Mercier 2022).

“Intersectional approaches to the study of inequality share the increasing recognition that multiple categorical identities may interact in complex ways and fundamentally alter the meaning of category membership” (Di Stasio and Larsen 2020, 231). To explain how these categories intersect with each other to produce unique forms of discrimination, scholars have used intersectional theory and conceptualized discrimination as a multidimensional process along a matrix of oppression (Faber and Mercier 2022; Pedulla 2014, 2018). Intersectionality arises when unequal treatment in two or more individual categories generates a unique form of disadvantage that fundamentally alters the effect of one, single categorical membership (Collins 2015). Although previous research has extensively discussed the potential drivers for intersectionality in discrimination (Derous, Ryan, and Serlie 2015; Mukkamala and Suyemoto 2018), research that investigates this theoretical explanation with sound empirical evidence about discriminatory behaviour is scarce.

In this study, we test discrimination against rental candidates with Moroccan and Polish names. Previous research in Europe has found extensive discrimination against these two ethnic origin groups. In Sweden, Norway, Belgium, Spain, France and Germany, candidates with Arab/Muslim names are found to be significantly discriminated against (Ahmed and Hammarstedt 2008; Andersson, Jakobsson, and Kotsadam 2012; Auspurg, Hinz, and Schmid 2017; Bosch, Carnero, and Farre 2010; Ghekiere and Verhaeghe 2022; Heylen and Van den Broeck 2016; Le Gallo et al. 2019). Additionally, significant but lower levels of discrimination are also found towards candidates with European-sounding names (Acolin, Bostic, and Painter 2016; Baldini and Federici 2011; Björnsson, Kopsch, and Zoega 2018; Carlsson and Eriksson 2014). The majority of these studies turn to the two dominant theoretical mechanisms of discrimination in both sociological and economic literature on rental discrimination, namely taste-based and statistical discrimination. Taste-based discrimination occurs when the agent’s own preference or bias is the main driver for a discriminatory selection (Becker 1971). Discrimination through the taste-based mechanism is considered a conscious, personal act. In the original formulation of taste-based discrimination, Becker (1971) argues that gatekeepers are willing to pay a certain price for not interacting with a member of an (ethnic) minority group. In labour market context, this cost is a percentage of their wage directly proportionate to their experienced distaste to avoid working alongside minority colleagues (Becker 1971; Lippens, Baert, and Derous 2021). On the housing market, it would be the cost of not renting out the dwelling because of a personal distaste towards certain candidates. Statistical discrimination, on the other hand, builds on the notion that unequal treatment arises when no or little information on the candidate is at hand, which pushes the agent to use general group characteristics to judge the individual candidate (Arrow 1973). However, as Bohren et al. (2019) argue, the underlying assumption of this mechanism is that they rely on accurate beliefs about ethnic groups, but reality shows that the occurrence of inaccurate beliefs and prejudices and thus inaccurate statistical discrimination is often present. Besides the preceding empirical findings and theoretical mechanisms of discrimination, we expect that the substantial electoral success of far-right political parties in Europe and in Belgium account for a large share of negative attitudes towards multiple ethnic minority groups (Castelli Gattinara and Pirro 2019). We hypothesize that both type of candidates, with Moroccan and Polish names, experience discrimination in access to housing (H1).

Secondly, we vary the homogeneity of the names used to signal the discrimination grounds. Previous research predominantly used homogenous names (first and last name signalling the same ethnic origin) as a proxy for ethnic origin when measuring discrimination (Gaddis 2017). However, in this study, we also include heterogenous names, a combination of an Anglo-Saxon first name and a last name stemming from another ethnic origin group (e.g. Belgian first and Moroccan last name). These mixed names have been used in research before but are rarely empirically tested (Gaddis 2019; Martiniello and Verhaeghe 2022). However, they do make for an interesting case. When considering the comparison between mixed and homogenous names, higher discrimination rates might be expected towards candidates with homogenous as compared to mixed names for three reasons. The first reason is what we name the spatial-distance explanation: the name a person carries might give indications as to whether or not the person is born in the studied country (Gaddis 2019). A person with a homogenous name might be perceived as a first-generation migrant. This as opposed to carrying a mixed name, which might raise the perception that the person is a 1.5, second or further generation migrant, and is thus born in the studied country (or migrated at a very young age) but has roots somewhere else. The second reason, here called the cultural-distance explanation, comprises an emotional-cultural aspect. A person with a homogenous name might be seen as someone (or a member of a family – as a person’s name is given by his/her parents) who is unwilling to adapt to the culture of the studied country. On the contrary, having a mixed name might be seen as a signal for the willingness to culturally adapt (Gaddis 2019; Tuppat and Gerhards 2021). Both the spatial-distance and cultural-distance explanations indicate that mixed names might be perceived as an increase in perceived nearness towards the ethnic majority group as compared to homogenous names. This perceived nearness might in turn result in lower discrimination rates, influenced by in-group favouritism (Greenwald and Pettigrew 2014). Thirdly, a more methodological reason is that signals of ethnic origin should be clear in order to trigger discriminatory behaviour (Tuppat and Gerhards 2021). It is harder to assess the ethnic origin of a name for mixed names (Martiniello and Verhaeghe 2022), possibly leading to lower levels of discrimination. We consequently hypothesize that among ethnic minorities, candidates with homogenous names experience more discrimination than those with mixed names (H2).

