The Long-term impact of the resettlement of the
Sudetenland on residential migration*
Martin Guzi, Peter Huber, and Štěpán Mikulat
March 29, 2021
We analyze the long-term impact of the resettlement of the Sudetenland after World
War II on residential migration. This event involved expulsion of ethnic Germans and
an almost complete depopulation of an area of a country and its rapid resettlement by
2 million Czech inhabitants. Results based on a regression discontinuity design show
a highly persistent higher population churn and thus a lower attachment of residents
to their region in resettled areas. Descriptive evidence also indicates that resettled
settlements still have fewer local club memberships, less frequently organize local
social events and had lower turnout in municipal elections until the 1990s. This thus
suggests persistently lower levels of local social capital. This finding is consistent with
recent theoretical models that suggest a highly persistent impact of the destruction of
local social capital on residential migration.
Keywords: Migration, Social Capital, Sudetenland
J E L Codes: N44, Z10, R23, J15
Declaration of interest: none.
*Financial support by the Czech Science Foundation (grant No. 18-1611 IS) is gratefully acknowledged.
We thank Alžběta Hanáčková, Barbora Kuklová and Marek Breza for excellent research assistance and
Martin Halla, Benjamin Eisner, Tomáš Dvořák and Peter Egger, as well the participants of the 2017 research
seminars at the University of Economics in Prague and the University of Stavanger, 2017, the 10th
Geoffrey
J.D: Hewings Regional Economics Workshop in Vienna, the 2018 Slovak Economic Association Meeting,
the 2018 Scottish Economic Society Annual Conference, the 2019 ISWG workshop in Vienna, the 2019
GFR Winterseminar in Matrei and the 2019 Annual Meeting of the Austrian Economic Association, the
33r d
Annual Conference of the European Society for Population Economics in Bath, the 2019 European
Economic Association Annual Congress in Manchester, and the 2019 European Association of Labour
Economists Conference in Uppsala for helpful comments. Computational resources were supplied by
the project "e-Infrastruktura CZ" (e-INFRA LM2018140) provided within the program Projects of Large
Research, Development and Innovations Infrastructures.
^Martin Guzi: Department of Public Economics, Masaryk University. Address: Lipová 41a, Brno,
Czech Republic. Email: martin.guzi@econ.muni.cz, Peter Huber: Austrian Institute for Economic Research.
Address: Arsenal Objekt 20, Wien, Austria. Email: peter.huber@wifo.ac.at, Štěpán Mikula: Corresponding
author. Department of Economics, Masaryk University. Address: Lipová 41a, Brno, Czech Republic. Phone:
+420 724 512 946. Email: stepan.mikula@econ.muni.cz.
1
1 Introduction
A growing body of economic and political science research documents that the disruption of
social structures caused by forced mass emigration has long-lasting and sizeable effects on
the political attitudes (Acemoglu et al. 2011; Grosfeld et al. 2013), institutional, economic
and educational outcomes (Acemoglu et al. 2011; Akbulut-Yuksel and Yuksel 2015; Pascali
2016; Bharadwaj et al. 2008; Testa 2020) and scientific achievements (Waldinger 2010,
2011) of the affected regions as well as on the forced migrants and their descendants
(Becker et al. 2020). This literature (see: Becker and Ferrara 2019, for a survey) suggests
that these impacts are substantially more long-lasting and sizeable than those found in the
related literature on the destruction of physical capital (Brakman et al. 2004; Waldinger
2016). The long-run demographic impact of mass emigration has so far been much less
analyzed. Among the exceptions Schumann (2014) focuses on population growth in a
receiving region to provide evidence of a highly persistent demographic impact of the mass
immigration to West Germany after World War II. The current paper adds to this literature
by using the post-WWII mass expulsion of ethnic Germans from the area of today's Czech
Republic (the so-called Sudetenland) to study the long-term impact of the subsequent
resettlement on migration to and from the sending resettled regions.
This focus may be interesting because migration is generally seen as a behavioural
measure of the attractiveness of a region as a place of residence and the attachment of
its population to the region (see e.g.: Greenwood et al. 1991). It may, also be interesting
because the expulsion and subsequent resettlement we study resulted in an almost complete
destruction of the social structures and networks in the affected region. Recent theoretical
contributions to the social capital literature (see e.g., David et al. 2010; Brauninger and
Tolciu 2011) suggest that such large shocks may have highly persistent effects on the
attachment of residents to the region. In these models there are typically two stable
long-run equilibria: One with low and one with high local social capital. Furthermore,
since individuals derive utility from income and social capital they may forego financially
profitable mobility. As a consequence, the high local social capital equilibrium is associated
with low mobility, and the low local social capital equilibrium with high mobility. A
sufficiently large shock to local social capital can move regions from one of these equilibria
to the other, thereby causing a persistent change in migration rates from and to a region
that goes hand in hand with persistent shift in local social capital levels.
The empirical part of this literature confirms this strong impact of current and previous
social contacts for mobility and suggests that in particular local contacts are an impediment
to mobility, while contacts to other regions enhance it. For example Kan (2007) and
Belot and Ermisch (2009) show that people with contacts in their region of residence are
2
less inclined to move elsewhere, while Biichel et al. (2020) show that mobile individuals
prefer to live in places with more nearby contacts. This literature, however, considers the
contemporaneous correlation between social capital and migration (i.e. the short-run). Yet
recent evidence by Costa et al. (2018) also indicates a persistent long-run impact impact
of social contacts on individual's location decisions, as civil war veterans serving in the
same military unit tend to live close to each other in the long-run. We therefore add to this
literature by analyzing the long-run impact of the resettlement of the Sudetenland, which
caused an almost complete destruction of the social structures and networks, on mobility
and presenting additional descriptive evidence on the long-run development of local social
capital levels in the resettled municipalities.
The resettlement of the Sudetenland is particularly well suited to identify such long-run
impacts due to the size of the population exchange, the extraordinary speed with which
ethnic Germans had to leave and the rapid resettlement that occurred in the follow-up. In
the municipalities we study at least 90% of the population were ethnic Germans, who were
subjected to expulsion and by and large completely replaced by Czech speaking settlers
within five years. With the exodus of the German speakers also all local social networks and
contacts of the resident population were destroyed and the newly arriving population had to
re-establish all social contacts anew. This provides for large scale and credibly exogenous
shock to local social capital in the affected region. In addition the fleeing population had to
leave behind all of their belongings as well as all physical capital, and was rapidly replaced
by settlers, who mostly came from other parts of the Czech Republic and took possession
of the belongings and physical capital left behind. This limits the role of some potential
explanations such as physical capital destruction or cultural or institutional differences,
which have been emphasized as alternative sources for long-term differences in regional
development in previous research (Grosfeld et al. 2013; Becker et al. 2015; Dell 2010;
Alesina and Giuliano 2015).
Finally, this paper also contributes to the still rather scant literature on the resettlement
of the Sudetenland (see e.g., Daněk 1995; Testa 2020). Earlier work on this resettlement
(Daněk 1995) used descriptive evidence on the district level to show that districts of the
Czech Republic that were at least partially located in the Sudetenland, still have a younger,
less well educated, more secularized and ethnically more diverse population than other
districts of today's Czech Republic. More recently, Testa (2020), in the paper most closely
related to this one, uses municipality level data and a spatial regression discontinuity
design based on the borders drawn by the Munich Agreement, to show that former German
municipalities still had a lower population density, higher rates of unemployment, less
skill-intensive industries, and lower levels of education in 2011. He argues that a lack of
agglomeration economies and the erosion of property rights caused by resettlement are
3
potential mechanisms that may have caused these differences. We augment these findings
by a detailed analysis of the impact of resettlement on migration and by proposing an
additional causal mechanism that may have contributed to these very long-run impacts.
We combine a unique municipality-level administrative data set that includes all
permanent residence changes in the years 1971 to 2015 with pre-WWII municipality-level
data on the ethnic composition of municipalities in 1930. This allows us to identify the
causal effects of resettlement on residential migration by comparing the most strongly
affected municipalities with more than 90% ethnic Germans in 1930 (which we refer to as
resettled municipalities) to municipalities with less than 10% ethnic German residents in
1930 (referred to as not resettled municipalities). We use a regression discontinuity (RD)
design identification strategy where the inference is based on a precise definition of the
border between ethnic German- and Czech-dominated areas before WWII. The results
indicate that resettlement led to a long-lasting increase in residential migration to and
from the resettled municipalities that survived such important institutional changes as the
transition from a planned to a market economy, the dissolution of Czechoslovakia and
accession to the European Union, and the many economic changes that occurred in the
Czech Republic in that period. Even at the end of our observation period, in 2015 (i.e., 70
years after the resettlement), emigration and immigration rates in resettled municipalities
were still substantially (by around 20%) higher than among not resettled municipalities. In
addition, at the beginning of the period studied the effects of resettlement on emigration
dominate over those on immigration, such that net emigration from resettled municipalities
initially increased as well. This effect, however, levels off to zero after the mid-1980s,
and is also less robust than the effect on gross-emigration and immigration rates. This
suggests that—consistent with theoretical predictions—resettlement mainly resulted in a
very long-run reduction in the attachment of the population to the affected regions.
