Rental Housing Discrimination and the Persistence of Ethnic Enclaves

We conduct a field experiment to show that discrimination in the rental market represents a significant obstacle for the geographical assimilation process by immigrants. We employ the Internet platform to identify vacant rental apartments in different areas of the two largest Spanish cities, Madrid and Barcelona. We send emails showing interest in the apartments and signal the applicants' ethnicity by using native and foreign-sounding names. We find that, in line with previous studies, immigrants face a differential treatment when trying to rent an apartment. Our results also indicate that this negative treatment varies considerably with the concentration of immigrants in the area. In neighborhoods with a low presence of immigrants the response rate is 30 percentage points lower for immigrants than for natives, while this differential disappears when the immigration share reaches 50%. We conclude that discriminatory practices in the rental housing market contribute to perpetuate the ethnic spatial segregation observed in large cities.


Introduction
Upon arrival to a new country immigrants often settle in segregated neighborhoods. Ethnic networks are useful to …nd a job and facilitate the adjustment to the new society (Bartel 1989;Zavodny 1997;Jaeger 2000;Bauer et al. 2002Bauer et al. , 2005. As the newcomers or their descendants assimilate -…nd a steady job, accumulate some wealth and form families -they may be willing to move out of the ethnic enclave. A di¤erent address in a less segregated neighborhood may signal that the immigrant family has improved economically and socially. However, a wellestablished empirical regularity is that immigrants in advanced societies tend to live spatially concentrated within large cities (Bartel 1989;Borjas 1998).
The most common theories to explain the formation of ethnic enclaves are based on the fact that immigrants prefer living near people with similar tastes and who speak the same language (Cutler et al. 1999). Hence, the concentration of immigrants in particular areas is demand driven. However, it has also been suggested that the native's behavioral response towards immigration may contribute to the surge of ethnic enclaves (Card et al. 2008 and Saiz and Wachter 2011). The literature has identi…ed two main mechanisms. First, natives may be willing to move to all-native neighborhoods and pay a premium to avoid immigrants (decentralized discrimination). Second, natives can …nd ways to e¤ectively restrict immigrant location choices to certain areas (centralized discrimination). This paper investigates the role of rental housing discrimination for the persistence of ethnic enclaves. To isolate the e¤ect that discriminatory practices have in determining residential sorting we conduct a …eld experiment where native and immigrant candidates apply to vacant rental apartments announced on the Internet in the two largest Spanish cities, Madrid and Barcelona. We employ Moroccan and Spanish-sounding names in the applications to signal the ethnicity of the candidate. 1 We then compare the response rate di¤erentials between the two groups across areas with di¤erent concentration of immigrants to identify the extent to which rental housing discrimination represents a barrier for the geographical assimilation process.
Our results uncover a signi…cant negative correlation between the immigration share in a particular neighborhood and the degree of discrimination against Moroccan applicants. That 1 By nationality the most numerous groups of immigrants come from Romania (14.2%), Morocco (12.7%), Ecuador (7.4%) and Colombia (5.2%). Source: Spanish Statistical O¢ ce, Local Population Registry, 2009. We restrict our analysis to Moroccan immigrants as their names, as opposed to those of Ecuadorians and Rumanians, are clearly distinguishable from those of natives.
2 is, discrimination against immigrants is particularly intense in areas where there are very few immigrants. In particular, the response rate to applications signed with a Moroccan-sounding name is, on average, 18 percentage points lower than to those signed by natives. However, in all-natives neighborhoods this di¤erential would increase up to 30 percentage points. As the share of immigrants increases the di¤erential treatment decays. Accordingly where this share is around 50%, immigrants and natives have the same probability of being contacted. 2 This result is robust to the inclusion of ‡at and applicant's socioeconomic characteristics. While we do not claim that discriminatory practices are behind the creation of ethnic enclaves, our evidence strongly suggests that discrimination in the rental market can perpetuate the spatial ethnic segregation pattern observed in large cities.