Thirdly, we include gender in our data collection and analysis, distinguishing between male and female rental candidates. While discrimination on the labour market is mostly present against female job candidates, most studies seem to prove that men are often disadvantaged on the rental housing market as compared to women (Ahmed and Hammarstedt 2008; Andersson, Jakobsson, and Kotsadam 2012; Baldini and Federici 2011; Flage 2018; Heylen and Van den Broeck 2016). Possible theoretical explanation could relate to the statistical mechanism of discrimination, in which group characteristics of female candidates are used to assess the fit of the individual candidate. However, these group characteristics could be subject to gender stereotyping. Female rental candidates may be considered cleaner and tidier, and therefore better in maintaining a dwelling in a good condition. Additionally, as Gusciute, Mühlau, and Layte (2022) argues, women could be favoured as tenants due to their “agreeable” and “conscientious” personalities (Hyde 2014). Therefore, we hypothesis that female rental candidates are favoured over male rental candidates (H3).

As for the intersect between ethnic origin and gender, Bengtsson and colleagues (2012) argue that ethnic discrimination overshadows gender discrimination, leading to the “gender advantage” for women to decrease or even disappear for rental candidates with Arab/Muslim names. In contrast, studies by Baldini and Federici (2011) or Heylen and Vandenbroeck (2016) find that discrimination is higher against male than female rental candidates with Arab/Muslim names. In general, a recent meta-analysis has shown that female candidates have higher chances to be invited for a viewing than male candidates (Flage 2018). Ethnic minority male candidates, however, experience more gender discrimination than the ethnic majority male candidates. Also, Andersson, Jakobsson, and Kotsadam (2012) show that in invitation rates for the Arab candidates, the female candidate is favoured over the male candidate. Same goes for the study by Bosch, Carnero, and Farré (2015), in which the results show that, while ethnic majority women have lower invitation rates than their male counterparts, ethnic minority female have higher chance of invitation than their male counter candidates. Importantly, in most studies, gender discrimination is measured within the same ethnic group. Female ethnic minority candidates are compared to male ethnic minority candidates and male ethnic majority candidates with their female ethnic majority counterpart. However, to measure the intersect between gender and ethnic origin, it is of importance to also study the presence of the two grounds (treatment group) against the baseline model (control group). Following the empirical evidence in the literature, we hypothesize that, ethnic minority men face most discrimination (H4).

Finally, as for the intersection between gender and the homogeneity of the name, we expect that the combination of both grounds of discrimination, gender and the homogeneity of names, will lead to more discrimination than the effects of one. More concretely, ethnic minority men with homogenous names are expected to be more disadvantaged than ethnic minority women or men with mixed names (H5).

In what follows, we will first discuss the research design and the applied multilevel logistic regression analysis. This is followed by a presentation of the results, whereby we built from individual identity-element effects on discrimination, to a combination of two elements to a full integration of ethnic origin, gender and the homogeneity of names. The findings are consequently discussed in the conclusion.

Research design

Between April and June 2021, we conducted 8.245 correspondence tests among real estate agents and private landlords on the private rental housing market in Flanders, the largest and Dutch-speaking region of Belgium (see Figure 1 for visual overview). The regional housing market is dominated by home owners with outstanding mortgages, which were for long heavily subsidized by the government. Private renters account for a relatively small proportion of tenure types and are especially popular among the lower income groups (Winters and Heylen 2014). The proportion of social housing track is even more marginal in Belgium. This study only focusses on discrimination on the private rental market.

Figure 1. Visual overview of effectuated tests in Flanders, Belgium.