To provide additional descriptive evidence on the potential causal mechanisms for this
long-run impact we analyze auxiliary data sets. We explore two mechanisms. The first
one assumes that the original settlers moving into the Sudetenland belonged to the more
mobile groups of the population at the time of resettlement and may have transferred their
values related to mobility to their children, who are the migrants we study at the end of
our observation period. The second is based on the assumption that resettlement led to a
persistent large scale reduction in local social capital of the resettled territory, and thus
moved these regions from the high social capital, low mobility to the low local social
capital, high mobility equilibrium suggested by the theoretical literature.
These analyses show that the population in the resettled and not resettled municipalities
share similar mobility related values and also do not differ in terms of their behaviour with
respect to proxy variables for social capital that is not clearly bound to the locality (such
4
their inclination to give donations to national organisations, as well as their willingness to
attend social events and to participate in voluntary work potentially outside their region of
residence). We, however, also find evidence that resettled municipalities had lower turnout
rates in municipal elections up to the 1990s and have fewer local activities and clubs as well
as lower membership levels in local clubs to this day. This suggests that local social capital
is less well developed in the resettled regions. While therefore we cannot fully preclude
some role for the parental transmission mechanism in explaining the long run impact of
resettlement on migration decisions, we, therefore, argue that—consistent with many of the
descriptions of these municipalities by historians (Glassheim 2006; Vaněk 1996; Čapka
et al. 2005; Spurný 2011; Matějka 2008) and with theoretical models of the impact of local
social capital on migration decisions—persistently lower local social capital in resettled
municipalities is likely to have at least contributed to the persistently higher residential inand
out-migration in these municipalities.
The remainder of the paper is structured as follows. Section two provides the historical
background of the resettlement. Section three presents the data. Section four introduces
the identification strategy used. Section five describes the results. Section six discusses the
mechanism driving the results, and section seven concludes.
2 Historical background
2.1 Ethnic Germans in the Czech Republic
Germans had been settling in the area of today's Czech Republic since the 13t h
century and,
except for a short period from 1938 to 1945 under German occupation, their territory of
settlement was always administered by the same state as the rest of today's Czech Republic.
Compared with the Jewish population studied in the literature, the ethnic Germans in the
Czech territory never belonged to a discriminated group (e.g., Alexander 2008; Meixner
1988). During the Austro-Hungarian Empire, of which today's Czech Republic was a part
from the 16t h
century to 1918, they tended to be more privileged relative to the Czech
population partly because they spoke the official language of the empire.
The German speaking population was, however, clearly segregated from the Czech
population. According to the last Czechoslovak census prior to World War II the German
speaking population mainly resided in a well defined territory in the north, west, and south
of today's Czech Republic, the so-called Sudetenland (see Figure l ) , 1
and the number
municipalities where both Germans and Czechs resided in equal numbers was rather low
1. The word Sudetenland is used in common language to denote the territories settled by the ethnic
Germans before World War II.
5
(see Figure 2). The differences in economic and population structure between the two
territories were small according to the results of this census. In particular, while the region
where German speakers dominated were less agricultural and more industrial (see Tables
A . l and A.2 in the Appendix) the share of migrants living in the two regions, which is the
only proxy for migration provided in the 1930 census, were rather similar in the Czech and
German speaking parts of today's Czech Republic (see Figure 3).2
Figure 1: Share of ethnic Germans in municipalities in 1930
Shire of ethnic
Germans (To)
1O0
Source: CZSO, own calculations.
Note: The ethnic German population is defined according to the primarily spoken language. The 1930
municipality-level data on ethnic Germans is harmonized with 6,168 municipalities defined in the 2011
census by using matching rules provided by the CZSO.
Historic accounts also suggest that the relationships between the two population groups
were relatively unproblematic for most of the time. Ethnic tensions between Czechs and
Germans arose only in the second half of the 19t h
century and continued after the break-up
of the Austro-Hungarian Empire, when ethnic Germans comprised 29.5% of the population,
according to the 1930 population census (see C Z S O 2014). Historical records document
a number of complaints from German representatives during the interwar period inter
alia about limited access to employment in the state bureaucracy, the closing of German
language schools, and the asymmetric impact of land reforms in the 1920s. Yet, according
to many accounts (e.g., Glassheim 2000) minority policy in Czechoslovakia was one of the
most liberal in Central and Eastern Europe at the time. Ethnic tensions severely intensified
2. Data on employment structure and share of migrants are not available at municipal level in 1930 census.
Therefore, we use data aggregated at the level of 330 judicial districts. For the definition of 1930 judicial
districts see Figure A.l in the Appendix.
6
Figure 2: Distribution of municipalities by the share of ethnic Germans in 1930
0 25 50 75 100
Share of ethnic Germans (%)
Source: CZSO, own calculations.
Note: The ethnic German population is defined according to the primarily spoken language. The 1930
municipality-level data on ethnic Germans is harmonized with 6,168 municipalities defined in the 2011
census by using matching rules provided by the CZSO.
only with the economic crisis in 1933 and the increasing popularity of German nationalist
political parties. Under the Munich Agreement in 1938 the Sudetenland was annexed by
the German Reich and remained under German rule until the end of WWII.
2.2 Expulsion of the Germans and resettlement
In the aftermath of WWII, the ethnic Germans were held responsible for the Nazi atrocities
and considered traitors. This perspective ultimately led to their expulsion from the country,
which started with the end of WWII in May 1945 and proceeded in two phases. The initial
phase, referred to as "wild expulsion", was poorly organized and controlled within a vague
legal framework. Up to 800,000 Germans left the country in this phase until the autumn
of 1945 (Wiedemann 2016). The second more organized phase continued from January
to October 1946 and followed the agreements of the Potsdam Conference. The mass
deportation during the second phase reduced the prewar population of ethnic Germans of 3
million to 200 to 300 thousand (Gerlach 2017). The share of ethnic Germans in the Czech
population decreased from 29.5% (based on the 1930 census) to 1.8%3
based on the first
postwar census in 1950 (CZSO 2014).
3. Data on ethnicity from the 1930 and 1950 censuses are not fully comparable due to methodological
changes. Nevertheless, estimates of historians are very similar to the 1950 census data. According to Stanek
(1991) there were 216,545 inhabitants of non-Slavic origin in 1947. The remaining Germans were also
subjected to an internal relocation policy (Dvorak 2013).
7
Figure 3: Share of immigrants from other regions in total population injudicial districts in
1930
[0,10] (10,90]
Share of ethnic Germans (%)
(90,100]
Source: CZSO, Urbánní a regionální laboratoř (URRlab) UK (2015), own calculations.
Note: The ethnic German population is defined according to the primarily spoken language. For definition of
1930 judicial districts see 330 see Figure A.l in the Appendix.
The process of resettlement occurred in parallel with the expulsion and was also rather
rapid. Initially, people were encouraged through newspapers and radio broadcasting to
seize German properties and were supported by Czech soldiers, militias, and security
forces (Glassheim 2000). During the wild expulsion, (i.e. within the first year of expulsion)
between 500,000 and 900,000 new settlers arrived in the Sudetenland.4
In addition
Wiedemann (2016) states that the most massive inflows of settlers occurred until 1947 (i.e.
one year after the beginning the second phase of deportation and two years after the end of
WWII) and that resettlement was largely over by the end of 1952, with only modest inflows
continuing on until the end of the 1950s.5
Similarly, Gerlach (2010) finds that almost 2
million new ethnic Czech settlers had arrived to Sudetenland by May 1947, while earlier
estimates by Radvanovsky (2001) suggest an influx of 1.5 million over the first 2 years of
resettlement. In sum thus resettlement was rather rapid and was clearly already completed
in 1971 when our period of analysis starts.
4. Radvanovsky (2001) estimates that 514,515 settlers arrived in Sudetenland by September 16, 1945.
Wiedemann (2016) states that their numbers reached 696,554 by mid-October and 862,706 by the end of
1945.
5. This extraordinary speed of resettlement is also documented in a number of anecdotes according to
which new settlers ended up cohabiting with those to be expelled when deportation trains were delayed (e.g.
Wiedemann (2016), p. 106).
8
2.3 The process of property acquisition by new settlers
For the Czech citizens resettlement to the Sudetenland offered a unique opportunity to
improve their economic and social status by acquiring a house and small piece of land
(the official limit was 0.13 square kilometers)6
and thus also seizing the expellees' other
property,7
as well as to obtain a better job, or to become a national administrator8
of
seized properties (Wiedemann 2016). The key institutions for assigning property to the
new settlers were about 100 local resettlement committees ("Osidlovacikomise"). They
were responsible for collecting applications and redistributing land and property. They
were also responsible for appointing national administrator for the vacated enterprises.
Wiedemann (2016) reports that, given the "spontaneous" nature of the wild expulsion
the work of the resettlement committees was rather complicated, inconsistent and also
intransparent. Committees had only limited knowledge of the situation in general and
occasionally allocated properties that were already taken and also caused many other
irregularities. Settlers had to pay for the acquired property. Prices were low, and amounted
to one to three yearly rents. Ten percent of the total price was due at the time of property
acquisition. The remainder was payable in the following 15 years. Settlers could pay in
cash or kind and a substantial part of the liabilities was never paid (Wiedemann 2016).
