# RECORD 01e2f1bf2915117a013dff8a924c07e2ed16f152
# TITLE: Living apart together: The economic value of ethnic diversity in cities
# YEAR: 2018.0  VENUE: Environment and Planning B Urban Analytics and City Science  DOI: 10.1177/23998083221082112
# AUTHORS: J. Bakens; R. Florax; H. D. de Groot; P. Mulder
# PAGES: 16  CANDIDATE TABLE PAGES (1-based): [5, 6]
# ABSTRACT: In consumer cities, the presence and location of immigrants impacts house prices through two channels, which both can be valued positively as well as negatively: (i) their presence and contribution to population diversity and (ii) the creation of immigrant-induced consumer amenities like those associated with ethnic restaurants in terms of both their quantity as well as diversity. We hypothesize that these two mechanisms create a trade-off in which city dwellers want to live apart yet consume together. We use unique microdata of house prices and ethnic restaurants in the city of Amsterdam over the 1996–2011 period and a generalized propensity score (GPS) matching technique for multiple continuous treatments to estimate the trade-off between consumers’ love for ethnic goods and their variety on the one hand, and ethnic residential composition on the other hand. We find evidence for the existence of such a trade-off: proximity to ethnically diverse restaurants indeed compensates for a negative correlation between the presence of immigrants and house prices. This trade-off mostly holds for neighborhoods featuring both a diverse population and a high level of amenities in terms of restaurants.


===== PAGE 1 (score 6, numeric-paren 0, table-words 0, var-words 6) =====

Manuscript
EPB: Urban Analytics and City Science
2022, Vol. 0(0) 1–16
© The Author(s) 2022
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/23998083221082112
journals.sagepub.com/home/epb
Living apart together: The
economic value of ethnic
diversity in cities
Jessie Bakens
Research Centre for Education and the Labour Market, Maastricht University, Maastricht, The Netherlands
Raymond JGM Florax
Department of Agricultural Economics, Purdue University, West Lafayette, IN, USA
Department of Spatial Economics, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands
Tinbergen Institute, Amsterdam, The Netherlands
Henri LF de Groot
Department of Spatial Economics, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands
Tinbergen Institute, Amsterdam, The Netherlands
Peter Mulder
TNO, Amsterdam, The Netherlands
Abstract
In consumer cities, the presence and location of immigrants impacts house prices through two
channels, which both can be valued positively as well as negatively: (i) their presence and con-
tribution to population diversity and (ii) the creation of immigrant-induced consumer amenities like
those associated with ethnic restaurants in terms of both their quantity as well as diversity. We
hypothesize that these two mechanisms create a trade-off in which city dwellers want to live apart
yet consume together. We use unique microdata of house prices and ethnic restaurants in the city
of Amsterdam over the 1996–2011 period and a generalized propensity score (GPS) matching
technique for multiple continuous treatments to estimate the trade-off between consumers’ love
for ethnic goods and their variety on the one hand, and ethnic residential composition on the other
hand. We find evidence for the existence of such a trade-off: proximity to ethnically diverse
restaurants indeed compensates for a negative correlation between the presence of immigrants and
house prices. This trade-off mostly holds for neighborhoods featuring both a diverse population and
a high level of amenities in terms of restaurants.
Corresponding author:
Jessie Bakens, Research Centre for Education and the Labour Market, Maastricht University, Tongersestraat 53, Maastricht
6211LM, The Netherlands.
Email: j.bakens@maastrichtuniversity.nl


===== PAGE 5 (score 20, numeric-paren 0, table-words 3, var-words 12) =====

In addition to the density of immigrants and restaurants, we look at their diversity. Diversity is
measured with the broadly used Hirschman-Herfindahl index, which allows to measure both
richness (the number of groups) and diversity (the distribution across groups).
Analogously to the density measure, we use the relative diversity index, that is, the diversity
relative to the overall diversity of the immigrant population and restaurant diversity in Amsterdam.
A summary of all variables included in the analysis and descriptive statistics can be found in Tables
1 and 2.
The majority of the observations in our data lie between a treatment intensity of 1 and 5 for
immigrant density, between a treatment intensity of 0.9 and 1 for immigrant diversity, and below a
treatment intensity of 1 for restaurants. As Table 3 shows, the correlation between immigrant and
restaurant density is negative and low, but statistically significant (the coefficient is 
0.14), which
indicates that within areas there is no strong correlation between the size of the immigrant pop-
ulation and the number of restaurants. Correlations between the treatments do show that restaurants
are more diverse in areas with more immigrants. The positive correlation between restaurant di-
versity and immigrant density is not investigated in this paper, but used as an assumption. The
statistically significant correlation points towards the fact that immigrant-induced diverse amenities
arise with the presence of immigrants because of a supply as well as demand effect of immigrants of
ethnic goods and services. The data also reveal a positive correlation between restaurant density and
diversity indicating that areas with a higher restaurant density on average have a higher diversity of
restaurants. The same holds for immigrant diversity. Figures 1 and 2 show the spatial distribution of
the treatments, specifically the treatment intensities. From these maps, it can be seen that the areas
that have both a large immigrant population share and many restaurants (in addition to high
immigrant and restaurant diversity)) are limited to the neighborhoods just outside the city center—
where, compared to the city center, immigrant density starts to increase, restaurant density starts to
decrease, and restaurant diversity is highest. Insofar as we would find a trade-off between immigrant
density and restaurant density it is most likely to happen in these neighborhoods because the joint
Table 1. Descriptive statistics.
mean
st. dev.
min.
max.
Price per square meter in e
3435
1000
770
12,190
Relative immigrant densitya
4.05
2.16
0.05
13.05
Relative restaurant densitya
1.28
1.47
0.08
9.11
Relative immigrant diversitya
0.85
0.15
0.14
1.02
Relative restaurant diversitya
0.91
0.16
0.18
1.20
Share social rent pc6
0.24
0.32
0
1
Share owner occupied pc6
0.43
0.29
0
1
Population density km2
10,930
4346
322
20,510
Distance to Amsterdam CS station (meter)
3617
2,01
250
1162
Distance to top10 (meter)b
32,500
18,167
11,000
108,000
Distance to nearest metro station (meter)
1660
925
21
4889
Data source: OIS and NVM.
aWe use the diversity and density relative to the overall diversity and density of the immigrant population and restaurants in
Amsterdam. Because of this scaling, the distribution of the measures is not normally distributed around 1, which would have
been the case if we had used the measures relative to the density or diversity in the sample. The immigrant density and
diversity are measured in a 250 m radius around the dwelling. The restaurant density and diversity is measured in a 1000 m
radius around the dwelling.
bThe ten main tourist attractions are based on visitor numbers. We use total distance from the dwelling to the ten attractions.
The ten tourist attractions used are the Rijksmuseum, Van Gogh Museum, Anne Frank Huis, Artis Zoo, Stedelijk Museum,
Madame Tussauds, Heineken Experience, Nemo Science Center, Hermitage, and Venustempel Museum.
Bakens et al.
5