Discriminatory behaviour on the rental housing market was measured with the field experimental technique of correspondence testing. In these experiments, two quasi-identical candidates apply for a rental advertisement (Gaddis 2018; Verhaeghe 2022). Only the characteristic that signals the discrimination ground (e.g. their name when measuring ethnic discrimination) differs between the two candidates. When the test group (e.g. ethnic minority) is systemically less invited to a viewing or treated differently from the control group (e.g. ethnic majority), discrimination is at stake (Verhaeghe et al. 2023). These field experiments are considered as the “golden standard” to objectively measure discrimination in the field (Heath and Di Stasio 2019; Verhaeghe 2022), and are inherent to the majority of studies on housing market discrimination (for example see Auspurg, Hinz, and Schmid (2017)). Moreover, its experimental design allows to establish causal relationships between the discrimination ground under scrutiny and the behavioural reaction of landlords and realtors. Finally, in contrast to face-to-face audit studies, correspondence testing allows to standardize the content and form of contact with the landlords or realtors, which already counters one of the so-called “Heckman and Siegelman (1993) critiques”.

First, as we test for multiple grounds of discrimination, 10 different types of profiles were created, whose names signal the tested discrimination grounds: ethnic origin, gender and the homogeneity of names. Similarly as in previous research, we use the name of the candidate as a proxy for these grounds (Gaddis 2017). The names we used stem from a survey on the perception of names conducted among a sample of 990 inhabitants of Belgian origin (Martiniello and Verhaeghe 2022). The name-combinations were created out of databases accessible through StatBel (Statistics Belgium). These databases provide lists with the most popular male and female first names between 2010 and 2019 and the most recurring last names in 2020. The results of Martiniello and Verhaeghe (2022) show that approximately 98.5 per cent of the respondents successfully differentiate between homogenous names as of Belgian origin or not. For mixed names, the rate is of 95.5 per cent. For the distinction between names as European or not, 88.6 per cent of the respondents are successful in categorizing the Belgian names, followed by 64.3 per cent and 61.8 per cent for the mixed and homogenous Polish names respectively. Respondents were less successful in classifying Moroccan names (48.2 per cent for homogenous and 36.3 per cent for mixed names). The rates for successfully recognizing the specific ethnic origin are 35 per cent and 32.6 per cent for homogenous and mixed Polish names. For Moroccan homogenous and mixed names, the rates are 34 per cent and 21.3 per cent. Nevertheless, the authors show that there is important within-group variation, depending on the specific name. We therefore used the names with the highest congruence rates and the ones validated in their research.

In order to avoid detection, multiple names were used for each type of profile. An overview of the used names can be found in Table 1. This specific research population was chosen because residents with Moroccan roots (including both first and second generation) contribute to the second biggest ethnic group in Flanders, consisting of 3.2 per cent of the total Flemish population, after the Dutch population (data accessed via statbel.be). Additionally, Polish rental candidates are studied, as in Flanders they represent 1 per cent of the total Flemish population (first and second generation), being the fifth biggest nationality group in relative terms (data accessed via statbel.be). The Moroccan community has a long migration history in Belgium, dating back to the labour migration in the sixties, followed by family migration since the mid-seventies (Timmerman 2017). In contrast, the Polish migration started more recently after the fall of the Iron curtain and was accelerated with the expansion of the EU in 2004. Anti-immigration attitudes in Belgium are worse about migrants from non-European, Muslim countries (such as Moroccan migrants) than about European, non-Muslim migrants (such as Polish migrants) (De Coninck 2020). In addition, the popular far-right party in the Flemish region especially stigmatizes the non-European migrants from the Maghreb and Turkey, and less the intra-European migrants.

Table 1. The used names in the correspondence testing.

Belgian homogenous man	Thomas Goossens	David Verhoeven	Bert Vermeulen	Kevin Lemmens	Pieterjan De Smet	Maarten Wauters	Davy Declerq	Kenny Cools
Belgian homogenous woman	Valerie Devos	Eva Segers	Nele Aerts	Evi Janssens
Moroccan homogenous man	Hassan El Battoui	Yassin Ben Aïssa	Youness El Malahi	Ayoub Haddioui	Mohamed Abdelaziz	Rachid El Khadji	Karim Azzouzi	Youssef El Ayadi
Moroccan mixed man	Maxim El Moussaoui	Loïc El Salhi	Maarten El Boujdaini
Moroccan homogenous woman	Dounia El Majdoub	Karima El Yahyaoui	Nisrine El Amrani	Norah El-Bazioui
Moroccan mixed woman	Lena El Makrini	Amy El Morabit	Vanessa Achbar	Nele El Hilali
Polish homogenous man	Wiktor Wozniak	Tomasz Sobkow	Lukasz Wieczorek
Polish mixed man	Jan Kowalski	Simon Wyrzykowski	Pieter Jaworski
Polish homogenous woman	Magda Piotrowska	Teresa Kwiecinska	Dorota Dabrowska	Gabriela Pawlak
Polish mixed woman	An Ziemczyk	Evi Kowalczyk	Kelly Mancewicz	Sara Wronkowska