The bulk of previous studies on ethnic enclaves and segregation have been conducted in the US. A recent study by Cutler et al. (2008) examines the residential integration of the foreign born between 1910 and 2000. Using decennial census data they document that segregation declined in the …rst part of the century, but has been rising over the past few decades. They argue that the increase responds mainly to two forces: First, recent immigrants arrive from countries with greater cultural distinction from the US natives and thus a higher propensity to segregate. Second, the rise of low-density suburban residence and employment areas forces socially and economically marginalized groups to live close to the public transit grid or potential carpool-mates. In addition, Saiz and Wachter (2011) document that natives have preferences for avoiding immigrant areas and are willing to pay a premium to live in predominantly native neighborhoods. 3 Card et al. (2008) provide further support for the view that segregation is driven at least in part by preferences of white families over the racial and ethnic composition of neighborhoods. They show that when the minority share in the neighborhoods exceeds a certain level, which ranges between 5% and 20% (i.e. the tipping point), all the white families leave.
Native preferences for spatial ethnic segregation or the so called "white ‡ight" can be viewed as a form of discrimination in highly mobile societies where residents are willing to change neighborhood as the minority population increases. However, in societies less prone to geographical 2 Similar results are obtained when the share of all immigrants is replaced by the share of only Moroccan immigrants. 3 Saiz and Wachter (2011) …nd that within metropolitan areas increases in a neighborhood's immigrant share are associated with lower housing price appreciation.
3 mobility, like Spain, negative attitudes toward the minority group may lead to other discriminatory practices that range from charging immigrants higher housing prices to limiting their housing search to speci…c areas.
A number of papers have used housing price di¤erentials to measure the extent of discrimination (see for example Bailey 1966, Yinger 1978or Chambers 1992. While in the 1960s there was evidence that African-Americans paid more for equivalent housing in US cities and metropolitan areas, this premium had entirely disappeared by 1990. Another approach to quantify the extent of discrimination is based on experimental audit studies to test the behavior of real estate and rental agents. These studies suggest some degree of discrimination against Hispanics in terms of the quality, price and quantity of housing units o¤ered (Yinger 1995).
In this paper we examine the extent to which natives are able to impose barriers to the geographical assimilation of immigrants by limiting the supply of rental housing units. Economic reasons and network e¤ects seem to be responsible for this regional concentration immigrants in 2008 was 31%, while it was less than 15% in the residential areas located in the north of the city (see Figure 1). Di¤erences in immigrant concentration across districts are even more pronounced in Barcelona (see Figure 2).
Spain hosts immigrants from a variety of ethnic origins. The bulk of the immigration ‡ow, however, comes from Latin America (30%), Eastern Europe (20%) and North Africa (13%).
Because we employ the soundness of the name to signal ethnicity, our experimental study focuses only on Moroccan immigrants whose names are clearly distinguishable from those of natives. The high concentration of immigrants in certain areas and the high premium they seem to pay for it may respond to the existence of natives'preferences for non-integrated neighborhoods.
Some studies suggest that natives' animosity against immigration goes beyond its economic impact (Dustman and Preston 2007; Mayda 2006). After all, most of the e¤ects of immigration can also be attributed to international trade, outsourcing or o¤-shoring. The singularity of 6 Rental prices in 2008 are obtained from the website used to conduct our experiment (www.idealista.com). 7 The Spanish National Immigrant Survey collects information about the socioeconomic characteristics of a representative sample of immigrants living in Spain in 2007. The survey is conducted by the Spanish Statistical O¢ ce (INE, www.ine.es). 8 We run the same regression using only the sample of Moroccans in the National Immigration Survey and obtain similar results.
6 immigration is the physical presence of foreigners in the host country. Accordingly natives may view immigrants not only as a threat to their labor market prospects but also to the country's cultural and social life. We …nd support for this conjecture in the "Attitudes towards Immigration" supplement of the European Social Survey, 2008. Figure 4 plots the distribution of natives' opinions towards immigrants in terms of their economic (short dashed line) and cultural (long dashed line) impact. The distribution of responses suggests that a substantial fraction of the population fears the economic consequences of immigration, but also that a non negligible one thinks that immigration tends to undermine a country's cultural life. The solid line indicates that natives'support to immigration is limited as a non-trivial fraction of them thinks that immigrants have made of Spain a worse rather than a better place to live.
By reducing the supply of rental housing units available to immigrants natives can e¤ectively block their entrance to certain areas. Next we examine the presence of this type of discrimination and its contribution to the persistence of ethnically segregated neighborhoods.

Experimental Design
Our experimental approach is similar to that in other studies that have attempted to identify We use the email correspondence testing method to examine the chances of natives and immigrants to rent a ‡at in areas with di¤erent concentration of foreign born population.