The process of property redistribution with respect to enterprises and arable land was,
however, also rather inconsequential, as the communist party decided to collectivize the
land and the industrial property throughout the Czech Republic after the 1948 coup d'etat.
This process started in 1949 and was by and large completed in the mid-1950s (i.e. shortly
after resettlement had been completed). After this the Czech Republic remained to be a
country with one of the lowest shares of private property of land and firms throughout
the communist era, even among C O M E C O N countries, and private ownership of "means
of production" was virtually non-existent until the early 1990s. The decisions of the
resettlement committees were therefore relevant only for home-ownership, as this was the
only part of the property of the expelled ethnic Germans that remained in private hands
after the collectivization.
6. Čapka et al. (2005) document that 91% of the settlers in the villages around Mikulov previously owned
land with acreage less than 0.03 square kilometers, and the typical acreage redistributed to settlers around
Mikulov was 0.05 to 0.08 square kilometers.
7. The expellees could keep 30 kg and later 50 kg of their belongings (excluding valuables) during the
more organized phase of the expulsion (Gerlach 2017).
8. This was a trustee who could manage an enterprise (or large farm) on behalf of the state with the
prospect of becoming the owner (Gerlach 2017).
9
2.4 Characteristics of settlers
The weak regulation (and often chaotic nature) of the resettlement makes it difficult to
map the socioeconomic characteristics of settlers. Wiedemann (2016), Čapka et al. (2005)
and Školí (1983) show, however, that settlers were young, often married couples, landless
persons, small farmers, second-born children (with low prospects for family inheritance),
or individuals who had worked in the civil services or non-agricultural sector. It is also
highly likely that the settlers belonged to the more mobile groups in population.
The resettlement policy aimed to attract people primarily from areas that were
geographically close and climatically similar in order to increase the chances of settlers
establishing economically and socially functional communities (Wiedemann 2016).
Historical research documents that settlers almost exclusively moved from other regions
of today's Czech Republic9
and over rather short distances. For example, Školí (1983)
documents that only 16% of the settlers in Břeclav district moved less than 10 km, and
39% moved more than 100 km (Figure A.2 and Table A.3 in the Appendix). Settlers likely
moved in smaller groups, however. Detailed statistics on the original municipalities of
settlers in the Břeclav district reveal that 12% of settlers were from the same miunicipality
of origin and 29% of settlers originated from the four most important municipalities (Table
A.4 in the Appendix). In addition, Čapka et al. (2005) describe Sobotin in North Moravia
as a municipality with a large group of settlers from the same original municipality, but
even in this example, the settlers from the same original municipality comprise around
15% of the 1950 population.
These major migratory movements resulted in two types of municipalities. Not resettled
municipalities were inhabited by ethnic Czechs before WWII and were therefore not
subject to expulsion and resettlement. The resettled municipalities were inhabited by ethnic
Germans before WWII and thus lost most of their original population together with their
human and social capital. The resettlement process brought new inhabitants to the emptied
municipalities, who were allowed to seize the property left behind by the expelled but had
to re-establish all local social contacts anew. In consequence resettlement also implied a
massive destruction of local social capital in the resettled regions.
9. Školí (1983) notes that in the Břeclav district, located at the border with Slovakia and Austria, 90.3°/c
of settlers were Czech, 2.1% were Slovak, and 7.6% were of other nationalities (Table A.5 in the Appendix).
10
3 Data
3.1 Migration and population data
We investigate how the resettlement of Sudetenland affected the subsequent migration
behavior in the resettled municipalities. The immigration and emigration rates as the key
dependent variables of this analysis are taken from administrative records of permanent
residence changes provided by the Czech Statistical Office (CZSO) for the period 1971 to
2015.1 0
These rates provide the number of movers from and to municipalities as a percentage
of the population on January 1 and apply to a period when according to historical sources
resettlement had clearly been completed, with the German-speaking population having
been gone from the municipality and the country for 20 years or more.1 1
This data is highly
reliable because residents of former Czechoslovakia (and of the current Czech Republic)
are legally obliged to register changes of their permanent address. This registration also
defines the constituency in municipal elections and is associated with preferential access to
local public services such as healthcare, elementary schools, kindergartens, and subsidized
accommodations for university students. We obtain the pre-WWII share of ethnic Germans
in each municipality from the 1930 population census (the last census before WWII), in
which ethnicity is defined according to the language primarily spoken in the household.1 2
We use this data to identify resettled and not resettled municipalities and to construct the
ethnic border.
Table 1 presents descriptive statistics of the pooled sample. We report averages and
standard errors of migration variables, population, and the share of ethnic Germans in
1930 separately for the resettled and not resettled municipalities. The emigration rates
were on average higher in the 1970s and 1980s compared with later periods, and the
immigration rates are more stable over time. Both emigration and immigration rates are
consistently higher in the resettled municipalities relative to not resettled ones throughout
the observation period, but this difference narrows in later periods. Figure 4 adds to this
information by presenting the size distribution of resettled and not resettled municipalities
for individual census years. It shows that although resettled municipalities were in general
larger prior to resettlement, their size reduced substantially after resettlement and they
10. Data is available at www.czso.cz/csu/czso/databaze-demografickych-udaju-za-obce-cr. We have data
for 6,168 municipalities due to some records being lost prior to digitization and due to definition changes;
this is an unbalanced panel. The baseline model is estimated with pooled data on all municipalities, and we
address concerns that relate to missing observations in the robustness analysis in the Appendix to the paper.
11. Annual population data are obtained from the CZSO. We remove obvious outliers (i.e., the top 1% of
emigration and immigration rates) from the data. In the robustness analysis we show that the inclusion of
outliers slightly increases the estimated effects.
12. Data from the 1930 census was digitized by the authors (see Appendix B). Matching rules provided by
the CZSO were applied to harmonize municipalities such as in the 2011 census.
11
also continued to loose population until the 1980s. Afterwards the size distribution of
both types of municipalities remained rather stable and the population of both types of
municipalities was also rather similar.
Figure 4: The evolution of the size distribution of resettled and not resettled municipalities
(1930 to 2011)
1930 1950 1961 1970 1980 1991 2001 2011
I—j—I Not resettled municipalities
| - H Resettled municipalities
Source: CZSO, own calculations.
3.2 Local social capital and values data
In addition, to provide evidence on potential causal mechanisms, we collected data on values
and different forms of social capital at a municipality level. This includes administrative
data on voter turnout (i.e., the percentage of eligible voters who cast a ballot) in all free
and voluntary municipal elections held in the Czech Republic provided by the CZSO. This,
following the social capital literature (Knack 1992; Hotchkiss and Rupasingha 2018), is
used as a proxy for local social capital. Further this includes survey data on values, civic
participation, and charitable activities collected in 2003 and 2004, respectively.1 3
These
survey data are, according to our review of the sources, the only publicly available data in the
Czech Republic that allow us to examine shared norms and values as well as social capital at
the municipality level. The 2003 survey (Majerova et al. 2003) was conducted in two waves
and provides information on both local social capital as well as forms of social capital not
13. This data is described in more detail in the Appendix B of this paper.
12
Table 1: Descriptive statistics
Period
All
Municipalities
Resettled Not resettled
Difference
(2)-(3)
Period (1) (2) (3) (4)
Municipalities (n) 6,168 751 4,808
Share of ethnic Germans (%) 1930 18.43 95.92 0.48 95.43***
(0.45) (0.10) (0.02)
Population (n) 1930 1,725.54 1,997.30 1,319.73 677.57**
(172.51) (117.27) (202.09)
Emigration rate (%) 1971-1979 3.50 5.05 3.19 1.85***
(0.008) (0.03) (0.008)
1980-1989 3.11 4.34 2.88 1.46***
(0.008) (0.03) (0.008)
1990-1999 2.44 3.12 2.29 0.83***
(0.006) (0.02) (0.007)
2000-2009 2.37 3.09 2.21 0.88***
(0.006) (0.02) (0.007)
2010-2015 2.51 3.26 2.34 0
(0.008) (0.03) (0.008)
Immigration rate (%) 1971-1979 2.38 3.38 2.14 1 24***
(0.009) (0.03) (0.009)
1980-1989 2.35 3.14 2.19 0.95***
(0.008) (0.03) (0.009)
1990-1999 2.41 3.21 2.24 0.97***
(0.008) (0.03) (0.009)
2000-2009 3.03 3.64 2.90 0.74***
(0.009) (0.03) (0.01)
2010-2015 3.05 3.53 2.94 0.59***
(0.01) (0.03) (0.01)
Net immigration rate (%) 1971-1979 -1.12 -1.67 -1.05 -0.62***
(0.01) (0.04) (0.01)
1980-1989 -0.76 -1.20 -0.69 -0.51***
(0.01) (0.03) (0.01)
1990-1999 -0.02 0.09 -0.05 0.14***
(0.009) (0.03) (0.01)
2000-2009 0.67 0.55 0.70 -0.14***
(0.009) (0.03) (0.01)
2010-2015 0.54 0.28 0.61 -0.33***
(0.01) (0.04) (0.01)
Note: Columns (1) to (3) contain means and standard errors (in parentheses). Column (4) contains differences
in means between resettled and not resettled municipalities and their statistical significance witht
= p < 0.1,
* = p < 0.05, **=/?< 0.01, ***=/?< 0.001.