Source: Martiniello and Verhaeghe 2022.

Subsequently, two (or three) fictitious rental candidates applied to the same rental property. As the sampling frame for rental advertisements, we used Immoweb.be, the largest online rental platform in Belgium. The platform was not informed about the testing. Following previous research (McLaren and Shanbhogue 2011; Verhaeghe and Ghekiere 2020), Google search data can be used to roughly estimate the relative rental demand. According to Google trends, 86 per cent of the rental online demand in Belgium during the period of data collection was on Immoweb, whereas the two alternative websites (Zimmo and Immovlan) only represented 14 per cent of the Google searches. Therefore, Immoweb can be considered as a representative sampling frame for the online rental demand. Nevertheless, it doesn’t cover advertisements who are only offered offline or informally, which might inhibit the representatives of this study (see Verhaeghe 2022).

The rental candidates have identical profiles with the exception of the investigated discrimination ground. Concretely, the two candidates differ only in their names. For both candidates, an e-mail address, telephone number and a profile are created on Immoweb.be, in order to execute the application but also to avoid detection. We first sent out the e-mail of the test person (e.g. the candidate with a Moroccan or Polish-sounding name) and a few hours later the e-mail of the control person (e.g. the candidate with a Belgian-sounding name). By doing so, we have strong evidence that in the case where the control person is invited and the test person is not, discrimination is at stake (Baldassarri and Abascal 2017).

We conducted one to two tests per week, with sufficient spacing between tests concerning the same type of profile. Afterwards, we kept track of the written responses of the realtors and landlords for ten days. Next, we examine the extent to which the candidates in the test group were treated systematically more unfavourably than the candidates in the control group. When there is systematic disadvantageous treatment, it is attributed to discrimination. The online application for housing proceeds in the same way each time. We used the standard message built into the contact function of the rental website, which could be translated as “Dear, I found your property on Immoweb and would like to schedule a viewing. Thank you”. This message was copied and signed by the candidate with his/her name.

The data is analysed by means of multilevel logistic regression analysis. The dependent variable is “invitation”, which is a dichotomous variable whereby 1 stands for an invitation and 0 for no invitation. Because rental dwellings (level 1) are nested in advertisements (level 2), we conduct multilevel analysis. The variances in the null model are significant (p < .001), with an ICC of 0.963 indicating a very good reliability. The independent variables are $x_1$: ethnic origin (Belgian, Polish or Moroccan), $x_2$: gender (man or woman) and $x_3$: homogeneity (homogenous or mixed). No cross-level interaction is needed as all interactions are situated on level 1. The specification of our three-way interaction model is expressed as: $y_{ijk}={\text{β}}_{0jk}+{\text{β}}_1{x}_1+{\text{β}}_2{x}_2+{\text{β}}_3{x}_3+{\text{β}}_4{x}_1{x}_2{x}_3+v_k+e_{ijk}$

Let $y_{ijk}$ be the invitation to a dwelling. Our random intercept model with a three-way interaction term contains two error terms, with $e_{ijk}$ being the residual error. We controlled all regression analysis for the rental price and size of the dwelling (expressed in number of rooms) as a test for robustness. Also, as a robustness tests and because of the multitude of interactions in our analysis, we added a regression analysis that accounts for all possible reference categories, in the Appendix.

Results

Descriptive results

In Table 2, we present the bivariate statistics on the rate at which the tested profiles are invited to view a rental dwelling in Flanders. When considering only ethnic origin, candidates with Moroccan names have the lowest invitation rate (27.8 per cent), followed by candidates with Polish (33.8 per cent) and Belgian (37.8 per cent) names. Besides, having a homogenous or mixed does not appear to have a strong influence on being invited, as for the former the invitation rate is of 34.0 per cent and for the latter 32.8 per cent. The invitation rates for men and women are comparable, with 34.2 per cent and 33.1 per cent, respectively.