Written applications are sent to rental vacant apartments advertised on www.idealista.com, which is the leading real estate website in Spain. 9 On this platform, private owners and real estate agencies can advertise properties for sell or rent. For private owners, the …rst ad is free of charge. Fees for agencies start at a minimum of 79 Euros per month. In contrast, individuals interested in a particular housing unit can send an electronic application containing the name, email address and a short message at no cost.
In our experimental setup, the potential tenants applied to all rental ads published by 9 According to this website almost 50% of people in Spain use the Internet to search for housing. Popular press such as The New York Times, The Telegraph, The Wall Street Journal and The Washington Post, identi…es idealista.com as the biggest Spanish online property advertising site (http://www.idealista.com/pagina/ranking).
private owners on idealista.com between December 2009 and June 2010. For each housing unit, the site contains information on the rental price per month, the exact address, the number of rooms, the size in squared meters and, in most cases, the name and, therefore, the gender of the person placing the ad. Each week, we collected information on available ‡ats on Tuesdays and sent the applications on the next day. One week later we recorded whether emails sent by the …ctitious applicants received a response. Those candidates invited to visit the apartment or to provide additional information politely declined the invitation. Previous studies show that information about the socioeconomic characteristics of the candidates a¤ect discriminatory practices. Accordingly, we send emails containing di¤erent amount of information about the occupation of the candidate. We consider two types of candidates: (1) an applicant who sends an email showing interest in the ‡at and without any information other the name; (2) an applicant whose email contains information about his/her highly reliable job and therefore represents the ideal tenant for property owners (i.e. university professor or banking clerk).
Our …ctitious applicants sent the Spanish version of the following emails: We use a random assignment procedure, where each vacant apartment is contacted by only one of the eight applicants. 10 We apply to 1186 apartments, and each type of applicant applied, approximately, to 150 apartments.
The focus of this paper however is not on the average discrimination that immigrants are subject to, but on how discrimination varies across neighborhoods with di¤erent concentration of immigrant population. From the Internet platform we obtain the complete address where each vacant ‡at is located. 11 We match this information with the geographical concentration of immigrants obtained from the Spanish Local Population Registry. 12 In particular, we employ 10 Hence we follow the methodology used by Ahmed et al. (2010). Alternatively, we could have used a matching procedure as in Bosch et al. (2010), where all property owners receive inquiries from all applicants. 11 The websites used in previous studies to investigate rental market discrimination do not contain the address of the housing units, hence it is not possible to conduct the type of analysis that we propose here. 12 The Registry is conducted at the municipality level and it provides a very accurate measure of immigrant 9 two levels of spatial disaggregation at the city level: the census district and the ZIP or postal code, being the latter a more disaggregated spatial subdivision. 13 Barcelona and Madrid add up to 31 census districts and 70 ZIP codes.
The randomness in our experimental design ensures that both immigrants and natives apply on average to similar apartments and hence, the di¤erential treatment that we observed is only attributable to the soundness of names. Tables A1 and A2 in the Appendix provide evidence on the validity of our randomization exercise. These tables present the mean di¤erences (and standard errors) in ‡at characteristics between rental units contacted by natives and immigrants. We do not …nd any systematic di¤erences in the type of ‡ats that the two groups apply for. 14  suggests that the response rates increase when positive information about the socioeconomic status of the applicant is revealed. Finally, there is evidence that this information reduces the response rate di¤erential between natives and immigrants, from 23.6 percentage points for those applicants without information to 15.45 for those in high-paying occupations.

Results
The main result of the paper is illustrated in Figure 5. We plot, by ZIP code, the di¤erential response rate in favor of natives against the share of immigrants in that particular ZIP code.
concentration, including undocumented immigrants. The reason is that registration is required in order to have access to public healthcare and education, but also to be eligible in the event of an amnesty. The process of registration does not require proof of legal residence and the data are con…dential (that is, cannot be used to expel undocumented migrants). Thus immigrants have strong incentives to register. 13  Positive numbers in the y-axis indicate that emails signed with a native-sounding name obtain a higher response rate than those signed with a foreign-sounding one. The …gure also displays the …tted values from regressing the response rate di¤erential on the share of immigrants, weighted by number of observations at the ZIP level. Although, arguably, there is some noise in the data, a negative relationship emerges, indicating that as the share of immigrants increases in a particular area rental housing discrimination decreases. This evidence suggests that while many factors are likely to be responsible for the geographical concentration of immigrants, the presence of arti…cial barriers to their mobility may contribute to the persistence of ethnic enclaves in large cities.