13
bound to the region of the resettled and not resettled municipalities and their inhabitants.
In the first wave, mayors of 1,324 municipalities were asked about the abundance of certain
forms of local social capital in the municipality (i.e the frequency of local events organized
by local clubs and local green activities). In the second wave, 1,287 residents in 223
municipalities were asked about memberships in local clubs, participation at social event
events and donations, and volunteer work in potentially national organisations. Among
these variables the first is a measure of local social capital, while the others pertain to
aspects of social capital that are not necessarily bound to the municipality of residence. The
2004 survey (Majerova et al. 2004) asked 1,518 respondents in 220 municipalities about the
importance of values in their life (i.e., nature and environment, a job, relationships, faith
and spiritual values, hobbies, housing, friendship, family life and children, and material
conditions).
3.3 Geographical data
Since our identification strategies require geographic information, we geocode all municipalitylevel
data using the reference points defined by the C Z S O 1 4
and obtain altitude and terrain
roughness data for each municipality using remotely sensed elevation data from Shuttle
Radar Topography Mission (SRTM) 1 Arc-Second Global.1 5
Altitude is measured as the
elevation at the reference point. Terrain roughness is calculated as the average of terrain
roughness index around the reference point1 6
such that higher values indicate rougher
terrain.
4 Identification strategy
Formerly German-dominated municipalities were clearly segregated from the Czechdominated
municipalities and mainly located in the borderlands of today's Czech Republic.
This segregation as well as the specific location of formerly German-dominated municipalities
makes direct comparison of resettled and not resettled municipalities problematic as they
could differ in unobserved characteristics that are correlated with their respective status
(such as local labor market conditions, market access, etc.). This may lead to O L S
estimation leading to biased estimates. To address this concern we identify causal effects by
14. These are placed in the social center of municipalities (i.e., typically in front of the town hall or church)
15. The SRTM data for the Czech Republic are transformed into approximately 30 x 40 meter tiles, which
is the maximum homogeneous resolution available for the SRTM data of the Czech Republic. The terrain
roughness index is calculated as the mean of the absolute differences between the altitude of a tile and the
altitudes of its eight surrounding tiles (Wilson et al. 2007).
16. We use a 1.6 km radius throughout because this is the median radius of municipalities in the Czech
Republic.
14
using a RD-design (Dell 2010; Becker et al. 2015; Egger and Lassmann 2015; Oto-Peralias
and Romero-Avila 2017). We estimate the following specification:
yit = y R M , + /3Zit + f(di) + <pr + <pt + <ps + ft* (1)
where yit is one of the three outcome variables (emigration, immigration, and net
immigration rates) defined for municipality i in year t, R M , is an indicator variable
for a resettled municipality, and Z,- includes geographical variables (log of altitude, the
terrain roughness, the log of shortest distance to the Czech border and second order
polynomials of the degree of latitude and longitude of the municipality). The year fixed
effects <pt are included to control for changes in institutions and economic developments
affecting all municipalities alike, <pr are region fixed effects that control for any regionspecific
features of the outcome variable (or equivalently for the border segment).1 7
As
the migration rates tend to be slightly higher in larger municipalities (see Figure A.4 in
the Appendix) we also include population fixed effects <ps (decile of the 1930 population).
These account for potential differences in migratory moves between municipalities of
different sizes.1 8
Variable rjn is the error term.
To define the ethnic border, we treat all municipalities with a minority (less than 50%)
of ethnic Germans in 1930 as Czech municipalities and all municipalities with a majority
of ethnic Germans in 1930 as German municipalities.1 9
The ethnic border is then located at
the border of German municipalities contingent to a Czech municipalities. This delineation
yields a compact Czech territory in the center of today's Czech Republic surrounded by
what once were German municipalities in the south, west, and north of the country and
a few former German enclaves surrounded by Czech municipalities and Czech enclaves
surrounded by former German municipalities. From these, we removed two enclaves
(one resettled and one not resettled municipality, respectively) that were smaller than 10
square kilometers to ensure that we could define a border region sufficiently wide for
an RD-regression. The distance of municipalities to the ethnic border is defined by the
distance of their reference points to the ethnic border.
One issue with respect to the RD-design based identification strategy is that we measure
distance to the border by defining municipality centers that are rarely located at the defined
border. This implies that we have only few municipalities in the vicinity of the border (see
17. The Czech Republic is administratively divided into 14 regions ("kraj"). We checked that the borders
of regions do not coincide with the ethnic border (Figure A.3 in the Appendix).
18. Population in 1930 is used as a control to avoid including endogenous controls and to account for the
fact that the once larger municipalities in the Sudetenland may provide more infrastructure and cultural
amenities than municipalities outside the Sudetenland.
19. We use the administrative areas of municipalities from a series of maps (ArcCR 500 Version 3.3)
developed by ARCDATA PRAHA to construct the ethnic border. These are also used for all geospatial
visualization (https://www.arcdata.cz/produkty/geograficka-data/arccr-500).
15
Figure 5: The ethnic border between Czech- and German-dominated municipalities and
municipalities within a 15 km band around the border
Source: CZSO, own calculations.
Figure 6). As pointed out by e.g. Cunningham (2021) this can make RD-estimates based
on non-parametric optimal bandwidth estimators unreliable and inrobust. In our baseline
estimates we therefore rely on a RD-approach that includes all resettled and not resettled
municipalities located within 15 kilometers of the ethnic border to ensure symmetry in
the number of observations at certain distances on both sides of the ethnic border (Figure
5). In our robustness section we, however, also provide estimates based on local linear
regressions as well as on a different identification strategy based on spatial matching, to
assess the impact of the border definition and the chosen RD-estimator on results.
The central identification assumptions of the regression discontinuity design are that
(1) all unobserved confounding variables affecting migration decisions vary smoothly at
the ethnic border but that (2) the share of ethnic Germans has a distinct jump at that border.
Figure 7 therefore shows discontinuity plots for the geographic variables and the share of
the German-speaking population in 1930 living on each side of the ethnic border for the
estimation sample as the only municipality level pre-treatment variables for which we have
data. There are no visible discontinuities in altitude, terrain roughness, distance to the
Czech border, and population in 1930, but there is a sizeable discontinuity in the share of
Germans at the cut-off.
16
Figure 6: Distance to ethnic border
-1.0-0.5 0.0 0.5 1.0
Distance to ethnic border (km)
Source: CZSO, own calculations.
Note: Resettled municipalities have negative distances to the ethnic border, and not resettled municipalities
have a positive one. Municipalities are aggregated to bins of 0.1 km.
5 Results
5.1 Baseline results
Figure 8 shows similar regression discontinuity plots for the three dependent variables for
the pooled data over all time periods in the analysis. This figure suggests a distinct jump in
the emigration an immigration rates as well as the net immigration rate at the ethnic border.
According to this figure municipalities on the former German speaking side of the border
were characterized by in average higher emigration and immigration rates as well as lower
net immigration rates over the years 1971 to 2015.
While these graphs are indicative of a potential causal effect of the resettlement on
mobility, the estimates they present fail to control for other confounding factors that may
have contributed to different migration levels between the two regions and also pool across
the entire time period from 1971 to 2015. In particular the later fact may be problematic on
account of the data in the early observation period also including the subsequent mobility
of the settlers, originally moving into the region. A s these settlers almost by definition
belonged to the more mobile groups of the Czech population at the time of resettlement
this may bias the causal estimates upward. Table 2 therefore presents several baseline
estimates of the parameter of interest (y) of Equation (1). The first panel (of two columns)
presents estimates for the average mobility rates over the entire observation period using
the complete complete panel information, while the second panel, to reduce potential issues
with serial correlation of error terms, focuses on the average mobility rate for the entire
17
Figure 7: Discontinuity plots of geographic variables, population, and share of ethnic
Germans
(a) Share of ethnic Germans in 1930 (b) Altitude
-10 -5 0 5 10
Distance to the ethnic border (km)
15 -10 -5 0 5 10 15
Distance to the ethnic border (km)
(c) Terrain roughness (d) Distance to the country border
Distance to the ethnic border (km) Distance to the ethnic border (km)
(e) Population in 1930
15 -10 -5 0 5 10
Distance to the ethnic border (km)
Source: CZSO, SRTM, own calculations.
Note: Resettled municipalities have negative distances to the ethnic border, and not resettled municipalities
have a positive one. Municipalities are aggregated to bins of 0.5 km represented by points. The smoothing
lines are parametric linear predictions and shaded areas represent 95% confidence intervals.
period. Panels three and four by contrast present estimates for two different sub-periods
(1971 to 1989 and 1990 to 2015). Further each of these panels presents estimates for
two different RD-polynomials (in their two columns) using standard errors clustered by
18
Figure 8: Discontinuity plots of migration rates at the ethnic border
(a) Emigration rate (b) Immigration rate
(c) Net immigration rate
_ ^ 0.4-
1 1 1 1 1 1 1
—15 -10 -5 0 5 10 15
Distance to the ethnic border (km)
Source: CZSO, own calculations.