Table 2. Descriptive statistics (n = 16.490).

		% Invitations for a viewing	Nnames	Cramer’s V
Ethnic origin	Belgian name	37.8%	12	0.085***
	Moroccan name	27.8%	18	0.089***
	Polish name	33.8%	15	0.005
Homogeneity	Homogenous name	34.0%	31	0.010
	Mixed name	32.8%	14	0.010
Gender	Man	34.2%	25	0.011
	Woman	33.1%	20	0.011
Triple intersection	Belgian – homogenous – woman	36.1%	4	0.024**
	Belgian – homogenous – man	38.8%	8	0.071***
	Moroccan – homogenous – woman	27.4%	4	0.031***
	Moroccan – homogenous – man	26.5%	8	0.072***
	Moroccan – mixed – woman	30.3%	4	0.017*
	Moroccan – mixed – man	31.2%	3	0.011
	Polish – homogenous – woman	30.9%	4	0.017*
	Polish – homogenous – man	34.2%	3	0.001
	Polish – mixed – woman	33.1%	4	0.003
	Polish – mixed – man	36.9%	3	0.014

Note: ***p < .001; **p < .01; *p < .05.

We also take the intersection on ethnic origin, gender and mixed versus homogenous names into account. Male candidates with a Belgian first and last name have the highest invitation rates, with 38.8 per cent. Female candidates with a Belgian first and last name (36.1 per cent) have comparable invitation rates to those of male candidates with a Belgian first name and a Polish last name (36.9 per cent). Candidates with a Moroccan first and last name, both man and woman, overall receive the lowest amount of positive response (respectively 26.5 per cent and 27.4 per cent). Candidates with mixed Moroccan names (respectively 30.3 per cent and 31.2 per cent for women and men) have comparable invitation rates as candidates with Polish homogenous names (respectively 30.9 per cent and 34.2 per cent for women and men).

Regression analysis

To arrive at an intersectional approach, we conduct multilevel logistic regression analysis on the odds to being invited for a viewing. First for the three discrimination grounds separately (models 1, 2 and 4), then for the interaction effect of two elements at a time (models 3, 5 and 6), and lastly for the three-way interactions (Table 4). The results of the one- and two-way interactions are shown in Table 3. To test our first hypothesis, which expects discrimination towards Moroccan and Polish rental candidates, we look at the effect of ethnic origin on the odds to be invited for a viewing. The results in model 1 show that candidates with Moroccan (0.491) and Polish (0.651) names have significantly lower odds to be invited for a viewing than candidates with Belgian names. Especially candidates with Moroccan names have the lowest odds. This result supports our first hypothesis.

Table 3. Multilevel logistic regression analysis on the odds to be invited for a viewing.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
	OR (SE)	OR (SE)	OR (SE)	OR (SE)	OR (SE)	OR (SE)
	n = 16.490	n = 16.490	n = 16.490	n = 8245	n = 8245	n = 8245
Intercept	0.341 (0.131)***	0.295 (0.130)***	0.413 (0.135)***	0.263 (0.117)***	0.305 (0.135)***	0.273 (0.120)***
Ethnic origin (ref. Belgian name)
Moroccan name	0.491 (0.050)***		0.437 (0.061)***		0.828 (0.085)*
Polish name	0.651 (0.061)***		0.666 (0.087)***		(ref.)
Gender (ref. man)
Woman		0.662 (0.058)***	0.598 (0.073)*			0.909 (0.081)
Homogeneity (ref. homogenous name)
Mixed name				1.214 (0.057)***	1.225 (0.097)*	1.288 (0.079)***
Ethnic origin*gender
Moroccan name*woman			1.265 (0.108)*
Polish name*woman			1.047 (0.124)
Homogeneity*gender
Mixed name*woman						0.925 (0.117)
Ethnic origin*homogeneity
Moroccan name*mixed					0.898 (0.143)
Polish name*mixed					–
Rent	YES	YES	YES	YES	YES	YES
AIC	79317.531	78877.267	79600.355	37295.855	37328.907	37313.180

Note: For the analyses including the variable “Homogeneity” (models 4–6), we excluded the Belgian names from the data as these are all homogenous.

***p < 0.001; *p < .05.

The second hypothesis expects a higher discrimination level against candidates with homogenous names, compared to mixed names. Our results in model 4 show that mixed names indeed have higher odds (1.214) to be invited for a viewing than homogenous names.