===== PAGE 6 (score 27, numeric-paren 6, table-words 4, var-words 13) =====

occurrence of both restaurants and immigrants is necessary for a trade-off. A more detailed dis-
cussion of the distribution of the treatments and the joint spatial occurrence of multiple treatment
intensities can be found in the Supplemental Material.
Method and diagnostic tests
Following the approach of Hirano and Imbens (2004) for GPS and the modification of the estimation
for multiple continuous treatments (Egger and Von Ehrlich, 2013) we estimate the dose-response
Table 2. Dummy variables.
Category
Observations
Inside A10 highway ring
Yes
32, 621
No
7, 519
Garden
Yes
2, 852
No
37, 288
Transaction year
2006
7, 785
2007
7, 640
2008
6, 385
2009
6, 179
2010
6, 279
2011
5, 872
Number of rooms
3 or less
27, 981
4
7, 505
5 or more
4, 654
Type of dwelling
Simple house
153
Canal house
209
Bungalow
29
Villa
94
Ground floor apartment
5, 640
Upper floor apartment
24, 732
Gallery apartment
1, 457
Maisonette
1, 464
Portico flat
3, 035
Ground and upper floor apartment
285
Single-family house
2, 266
Town house
734
Rest category
42
Observations
40, 140
Data source: OIS and NVM.
Table 3. Correlations between relative treatments.a
Immigrant density
Restaurant density
Immigrant diversity
Restaurant density

0.14***
Immigrant diversity
0.45***

0.55***
Restaurant diversity
0.36***
0.29***
0.04***
Data source: OIS and NVM.
aThe statistical significance of the correlation coefficients is indicated by ***, **, * for the 0.01, 0.05, and 0.1 significance
levels, respectively.
6
EPB: Urban Analytics and City Science 0(0)


===== PAGE 7 (score 11, numeric-paren 0, table-words 2, var-words 7) =====

function in three steps. A detailed description is given in the Supplemental Material. In the first step,
the conditional distribution of the treatments given the covariates is modeled. In the second stage,
we estimate the conditional expected house prices given the multivariate treatment intensities and
the estimated GPS. In the final stage, the average potential outcome at different treatment level
combinations is estimated.
The covariates in the selection equation explain about 53 percent of the variation in immigrant
density and 81 percent of the variation in restaurant density.9 The R2-values of the immigrant and
restaurant diversity estimations are relatively low, 25 percent and 37 percent, respectively. Although
the spatial distribution of the treatment intensity of restaurant and immigrant diversity does not
indicate the nature of the omitted variables, there is likely an underlying pattern explaining res-
taurant and immigrant diversity that we cannot perfectly capture with our data. Restaurant diversity
might also be related to market factors such as competition and quality, while immigrant diversity
might relate to other spatial aspects, that we do not observe. For the GPS to obtain unbiased
estimates of the dose-response function, the covariates in the first step need to be normally dis-
tributed. A Henze-Zirkler multivariate normality test of a random 10 percent sample of the errors of
the GPS estimation shows that the conditional treatments are not multivariate normally distributed.
The univariate Q-Q plots of the residuals in the Supplemental Material show that the non-normality
in the data is present for all treatments. The Q-Q plots show high kurtosis for all treatments,
especially in the left tails, but the effect is most apparent for restaurant diversity.
In the second stage, we estimate the conditional expected house prices given the multivariate
treatment intensities and the estimated GPS. The likelihood ratio test shows that including the GPS leads
to a better model specification for estimating the dose-response function, with a χ2 of 1941. We reject the
null hypothesis that the GPS does not improve the specification, with a p-value less than 0.1 percent.
Figure 1. Average per square meter house prices between 2006–2011 in constant 2088 prices.
Bakens et al.
7