Note: Resettled municipalities have negative distances to the ethnic border, and not resettled municipalities
have a positive one. Municipalities are aggregated to bins of 0.5 km represented by points. The smoothing
lines are parametric linear predictions and shaded areas represent 95% confidence intervals.
municipality for panel estimates and robust standard errors for estimates using aggregated
data.
The estimates focusing on the entire period, irrespective of whether they apply to
the full panel or averages over the period, suggest that resettlement increased emigration
rates from the resettled municipalities by 0.7 percentage points and immigration rates by
approximately 0.5 to 0.6 percentage points per year on average. Both these effects are
statistically significant at the 0.1% level in all estimations. In addition, according to the
results net immigration decreased by 0.1 to 0.2 percentage points per year on average, with
these effects being statistically significant only when considering panel estimates.
19
Results, however, differ for the sub-periods from 1971 to 1989 and 1990 to 2015. In
the first period the effects on emigration rates are substantially and effects on immigration
rates slightly higher than for the overall period and also effects on net immigration are
negative and statistically significant at the 1% significance level. In this earlier period
resettlement led to a substantial increase of both emigration (of 1.2 to 1.3 percentage
points) and immigration (0.6 to 0.7 percentage points) and also caused net immigration
rates to these regions to fall (or equivalently net emigration from these regions to rise).
This may, however, be due to the original settlers belonging to the more mobile groups of
their population and some of them potentially also having been disappointed with their
new live in the resettled municipalities.
20
Table 2: Impact of the resettlement on migration rates
Panel estimates Cross-sectional estimates using data aggregated by period
1971-2015 1971-1989 1990-2015
RD-polynomial RD-polynomial RD-polynomial RD-polynomial
1s t
order 2 n d
order 1s t
order 2 n d
order 1s t
order 2 n d
order 1s t
order 2 n d
order
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent variable: Emigration rate (%)
Resettled municipality (==1) 0.735*** 0.690*** 0.685*** 0.664*** 1.303*** 1.196*** 0.474*** 0.400***
(0.063) (0.092) (0.066) (0.096) (0.105) (0.156) (0.062) (0.090)
Adjusted R 2
0.169 0.169 0.444 0.444 0.453 0.454 0.359 0.359
Observations 82423 82423 2104 2104 1882 1882 2104 2104
Dependent variable: Immigration rate (%)
Resettled municipality (= 1) 0.504*** 0.465*** 0.573*** 0.506*** 0.589*** 0.675** 0.512*** 0.417**
(0.070) (0.102) (0.084) (0.122) (0.129) (0.207) (0.090) (0.129)
Adjusted R 2
0.091 0.091 0.320 0.319 0.268 0.269 0.314 0.314
Observations 82423 82423 2104 2104 1882 1882 2104 2104
Dependent variable: Net immigration rate (%)
Resettled municipality (==1) -0.231*** -0.225* -0.111 -0.159 -0.714*** -0.521** 0.038 0.017
(0.062) (0.092) (0.083) (0.121) (0.133) (0.198) (0.085) (0.126)
Adjusted R 2
0.096 0.096 0.102 0.102 0.234 0.235 0.124 0.124
Observations 82423 82423 2104 2104 1882 1882 2104 2104
Note: Estimates from Equation (1). Only municipalities within a 15 km band around the ethnic border are considered. Results control for year fixed effects, an
RD-polynomial, region fixed effects, population fixed effect, log of altitude, terrain roughness, log of the distance to the country border, longitude, latitude, and squared
longitude and latitude squared. Symbols represent statistical significance of differences in means between resettled and not resettled municipalities witht
= p < 0.1,
* = p < 0.05, ** = p < 0.01, *** = p < 0.001. Values in brackets are standard errors clustered by municipality.
Somewhat more surprisingly the increased mobility also applies to the 1990 to 2015
period, where the original adult settlers were over 60 years old and the majority of migration
moves was accounted for by the children of the original settlers. In particular while the
effects of resettlement on net migration are statistically insignificant and also very close
to zero for this period, emigration and immigration rates in the resettled municipalities
still exceeded those in not resettled municipalities by 0.4 to 0.5 percentage points. This
suggests that in this very long-run, resettlement mainly resulted in increased population
churning and thus a reduction in the attachment of the population to the affected regions.
5.2 Effects over time
Figure 9 extends these findings by plotting estimates of year-treatment interaction terms,
based on a linear RD-specification.2 0
It, therefore, provides a more complete picture of
the dynamics of the effects of resettlement on migration. The effects once more indicate
a particularly pronounced increase in emigration and immigration rates in the treated
municipalities in the 1970s and 1980s. In the 1970s, emigration rates were as much as 2
percentage points and immigration rates as much as 1.5 percentage points higher in the
resettled municipalities (relative to a baseline emigration rate of 3.2% and an immigration
rate of 2.1% in the not resettled municipalities). The overall effect on net immigration in
this period was negative, that is, resettled municipalities' population decreased because of
emigration.
By the 1990s the gap in net immigration rates, however, narrowed and stabilized
at a level comparable to those of the not resettled municipalities for the remainder of
the observation period. B y contrast, gross emigration from the resettled municipalities
remained statistically significantly higher than in not resettled municipalities in all years
except for 2013. Furthermore, while these differences declined during the 1970s and
1980s, they remained at approximately 0.4 percentage points higher in the resettled than in
not resettled municipalities thereafter. Similar observations apply to immigration rates,
although these differences are not statistically significant in all years.
Thus while impacts on net migration rates levelled off during the 1980s, the impact on
population churning persisted over a period over which the original settlers in the resettled
areas went through almost their entire life cycle, as an individual born at the beginning of
resettlement in 1945 would have been 70 years old in 2015. This stability of the estimates
for the effects on immigration and emigration from the 1990s on, implies that the effect was
affected neither by the many political and institutional changes experienced by the Czech
Republic in that time period nor by the rapidly changing macroeconomic environment.
20. This is chosen because it minimizes the Bayesian Information Criterion (BIC).
22
Figure 9: Impact of the resettlement on migration rates by year
(a) Emigration rate
2.5
2.0
-7 1.5
OH
s & i.o
a a 0 . 5
s a
<£ .S o.o
Q m - 0 . 5
I -1.0
-1.5
-2.0
2.5
2.0
d 1.5
d
a " 1.0
§1 0.5
£ -a o.o
S .5? -0.5
-1.5
-2.0
2.5
—, 2.0
| 1.5
• S i '-o
g g 0-5
u -a
£ 2 o.o
EC
Q S -0.5
1
'Z -i.o
w
-1.5
-2.0
(b) Immigration rate
(c) Net immigration rate
Note: Figure reports estimates of the impact of the resettlement on the migration rates from Equation (1).
Only municipalities within a 15 km band around the ethnic border are considered. Results control for
the interaction of the year fixed effects with the dummy variable for resettled municipality, interaction of
year fixed effect with a first-order polynomial forcing function, and region fixed effects, year fixed effects,
population fixed effect, log of altitude, terrain roughness, and log of the distance to the country border. Points
represent the point estimates and the associated bars the 95% confidence intervals of the estimates. Standard
errors are clustered by municipality.
23
There are no visible changes in the causal impact at the time of fall of the Iron Curtain
and the start of economic transition (1989), dissolution of Czechoslovakia (1993), and the
accession to the European Union (2004). Similarly, the estimates are very stable throughout
the deep (so-called transition) recession at the beginning of the 1990s, the Czech currency
crisis in 1997 and the following recession, the economic crisis in the late 2000s, and the
great depression of 2009.
5.3 Effects by education groups
Table 3 further extends these finding, by using data on changes in permanent residence by
education groups available from C Z S O for the years 1993 to 2004, to estimate Equation
(1) separately for the highly and less highly educated.2 1
This data pertains to a shorter
time period than that used in baseline estimates.2 2
Nonetheless it indicates that the
impact of resettlement on population churning in the later period of analysis was more
pronounced among the more highly educated, who according to most results in the social
capital literature (see Putnam 1995; Helliwell and Putnam 1999; Glaeser et al. 2002) both
contribute and depend less on social capital.
Thus, resettlement led to a sizeable and statistically highly significant increase in
emigration rates (of around 2.1 percentage points) of the highly educated. Similarly their
immigration rates increased 2.8 to 3.8 percentage points. Among the less educated effects
were more modest, as their emigration rate was by 0.4 to 0.5 percentage points higher in
the resettled than in not resettled municipalities and their immigration rates were by 0.4 to
0.6 percentage points higher. Results for net immigration rates, by contrast, are once more
much less robust for both the highly and the less educated and statistically significant only
for the highly educated, when using a first-order RD-polynomial.
5.4 Falsification tests and robustness
One concern regarding these findings is that—as aforementioned—the resettled municipalities
are located closer to the German border. This could threaten the causal interpretation
of our results if municipalities closer to the border of the country are in general less
attractive for living. This may be of particular relevance in the context of communist
Czechoslovakia before 1989 because in those times, due to heavy policing of the border,
borders to "Western countries" were particularly unattractive places to live. We therefore
conduct two falsification tests to address this concern. In these we counter-factually shift
21. Less highly educated are persons with compulsory education and highly educated with more than
compulsory education.