To test the third hypothesis, whereby we expect that female rental candidates are favoured over male rental candidates, we analyse the effect of gender on the odds to be invited for a viewing. Contradicting the hypothesis, we find that women are significantly less likely (0.662) to be invited for a viewing as compared to men. While this result is conflicting with Flage’s (2018) meta results, we do find some individual studies in Flanders that comply with our results (Verhaeghe, Martiniello, and Ghekiere 2020). Importantly, gender is measured as an overarching variable that includes both ethnic majority and ethnic minority candidates. Results from the interaction terms will provide more insight in the gender effects for ethnic groups separately.

When looking at the interaction effects to control for a two-way intersection, we find that having a Moroccan name and having a female name is the only intersect that significantly interacts. The effect of gender is opposite (1.265) for candidates with Moroccan names compared with candidates with Belgian names (0.598). Further analyses show that Moroccan women do not significantly differ from Moroccan men in their odds to be invited for a visit. Models 5 and 6 show that the intersection of the homogeneity of names with ethnic origin or gender appears to be less influential. With no effect from the intersect of homogeneity of names and gender, we reject hypothesis 5.

In Table 4, we present the relative chance to be invited for a viewing by taking the three-way intersection of gender, ethnic origin and homogeneity of rental candidates’ name into account. A red colour portrays a negative effect and the green colour a positive effect. The stronger the shade, the stronger the effect (no colour means no significant effect; see Table A1 in Appendix for more detailed results with significance levels). From this table, it appears that (male) candidates with Belgian homogenous names are treated the most advantageously on the Flemish rental housing market, followed by candidates with Polish mixed names. In addition, male and female candidates with Moroccan homogenous names are confronted with the most rental discrimination, as all other profiles have significantly higher relative chances to be invited. Ordering the different tested profiles from least to most rental discrimination, taking the three-way intersection into account, we become the following ethnic hierarchy on the rental housing market: stratum (1) candidates with Belgian male names; stratum (2a) Polish mixed male names, (2b) Polish homogenous male names, Belgian female names, (2c) Moroccan mixed male names, Polish mixed female names; stratum (3a) Moroccan homogenous male names, Polish homogenous female names, (3b) Moroccan mixed female names; and stratum (4) Moroccan homogenous female names. This classification is visualized in Figure 1.

Table 4. The odds to be invited for a viewing: three-way interactions.

Note: Calculated by means of multilevel binary logistic regression analysis. All results are controlled for the rent.

The green colour stands for a significant positive effect, the red colour for a significant negative effect and the white colour for no significant effect.

The shade of the colour visualizes the strength of the effect. A darker shade means a stronger (either positive or negative) effect.

Paying attention to a full intersectional approach appears to be relevant, as the relative chance to be invited for candidates with Belgian female names does not significantly differ from those of candidates with, mixed Polish male and female; Polish homogenous male names and mixed Moroccan male names (stratum 2). This although Belgian names, without taking an intersectional approach, appear to be the most invited (Table 2). Also, candidates with Polish homogenous female names do not have significantly different invitation rates than candidates with Moroccan homogenous male names or mixed female names (stratum 3).

Adding the level of homogeneity of names to the picture further contributes to our understanding. In line with the double jeopardy theory as formulated in hypothesis five, Moroccan women face more discrimination than Moroccan men, when they both have mixed names or homogenous names. These findings contradict hypothesis 4. This same pattern of double jeopardy could be found among Polish candidates: no matter whether they have homogenous or mixed names, Polish women are always discriminated more than Polish men. Moreover, contradicting hypothesis 5, Polish women with homogenous names are more disadvantaged than Polish women with mixed names and Polish men. This confirms the relevance of applying a full intra- and inter-categorical lens on studying ethnic disadvantages on the housing market (Figure 2).

Figure 2. Odds to be invited to a viewing.

Discussion and conclusion

Although many studies have already shown persistent patterns of ethnic and gender discrimination on the housing market (Auspurg, Hinz, and Schmid 2017; Flage 2018), there is surprisingly little research that investigates these disadvantages from an intersectional approach, in which multiple grounds of discrimination are examined together. The goal of this paper is to examine the racialized and gendered nature of discrimination on the housing market with an intersectional design and analysis. Moreover, in line with recent research (e.g. Gaddis 2017; Gaddis 2018; Martiniello and Verhaeghe 2022) we further distinguish intra-categorically between homogenuous and mixed minority names of the rental candidates. For these aims, we collected data on discriminatory behaviour against Moroccon and Polish candidates by means of correspondence tests with 16.490 rental enquiries.