We next estimate a set of econometric models to investigate the statistical signi…cance of the previous evidence. Let us …rst discuss the results for our baseline discrimination model.
Following previous studies we run a regression to estimate the probability of being contacted as a function of a set of socioeconomic characteristics including the applicant's ethnicity: where C i is an indicator variable that takes value 1 if the applicant is contacted and 0 otherwise; Img i is an indicator that takes value 1 if the email is signed with a foreign-sounding name; This is evidence of the large penalty that male immigrants face in the rental housing market.
Column (4) estimates the same model including ‡at characteristics and, as expected, the results are una¤ected.
Next we study how the discriminatory behavior changes with the amount of information disclosed in the application. The model in column (5) contains as additional regressors the information dummy and its interaction with the immigrant indicator to capture di¤erences in the response rate between "high-quality" candidates and those who do not provide any information about their socioeconomic status. According to our estimates candidates signaling a high-paying occupation are 8 percentage points more likely to be contacted than those who do not report any information about their jobs. The interaction of this variable with the immigrant indicator suggests the presence of some additional informational premium for immigrants of around 8 percentage points, which is statistically insigni…cant. However when this informational premium is interacted with the gender indicator (see column (6)), the returns to information becomes positive and statistically signi…cant for male immigrants. 15 Note that despite this positive premium information does not eliminate the di¤erence in response rate between natives and immigrants.
In all, the results in table 2 con…rm the previous …ndings in the literature. Agents in the rental market use the informational content of names to di¤erentially treat immigrants. This di¤erential treatment is substantially larger for males and it does not disappear when information about the socioeconomic status of the candidate is revealed. This last result indicates that either information other than the socioeconomic status is relevant for the property owners or that negative attitudes towards immigrants are behind the substantial amount of discrimination observed in the rental market. 16 immigrants are on average 30 percentage points less likely to be contacted than natives. However this di¤erential decreases as the presence of immigrants in the area increases. In particular, a 10 percentage points increase in the immigration share at the ZIP code level increases the chances of being contacted (relative to those of natives) by 5.5 percentage points. Accordingly discrimination will disappear in areas where the concentration of immigrants is around 50%.
The remaining columns in table 3 investigate the robustness of the previous result. Column (2) adds ZIP code …xed e¤ects to control for unobserved characteristics that may a¤ect the probability of being contacted. However, due to the experimental nature of our design the results are unaltered. Column (3) investigates the e¤ect of outlier observations. According to is larger, but this is due to the fact that the mean and the variance of this share are smaller. 18 The point estimate suggests that a 1 percentage point increase in the share of Moroccans at the 17 The results using a probit model instead of a linear probablity model are extremely similar and are available upon request. 18 The share of Moroccan immigrants in the sample has mean 1.14 and standard deviation 0.97. The share of all immigrants in the sample has mean 22.05 and standard deviation 9.93.

13
ZIP code level, increases the chances of response to an email signed by a Moroccan applicant by 5 percentage points. This e¤ect is large and reinforces the view that while several factors may be responsible for the important geographical concentration of immigrants in certain areas of big cities, part of this segregation responds to discriminatory practices in the rental housing market. In particular, property owners through the Internet platform seem to be e¤ectively blocking the supply of housing units immigrants have access to.

Robustness Checks
We now investigate the e¤ect of a series of confounding factors that could threat the validity of our previous results. One possibility is that the quality of ‡ats in areas with few immigrants is di¤erent from that in areas with a high concentration of them. If the quality/price of the ‡at determines the level of discrimination (maybe because of the owner's risk aversion) this could be causing the observed correlation. We investigate this possibility in the …rst column of table 5. We allow the coe¢ cient on the interaction between the share of immigrants and the immigrant indicator to vary by ‡at characteristics. None of those interactions is statistically signi…cant and our main result remains invariant, suggesting that the reason for the observed spatial pattern is not that discrimination occurs in expensive/high-quality ‡ats that happen to be in areas where there are few immigrants.