22. There are no annual municipality-level data on population by education. Therefore, in this analysis we
calculate yearly migration rates using 1991 census data on population in each education group.
24
Table 3: Impact of the resettlement on migration rates by education
Sample
Primary education Secondary and
tertiary education
RD-polynomial RD-polynomial
1s t
order 2 n d
order 1s t
order 2 n d
order
(1) (2) (3) (4)
Dependent variable: Emigration rate (%)
Resettled municipality (= 0.533*** 0.368** 2.139*** 2.058***
(0.079) (0.114) (0.341) (0.472)
Adjusted R 2
0.077 0.078 0.044 0.044
Observations 25188 25188 25164 25164
Dependent variable: Immigration rate (%)
Resettled municipality (==D 0.605*** 0.372f
3.772*** 2.841**
(0.173) (0.208) (0.733) (0.993)
Adjusted R 2
0.042 0.042 0.062 0.062
Observations 25188 25188 25164 25164
Dependent variable: Net Immigration rate (%)
Resettled municipality (= 0.072 0.004 1.632** 0.783
(0.147) (0.177) (0.575) (0.824)
Adjusted R 2
0.008 0.008 0.016 0.016
Observations 25188 25188 25164 25164
Note: Panel estimates for Equation (1) for period 1993-2004. Only municipalities within a 15 km band
around the ethnic border are considered. Results control for year fixed effects, an RD-polynomial, and region
fixed effects, population fixed effect, log of altitude, terrain roughness, log of the distance to the country
border, and 2n d
order polynomials of longitude and latitude. Symbols represent statistical significance of
differences in means between resettled an not resettled municipalities with ' = p < 0.1, * = p < 0.05,
** = p < 0.01, *** = p < 0.001. Values in brackets are standard errors clustered by municipality.
25
the ethnic border of the Sudetenland 5 kilometers away from, respectively, 5 kilometers
toward the border of the Czech Republic in sub-samples of resettled and not resettled
municipalities. These placebo treatments (Table 4) indicate no statistically significant
differences between municipalities at the counterfactual ethnic border for emigration and
net immigration rates irrespective of where the border is shifted. The point estimates
level off to zero for the counter-factual shift of the ethnic border away from the border of
the Czech Republic and are only slightly larger in case of counterfactual shift toward the
country border. However, none of these estimates is statistically significant.
To address other concerns about the baseline estimates we conduct a series of further
robustness checks based on panel data estimates. These address issues related to the potential
impact of spatial auto-correlation of error terms on the results (see Kelly 2019; 2020, for a
discussion), the choice of the estimation method for the regression RD-specification and
other potential methodological choices made in the estimation of the baseline specification
The main results of these robustness tests are presented in Table 5 and more detailed
descriptions are available in the Appendix C, which also reports results for more settings
and estimates with higher order RD-polynomials.2 3
To assess the potential impact of spatial auto-correlation on results we adopted the
correction suggested by Kelly (2020) and additionally implemented a permutation test
where the empirical distribution of treatment effects under the null hypothesis is obtained
from 1000 placebo treatments resulting from counter factually shifting the ethnic border
between 20 and 35 k m away from the border of the Czech Republic.2 4
The results
(reported in the first two rows of Table 5) suggest only minor increases in standard errors
for immigration and net immigration rate, which do not affect statistical significance, for
estimates based on the spatial auto-correlation corrected standard errors and also no change
for the inference based on permutation tests.
Further to assess the impact of the RD-estimation method on results we re-estimated
the regression in Equation (1) using more flexible specifications allowing for different
RD-polynomials on both sides of the cut-off (by allowing for all combinations of first
to third-order polynomials on both sides of the border) and also considered results
from estimating local linear regressions based on several optimal bandwidth selection
procedures.2 5
Results for estimates based on different functional forms of the RD-Polynomials
at the two sides of the border (presented in the Appendix in Table C.3) do not differ from
23. Equivalent baseline results estimated with 3r d
order RD-polynomials are presented in Table A.6 in the
Appendix.
24. Artificial shifts in the other direction cannot be considered, because the ethnic border lies too close to
the border of the Czech Republic. Consequently this test rests on the assumption that the auto-correlation
structure at the ethnic border is identical to that the counter factually constructed border.
25. These included using the Mean Square Error (MSE) and CER-optimal bandwidth selector to define
optimal bandwidth.
26
Table 4: Placebo tests
Placebo test
Placebo test on the sample
of resettled municipalities
(shift of the ethnic border
by 5 km towards
the country border)
RD-polynomial
1s t
order 2 n d
order
Placebo test on the sample
of not resettled municipalities
(shift of the ethnic border
by 5 km outwards
the country border)
RD-polynomial
1s t
order 2 n d
order
(1) (2) (3) (4)
Resettled municipality (=1)
Adjusted R 2
Observations
Resettled municipality (=1)
Adjusted R 2
Observations
Resettled municipality (=1)
Adjusted R 2
Observations
Dependent variable: Emigration rate (%)
0.100 0.054 0.010 0.002
(0.121) (0.142)
0.227 0.227
15070 15070
Dependent variable:
0.163 0.179
(0.126) (0.148)
0.070 0.070
15070 15070
(0.042) (0.053)
0.112 0.112
67353 67353
Immigration rate (%)
-0.006 -0.024
(0.063) (0.086)
0.081 0.081
67353 67353
Dependent variable: Net immigration rate (%)
0.063 0.125 -0.016 -0.026
(0.119) (0.128) (0.059) (0.082)
0.126 0.126 0.091 0.091
15070 15070 67353 67353
Note: Panel estimates from specifications equivalent to Equation (1) with counter-factually shifted ethnic
border. Only municipalities within a 15 km band around the ethnic border are considered. Results control for
year fixed effects, an RD-polynomial, region fixed effects, population fixed effect, log of altitude, terrain
roughness, log of the distance to the country border, longitude, latitude, and squared longitude and latitude
squared. Symbols represent statistical significance of differences in means between resettled and not resettled
municipalities with ' = p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.001. Values in brackets are
standard errors clustered by municipality.
27
Table 5: Overview of robustness tests
Dependent variable
Emigration Immigration Net
rate (%) rate (%) Immigration
rate (%)
(1) (2) (3)
(1) Standard errors adjusted for spatial autocorrelation 0.735*** 0.504*** -0.231***
(0.062) (0.076) (0.068)
(2) Statistical significance determined in a permutation test 0.735*** 0.504*** -0.231***
(0.021) (0.069) (0.068)
(3) Local linear regression 0.209f
0.398*** -0.156^
(0.107) (0.093) (0.091)
(4) Local linear regression: Exclusion of 0.5 km band 0.778*** 0.519*** -0.285***
(0.060) (0.064) (0.068)
(5) Local linear regression: Optimum bandwidth doubled 0.568*** 0.464*** -0.182**
(0.061) (0.054) (0.056)
(6) (Spatial) matching estimates 0.782*** 0.607*** -0.175**
(0.053) (0.061) (0.053)
(7) Defining resettled and not resettled municipalities 0.647*** 0.414*** -0.233***
using a 50°/o cutoff (0.043) (0.048) (0.044)
(8) Exclusion of a part of ethnic border close 0.659*** 0.456*** -0.203**
to Bavaria (West Germany) (0.064) (0.072) (0.065)
(9) Inclusion of linear time trends for each region 0.738*** 0.505*** -0.233***
and period (before/after 1990) (0.063) (0.070) (0.063)
Note: Table contain estimates of impact of resettlement on residential migration. Rows (1), (2), (7)-(9)
contain estimates from (extended) Equation (1). Only municipalities within a 15 km band around the
ethnic border are considered. Results control for year fixed effects, an RD-polynomial, region fixed effects,
population fixed effect, log of altitude, terrain roughness, log of the distance to the country border, longitude,
latitude, and squared longitude and latitude squared. Rows (3)-(5) contain results obtained from local linear
regression using MSE optimum bandwidth selector. Only municipalities within a 15 km band around the
ethnic border are considered. Results control for yearfixedeffects,first-orderRD-polynomial, region (defined
by nearest foreign country) fixed effects, population fixed effect, log of altitude, terrain roughness, log of
the distance to the country border, longitude, latitude, and squared longitude and latitude squared. Row (6)
contains estimates based on spatial matching strategy. Results control for year fixed effects, matched-pair
fixed effect, log of altitude, terrain roughness, log of the distance to the country border, longitude, latitude,
and squared longitude and latitude squared. Values in brackets are standard errors adjusted for spatial
auto-correlation in row (1); standard deviations of the empirical distributions obtained in permutation
tests in row (2); robust standard errors in rows (3)-(5); standard errors clustered by municipality in rows
(6)-(9). Symbols represent statistical significance of differences in means between resettled and not resettled
municipalities witht
= p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.001.
28
the baseline. Estimates obtained from local linear regressions show a substantial decline in
estimated effects and are highly sensitive to the choice of specification (see row 3 in Table
5 and Table C.4 in the Appendix). This is due to the low number of observation located
directly at the border, which makes local linear estimates based on optimal bandwidth
selection procedures unstable and sensitive to outliers. Thus when we exclude the 0.5 km
band along the ethnic border, which contains 10 not resettled and 7 resettled municipalities,
and adjust distances accordingly the local linear estimates in row (4) converge to parametric
ones. Similarly doubling the bandwidth, that reduces sensitivity to sparse observations in
the proximity of ethnic border, yields point estimates reported in row (5), that are close to
baseline results.