In line with previous research (e.g. Heylen and Van den Broeck 2016; Ghekiere and Verhaeghe 2022; Verhaeghe and De Coninck 2022), we found pronounced discrimination against Moroccan and Polish candidates on the rental housing market in Flanders. Moreover, it appears that Moroccan candidates are more discriminated than their Polish counterparts. This resembles previous studies which indicate a bright European vs. non-European boundary in rental discrimination patterns (e.g. Acolin, Bostic, and Painter 2016; Baldini and Federici 2011; Carlsson and Eriksson 2014). Further research could shed light to which extent this boundary is based on ethnic-national, religious and/or skin colour factors.

Gender discrimination appeared to exist in favour of male rental candidates, as opposed to what we hypothesized in the theoretical framework. While this result contrasts multiple individual and review studies on the housing market (Flage 2018, 2020), we do find similar results in other studies on the Belgian rental market (e.g. Verhaeghe, Martiniello, and Ghekiere 2020). Also, Massey and Lundy (2001) who used a similar intersectional model in their analysis, found similar results. They found that gender correlated strongly with financial strength, resulting in excessive discrimination towards poor, black, female renters. This form of statistical discrimination could be at stake in our sample too. However, it is important to note that although we found significant gender effects, the difference in invitation rates is very small (33.1 vs. 34.2 per cent), considerably smaller than the difference in invitation rates for ethnic origin (37.8 vs. 27.8 per cent).

For the homogeneity of names, candidates with homogenous names experience significant more discrimination then the counter candidates with mixed names. Consequently, a perceived nearness to the ethnic majority leads to less discrimination. Besides, the most substantial contribution of this study lies in its full intersectional design along the axes ethnic origin, gender and homogeneity of names. Going from simple bivariate analyses to two- and three-way interaction models we find that the levels of discrimination vary much when we add intra-categorical distinctions within each ethnic group. Next to clear gender inequalities within each ethnic group where women are discriminated more than men, we find four strata on the rental housing market in Flanders. At the top are the male candidates with Belgian names, who are invited the most for a visit. Next are the Polish men with mixed or homogenous names, the Belgian women, the Moroccan men with mixed names and the Polish women with mixed names. This second stratum is followed by a third statum composed by Polish women with homogenous names, Moroccan men with homogenous names and Moroccan women with mixed names. The most discriminated stratum consists of Moroccan women with homogenous names. These findings support and extend the double – here even triple – jeopardy theory (Beal 2008), which states that multiple discrimination grounds enforce each other. Without a full intersectional approach, these layered patterns of exclusion would have be hidden behind the bold boundaries of ethnic or gender distinctions.

Our results expose the relevance and importance that the intersection of different categories – as opposed to simple additive effects of different categories – has on measured discrimination. These findings are relevant for policy, indicating the necessity to move beyond the focused attention on single categories (ethnic origin or gender or sexual orientation …). However, this does not mean a move towards policies that take different categories individually into account, as is the case in diversity policies. Although the latter recognizes different categories of discrimination, these categories are often analysed parallel to each other, rather than as intersecting with each other (Schiller 2017). Better, policy – and specifically anti-discrimination policy – should address discrimination based on the intertwined effect of belonging or being ascribed to multiple categories.

The conclusions should, however, be considered within the confines of a few limitations. Firstly, we only examined the first phase of the rental process – getting an invitation for a viewing or not – whereas more and other forms of discrimination could occur, also in the latter phases of the process (Ghekiere et al., 2022). In addition, this study only investigated discrimination for rental properties advertized on a popular online platform in Belgium, leaving the informally let properties out of the picture. If there is more discrimination occurring via informal selection processes, this study might have underestimated the level of rental discrimination (Verhaeghe 2022). Secondly, we discuss only the intersect between a few discrimination grounds. Future research should include the intersect with other grounds, like income, sexual orientation, family status or disability. Also the discrimination grounds used for this research could be elaborated in future research. We, for example, solely consider rental candidates with Polish and Moroccan names, although people with other ethnic backgrounds are also confronted with discrimination. Thirdly, we hypothesize that a mixed name is an indicator for the nearness towards the ethnic majority group. However, whether this nearness is perceived as cultural, social or generational, should be subject of future research. Fourthly, by means of correspondence testing we can solely measure the degree of discriminatory behaviour, but not infer the motives underlying that behaviour. Lastly, perceived discrimination remains in the shadow, as the focus of this study is on objectively measured discrimination. Future research could focus on experiences of perceived discrimination, though preferably considering an intersectional lens.