The geographical pattern of discrimination that we observe may also respond to the fact that, because of urban segregation, owners in areas with a high concentration of immigrants are immigrants themselves and less prone to discriminate against those of their own kind. However, the immigration phenomenon in Spain is relatively recent and originates mainly from low income countries. Hence, the home ownership rate among immigrants is relatively low. According to the National Immigrant Survey (2007) this rate is around 30%. Thus it is unlikely that a substantial share of the immigrant population is operating on the supply side of rental market and driving our results. We can actually test this hypothesis with our data. We have the name of approximately 80% of the property owners or renters in our sample, either because they were advertising it in the rental ad or because they would sign the reply email. With this information we can infer the nationality of the owner and test whether it is responsible for the observed discriminatory patterns. In our sample 85% of all the owners (for which we have names) have a Spanish-sounding name. We then compute the share of "non-Spanish"owners by ZIP code and 14 interact it with the immigrant indicator. The results for this speci…cation appear in column (2) of table 5. We do not …nd any signi…cant e¤ect for this variable suggesting that our results identify the behavior of native owners.
Another possible explanation is rooted in the fact that attitudes towards immigration may di¤er across neighborhoods. In Spain the more conservative parties have been strong supporters of laws restricting immigration. They were also against the general amnesty in 2005, whereby immigrants who had lived in Spain for at least one year, had a job and no criminal record could obtain a residence permit. We argue that the share of "conservative" vote in the 2008 general election in a particular neighborhood is a good proxy for the views against immigration in that area. 19 Information on the electoral results is only available at the census district level. To see whether this higher level of aggregation a¤ects our results in column (3)  interacted with the immigrant indicator. This allows us to study whether more conservative districts discriminate more. We do not …nd support for this hypothesis in our data as the new interaction is statistically insigni…cant. In contrast, the coe¢ cient on the interaction between the concentration of immigrants and the immigrant indicator variable slightly falls from 0.56 to 0.52, but remains signi…cant at the 10% level of signi…cance. Thus we conclude that political preferences are not responsible for the persistence of ethnic enclaves.
Finally, we discuss two possible channels that can explain the correlation between discrimination and immigrant concentration by studying the evolution of the latter during the last decade. Again we use the census district as a unit of analysis since we only have data on the past spatial concentration of immigrants at this level. The idea is to relate the increase in immigrant concentration in a particular neighborhood with the current level of discrimination.
We employ as explanatory variable the growth in the share of immigrants by district between 2000 and 2008 interacted with the immigrant indicator. Column (5) shows that there is a very strong correlation between the increase in the immigrant population and the current levels of discrimination. In particular, a 1 percentage point increase in the stock of immigrants in one district is associated to a fall in discrimination of 0.85 percentage points. One possible explanation for this result is that districts discriminating more in 2008 were also over discriminating in 2000, thus generating a lower in ‡ux of immigrants. Alternatively, one could argue that immigrants moving into certain districts brought in new information and increased acceptance of the foreign born population. Unfortunately our data do not allow us to disentangle these two explanations. The former would suggest that discriminatory practices are crucial for shaping spatial segregation. The latter would imply that as immigrants move into particular areas, assimilation reduces discriminatory practices.
On the whole, our results indicate that the degree of discrimination varies substantially with the ethnic composition of the neighborhood. We …nd evidence that property owners or renters discriminate more in areas with a higher concentration of natives. Although possible there are other factors determining ethnic segregation at the city level, we can conclude that the presence of rental housing discrimination contributes to the existence of ethnic enclaves in large cities and makes the geographical assimilation process of immigrants an arduous task.

Conclusions
In this paper we conduct a …eld experiment to show that discrimination against immigrants in the rental market is strongly correlated with their spatial concentration in the two largest Spanish cities, Madrid and Barcelona. Our estimates indicate that in areas with very few immigrants the di¤erential in response rates between natives and immigrants reaches a magnitude of 30 percentage points. As the concentration of immigrants increases, this di¤erential is reduced. In particular, a 10 percentage points increase in the share of immigrants at the ZIP or postal code level increases the chances that an immigrant will be contacted by the property owner or renter by 6 percentage points (relative to their native counterpart). We also show that this spatial pattern does not respond to di¤erences in the quality and price of ‡ats, political preferences, or ethnic origin of the owners across geographical areas.
These results do not allow us to conclude that discriminatory practices generated the current distribution of immigrants across neighborhoods. Probably other factors, like housing prices and immigrants'preferences to live close to each other played a substantial role in shaping the 16 spatial distribution we observed today. Nonetheless our results show that, even if other forces would have been responsible for triggering ethnic segregation, the discriminatory behavior of property owners and renters would have created persistency once segregation started and thus prevent the geographical assimilation of immigrants.       Flat characteristics include price per squared meter, number of rooms and city …xed e¤ects.
The estimates correspond to a linear probability model.