Further, to assess the impact of additional methodological choices we made in
implementing the baseline RD-analysis, we conducted a number of additional robustness
tests. These included implementing an alternative identification strategy based on spatial
matching (Becker et al. 2015)2 6
to test whether our results are corroborated by a method
which does not require a definition of the ethnic-border (see row 6 in Table 5 for results);
defining resettled municipalities as municipalities with a share of ethnic Germans of 50%
or more and all others as being not resettled, to assess the impact of focusing only on
municipalities with more than 90% respectively 10% ethnic Germans on results (see row
7 for results) and excluding the municipalities located close to Bavaria (West Germany)
to avoid potential asymmetric impact from the opening of the border in 1990 on the
attractiveness of the Czech-German border region as a place of residence (see row 8 for
results); controlling for different trends in regional development by extending (1) with
linear time trends for each region and period prior and after 1990 (see row 9 for results).2 7
These additional estimates are once more highly robust and comparable to our baseline
results. Throughout the estimated coefficients for the emigration rate range between 0.7 and
0.9 percentage points, and those for the immigration rate between 0.5 and 0.7 percentage
points. The changes in results are thus mostly smaller than 0.1 percentage points relative to
our baseline estimates. The estimates for the net immigration rate are less robust.
In summary, this evidence is consistent with a strong medium-term causal negative
impact of resettlement on net immigration on new municipalities, which could be associated
with return or onward migration of the original settlers. More important, this evidence
26. Row (6) in Table 5 shows estimation results on within-pair comparison of resettled and not resettled
municipalities that are located within 10 km from each other and did not differ in population by more than
250 in 1930.
27. We also conducted robustness tests to check how ourresults are affected by the ethnic border irregularities
in the northeast of the Czech Republic, the use of a balanced panel data, the exclusion of outliers and the use
of municipality size classes based on the 1950 population as a control affect results. Moreover, we estimate
the baseline specification separately for northern, western, and southern part of the country. These additional
checks provide no further insights over the previous analysis. They are, therefore, reported in the Appendix
to the paper only.
29
suggests a long-lasting effect of resettlement on immigration and emigration rates and
thus the population churn. This implies a persistently lower attachment of people in new
municipalities that continued to persist even 70 years after the beginning of resettlement.
6 Mechanisms
The fact that the fleeing population from Sudetenland left behind all of their belongings
as well as all physical capital, and was replaced by settlers, who mostly came from other
regions of the Czech Republic and took possession of the belongings and physical capital
left behind suggests that cultural differences, differences in administrative rules and effects
from the destruction of physical capital, which have been emphasized as alternative sources
for long-term differences in regional development in previous research (Grosfeld et al. 2013;
Becker et al. 2015; Dell 2010; Alesina and Giuliano 2015), are unlikely to explain the
long-run impacts found.
As a consequence we consider the plausibility of two alternative potential causal
mechanism for the long-term reduction in the attachment of the population to their
region of residence caused by resettlement. The first, is based on the self-selection of
settlers to resettled municipalities: Since the Sudetenland was in all likelihood resettled
by the more mobile groups among the Czech population at the time, it could be that
migration-related values were transferred across generations among the population of the
resettled municipalities such that these are still populated by more mobile population groups
to this day. The second, accords with many of the descriptions of the resettled municipalities
by historians (Vaněk 1996; Glassheim 2006; Matějka 2008; Čapka et al. 2005; Spurný
2011) and is derived directly from the predictions theoretical models of the impact of social
capital on migration decisions (e.g., David et al. 2010; Bráuninger and Tolciu 2011). It
assumes that the destruction of local social capital in resettled municipalities could have
moved these municipalities to a low social capital and high mobility equilibrium.2 8
Although, due to data constraints, we cannot provide a causal inference with respect
to these mechanisms, we examine a number of further data sets (described in Section 3),
to assess their plausibility. These provide information on the differences in values of the
population of resettled and not resettled municipalities and their activities related to local
social capital (the participation in local elections, the number of clubs and local green
activities in a municipality and membership in local clubs) as well as to forms of social
28. The empiricalfindingsby Charnysh (2019) also supports the plausibility of this causal mechanism.
She studies the impacts of the settlement process that followed post-WWII westward shift of Poland and finds
that districts settled by migrants heterogeneous in place of origin are lower in prevalence of volunteer fire
brigades, characteristic linked to local social capital level, as of 1989.
30
capital that are not bound to a specific locality (such as attending social events in general
and volunteering to work with or to give donations to potentially national aid organisations).
6.1 Values
Regarding social values, we examine data from a survey on the values of the old (aged
60-74 years) and young (aged 18-29 years) generation in the Czech Republic {Český
venkov 2004: život mladých a starých lidí"). We run a linear regression on the responses
to the value-related questions in this survey on a dummy variable that takes the value
of one if the respondent resided in a resettled municipality and zero if the respondent
resided in an not resettled municipality. In these regressions we control for gender, age,
previous migration, education, labor market status, marital status, household size and
municipality population, and county fixed effects.2 9
The results (Table 6) suggest no
statistically significant differences in the values between the resettled and not resettled
municipalities among the old or young residents. This finding applies to all values except
the importance attributed to hobbies by old inhabitants; however, this could just as well be
caused by a type II error.
29. Due to the smaller number of municipalities covered in this survey the regression discontinuity design
approach is feasible - see Figure B.l in the Appendix.
31
Table 6: Differences in values between resettled and not resettled municipalities
Things and values very important in personal life (=1)
Nature, Job, Relationships Faith, Hobbies Housing Friendship Health Family life Material
environment occupation spiritual values and children conditions
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Regression results
Panel A: Young cohort (18-29)
Resettled municipality (=1) 0.049 0.113 -0.032 -0.029 0.076 0.042 0.072 -0.046 -0.062 0.004
(0.075) (0.080) (0.084) (0.046) (0.077) (0.075) (0.066) (0.090) (0.087) (0.088)
Adjusted R2
0.039 0.050 0.043 0.039 0.021 0.018 0.037 0.044 0.226 0.016
Observations 680 680 680 680 680 680 680 679 680 679
Panel B: Old cohort (60-74)
Resettled municipality (=1) -0.019 0.030 -0.045 -0.015 0.231*" 0.006 0.004 -0.078f
-0.076 0.108
(0.087) (0.096) (0.094) (0.080) (0.086) (0.081) (0.092) (0.045) (0.084) (0.103)
Adjusted R2
0.026 -0.011 0.022 0.084 0.006 0.022 0.010 0.043 0.109 0.036
Observations 611 602 609 608 610 611 609 610 610 610
Descriptive statistics
Panel C: Young cohort (18-29)
Not resettled municipalities 0.579 0.661 0.441 0.073 0.253 0.659 0.511 0.822 0.657 0.394
(0.020) (0.019) (0.020) (0.011) (0.018) (0.019) (0.020) (0.016) (0.019) (0.020)
Resettled municipalities 0.591 0.750 0.443 0.057 0.250 0.727 0.534 0.747 0.602 0.398
(0.053) (0.046) (0.053) (0.025) (0.046) (0.048) (0.054) (0.047) (0.053) (0.053)
Panel D: Old cohort (60-74)
Not resettled municipalities 0.680 0.304 0.564 0.256 0.218 0.568 0.543 0.930 0.762 0.270
(0.019) (0.019) (0.020) (0.018) (0.017) (0.020) (0.020) (0.010) (0.017) (0.018)
Resettled municipalities 0.674 0.287 0.506 0.227 0.295 0.607 0.568 0.932 0.670 0.386
(0.050) (0.050) (0.053) (0.045) (0.049) (0.052) (0.053) (0.027) (0.050) (0.052)
Table reports estimated coefficients on an indicator variable for resettled municipalities after controlling for municipality (log of altitude, terrain roughness, log of distance
to the country border, region fixed effect, and population fixed effect) and personal (age group, education, labor market status, marital status, household size, and for
being born in the municipality of residence) characteristics. Standard errors clustered by municipality are reported in parentheses with: ' = p < 0.1, **=/?< 0.05,
*** = p < 0.01, *** = p < 0.001. Descriptive statistics are the mean and the standard deviations (in parentheses).
6.2 Social Capital
With respect to social capital we follow the literature (Knack 1992; Hotchkiss and
Rupasingha 2018) and use voter turnout in municipal elections at the municipality level as
one dependent variable3 0
In addition, from the first wave of the survey "Český venkov 2003
- situace před vstupem do EU." we use the responses of mayors to a question on the number
of events organized by local clubs and green activities (i.e., collective activities to improve
the environment and living conditions in the municipality) in their municipality. These
data are explicitly related to a specific municipality and are therefore measures of local
social capital. From the second wave of the survey we consider the responses to questions
on whether the interviewed individuals are a member of a local club in their municipality
and whether they participated in an event, have made a donation, or have committed to
conducting voluntary work that are not necessarily located in the respondent's municipality.