Acknowledgements

This work was supported by FWO under Grant S004119N. Ethical approval was granted by the ethical commission of human science of the authors institution under reference number ECHW_289. The principle of informed consent was waived by the ethical committee as this might bias the correspondence tests. The two first authors contributed equally to this manuscript. Authors were ordered alphabetically.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by Research Foundation Flanders [grant number S004119N].

Appendix

AppendixTable A1. Multilevel logistic regression analysis on the odds to be invited for a viewing for the intersection on gender, ethnic origin and homogeneity (n = 16.490).

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10
	OR (SE)	OR (SE)	OR (SE)	OR (SE)	OR (SE)	OR (SE)	OR (SE)	OR (SE)	OR (SE)	OR (SE)
Intercept	0.409 (0.135)***	0.248 (0.141)***	0.126 (0.172)***	0.179 (0.140)***	0.149 (0.171)***	0.206 (0.176)***	0.168 (0.159)***	0.270 (0.219)***	0.204 (0.169)***	0.302 (0.175)***
Belgian – homogenous – man	(ref.)	1.651 (0.073)***	3.238 (0.122)***	2.287 (0.066)***	2.752 (0.120)***	1.982 (0.123)***	2.430 (0.101)***	1.516 (0.178)*	2.007 (0.118)***	1.355 (0.119)*
Belgian – homogenous – woman	0.606 (0.073)***	(ref.)	1.961 (0.121)***	1.385 (0.084)***	1.667 (0.119)***	1.200 (0.130)	1.472 (0.102)***	0.918 (0.184)	1.216 (0.117)	0.821 (0.128)
Moroccan – homogenous – woman	0.309 (0.122)***	0.510 (0.121)***	(ref.)	0.706 (0.129)**	0.850 (0.160)	0.612 (0.164)**	0.751 (0.147)*	0.468 (0.210)***	0.620 (0.146)***	0.419 (0.160)***
Moroccan – homogenous – man	0.437 (0.066)***	0.722 (0.084)***	1.416 (0.129)**	(ref.)	1.204 (0.128)	0.867 (0.133)	1.063 (0.111)	0.663 (0.185)*	0.878 (0.125)	0.593 (0.125)***
Moroccan – mixed – woman	0.363 (0.120)***	0.600 (0.119)***	1.177 (0.160)	0.831 (0.128)	(ref.)	0.720 (0.160)*	0.883 (0.134)	0.551 (0.207)**	0.729 (0.157)*	0.492 (0.161)***
Moroccan – mixed – man	0.505 (0.123)***	0.833 (0.130)	1.634 (0.164)**	1.154 (0.133)	1.389 (0.160)*	(ref.)	1.226 (0.144)	0.765 (0.207)	1.013 (0.161)	0.684 (0.166)*
Polish – homogenous – woman	0.411 (0.101)***	0.679 (0.102)***	1.332 (0.147)*	0.941 (0.111)	1.132 (0.134)	0.815 (0.144)	(ref.)	0.624 (0.197)*	0.826 (0.144)	0.558 (0.148)***
Polish – homogenous – man	0.660 (0.178)*	1.089 (0.184)	2.137 (0.210)***	1.509 (0.185)*	1.816 (0.207)**	1.308 (0.202)	1.604 (0.197)*	(ref.)	1.324 (0.207)	0.894 (0.210)
Polish – mixed – woman	0.498 (0.118)***	0.823 (0.117)	1.613 (0.146)***	1.139 (0.125)	1.371 (0.157)*	0.988 (0.161)	1.211 (0.144)	0.755 (0.207)	(ref.)	0.675 (0.157)*
Polish – mixed – man	0.738 (0.119)**	1.218 (0.128)	2.389 (0.160)***	1.687 (0.125)***	2.031 (0.161)***	1.462 (0.166)*	1.793 (0.148)***	1.118 (0.210)	1.481 (0.157)*	(ref.)
Rent	1.000 (0.001)***	1.000 (0.001)**	1.000 (0.001)**	1.000 (0.001)**	1.000 (0.001)**	1.000 (0.001)**	1.000 (0.001)**	1.000 (0.001)**	1.000 (0.001)**	1.000 (0.001)**
AIC	79582.296	79,582.296	79,582.296	79,582.296	79,582.296	79,582.296	79,582.296	79,582.296	79,582.296	79,582.296

***p < .001; **p < .01; *p < .05.