Of these variables only the first refers to a specific municipality and therefore measures
local social capital. For the other three the questions posed in the questionnaire (explicitly
or implicitly) also include contacts outside the specific municipality and therefore refer to
forms of social capital not directly related to a specific locality.
Among these variables the voter turnout allows for using RD-design as in Section 5, on
account of applying to all Czech municipalities. According to the results (Table 7),3 1
voter
turnout in municipal elections was statistically significantly lower in the resettled than the
not resettled municipalities with the point estimates of this impact ranging from 2.5 to 2.7
percentage points on average. In addition, the results of year by year RD-design estimates
for voter turnout (Figure 10) suggest that these differences between the resettled and not
resettled municipalities have reduced continuously since the times of the fall of the Iron
Curtain and attained a similar value in both types of municipalities in the 2010 and 2014
elections. Thus, resettlement also led to a reduced voter turnout in municipal elections,
that was, however, not as long-lived as the impact on the population churn.
For the survey based variables, by contrast, an RD-design analysis is not feasible, on
account of the surveys being based on only a sample of municipalities (see Figure B.3
in the Appendix), as for the values data, we therefore, only regress the responses at the
municipality level to these questions on a dummy variable indicating that a particular
municipality was a resettled municipality, and the same municipality-level controls also
used in the remainder of this analysis. For the individual-level data, we additionally control
for gender, age, home-ownership status, education, marital status, household size, and labor
market status.
30. In municipal elections citizens are not allowed to vote outside their constituency. This rules out potential
bias caused by, e.g., correspondence voting.
31. For descriptive statistic on voter turnout see Table A.7 in the Appendix.
33
Table 7: Impact of the resettlement on migration rates on voter turnout in municipal
elections
RD-polynomial
1s t
order
(1)
2 n d
order
(2)
Resettled municipality (=1) -2.538*** -2.718**
(0.653) (1.006)
Adjusted R 2
0.573 0.574
Observations 14423 14423
Note: Estimates from the specification equivalent to Equation (1) with voter turnout in municipal elections
held between 1990 and 2014 being the dependent variable. Only municipalities within a 15 km band around
the ethnic border are considered. Results control for year fixed effects, an RD-polynomial, region fixed
effects, population fixed effect, log of altitude, terrain roughness, log of the distance to the country border,
and 2n d
order polynomial of longitude and latitude. Symbols represent statistical significance of differences
in means between resettled and not resettled municipalities with ' = p < 0.1, * = p < 0.05, ** = p < 0.01,
*** = p < 0.001. Values in brackets are standard errors clustered by municipality.
Figure 10: Impact of the resettlement on voter turnout in municipal elections by year
§ g - 2
,8 3
Note: Figure reports estimates of the impact of the resettlement on the voter turnout in municipal elections
from the specification equivalent to Equation (1). Municipalities within a 15 km band around the ethnic
border are considered. Results control for the interaction of the year fixed effects with the dummy variable
for resettled municipality, interaction of year fixed effect with a first-order polynomial forcing function, and
region fixed effects, year fixed effects, population fixed effect, log of altitude, terrain roughness, log of the
distance to the country border, and second order polynomials of longitude and latitude. Points represent the
point estimates and the associated bars the 95% confidence intervals of the estimates. Standard errors are
clustered by municipality.
34
The results in Table 8 show statistically significant differences between the resettled
and not resettled municipalities in variables associated with local social capital (i.e. the
participation in events organized by local clubs, green activities, and club membership
in the municipalities). These differences are also quantitatively important because they
indicate that there are on average 2.8 fewer events organized by clubs and 0.9 fewer green
events organized in the resettled compared with not resettled municipalities in 2003 and
that the probability to participate in a local club or civil society organization is 0.1 to 0.2
percentage points lower in the resettled compared with the not resettled municipalities. The
same does, however, not apply to indicators associated with non-localized forms of social
capital (the frequency with which people attend events, provide donations, and participate
in voluntary work). Consistent with our hypothesis this implies lower local social capital
in resettled municipalities, and thus suggests that a persistently lower level of local social
capital in the resettled municipalities may have contributed to the lower attachment of the
population to their region.
Table 8: Differences in events organization, attendance, club membership and charitable
activities between resettled and not resettled municipalities
Panel A: Wave 1 Panel B: Wave 2
Events
organized by
local clubs (n)
Green
activities
(n)
Club
membership
(=1)
Events
attendance
(=1)
Donations
(=1)
Voluntary
work
(=1)
(1) (2) (3) (4) (5) (6)
Regression results
Resettled municipality (=1) -2.909* -0.882"* -0.1501,
0.003 -0.032 -0.030
(1.297) (0.310) (0.082) (0.062) (0.083) (0.066)
Adjusted R2
0.086 0.032 0.055 0.115 0.094 0.069
Observations 1,181 1,181 1,525 1,525 1,523 1,523
Descriptive statistics
Not resettled municipalities 10.598 2.233 0.428 0.596 0.586 0.248
(0.462) (0.093) (0.013) (0.013) (0.013) (0.012)
Resettled municipalities 8.535 1.208 0.314 0.599 0.465 0.256
(0.802) (0.177) (0.035) (0.037) (0.038) (0.033)
Regression results report estimated coefficients on an indicator variable for resettled municipalities.
Regressions in panels A and B control for the log of altitude, terrain roughness, log of distance to
the country border, population fixed effects, and region fixed effects. Municipality-level regressions in
the Panel A also control for age structure (shares of 20—29, 30—39, 40—49, 50—59, and 60—69 in
total population) and the shares of the low and high educated in population 15+ in 2001. Individual-level
regressions in Panel B control for gender, age group, home-ownership status, education, marital status,
household size, and labor market status. Values in parentheses are robust standard errors in Panel A, and
standard errors are clustered by municipality in Panel B: ' = p < 0.1, ** = p < 0.05, *** = p < 0.01,
*** = p < 0.001. Descriptive statistics are the mean and the standard deviations (in parentheses) for resettled
and not resettled municipalities.
35
7 Conclusions
In this paper, we exploit the resettlement of the Sudetenland after WWII as a natural
experiment to analyze the long-term impact of the destruction of the social structures
of municipalities affected by this resettlement on residential migration. This large
and unexpected resettlement involved the relocation of 2 million Czech inhabitants to
municipalities previously populated by ethnic Germans. We find that resettlement led to a
very long-term increase of the population churn from and to the affected municipalities.
Even at the end of our observation period (i.e. 70 years after the resettlement) annual
immigration and emigration rates were still by 0.4 to 0.5 percentage points higher (relative
to a baseline of around 2.5%) in the affected resettled municipalities than in the unaffected
not resettled ones without there being any indication of a decrease in these differences. In
addition, in the period from 1970 to about 1989 next to a more pronounced increase in
emigration and immigration also net immigration rates to the resettled municipalities was
lower by 0.5 to 0.7 percentage points. This effects, however, levels off by the 1990s and
may due to the migratory behaviour of the original settlers.
The longer-lived impact of resettlement on the population churn implies that even
in 2015 the population in resettled municipalities was still less attached to their place
of residence. Furthermore, the higher population churn survived important institutional
changes and major macroeconomic crises that occurred in the Czech Republic in that time.
These include the fall of the communist regime and the Iron Curtain, transition from a
planned to a market economy, separation from the Slovak Republic, and accession to the
European Union. Our evidence therefore suggests a persistent and long-term causal impact
of resettlement on the attachment of the population to the regions affected.
Exploring the plausibility of different potential causal mechanisms for this long-run
and stable impact of resettlement on the attachment of residents, we provide descriptive
evidence that inhabitants of resettled municipalities share the same values as residents of
municipalities that were not resettled and also do not differ in terms of their behaviour
with respect to proxy variables for social capital not bound to a specific locality (such as
their inclination to give donations to national organisations, as well as their willingness to
attend social events and to participate in voluntary work potentially outside their region of
residence). We, however, find evidence that resettled municipalities to this day are lower
values in measures of local social capital (such as the number of local social activities
organized and membership in local clubs). In addition, up until the early 2000s, resettled
municipalities had lower voter turnout in municipal elections than municipalities not
resettled. This finding leads us to conclude that lower local social capital was a likely
contributor to the lower attachment of the population to the resettled municipalities.
36
These findings thus corroborate previous empirical evidence on the long-term impact
that the destruction of the social structures of a region has on its economic, social, and
political development. In particular, the finding of higher population churn and lower
civic participation in clubs and similar communal activities in the resettled municipalities
is consistent with a number of recent theoretical and empirical contributions on the role
of local social capital in shaping migration decisions. These show that if local social
capital yields utility and is endogenously invested in by the resident population two stable
equilibria arise: one with low local social capital and high migration and one with high
local social capital and low migration (Brauninger and Tolciu 2011 or David et al. 2010).
Interpreting these results from this perspective, the destruction of local social capital in
resettled municipalities may have moved the resettled municipalities to a low local social
capital and a high migration equilibrium while not resettled municipalities continued to be
in high local social capital and low migration equilibrium. From a wider perspective this
evidence points to the highly persistent effects that destruction of the local social structure
of a region (as was the case in Sudetenland) can have on the mobility of residents from and
to that region in the long-run.
37
References
Acemoglu, Daron, Tarek A Hassan, and James A Robinson. 2011. "Social structure
and development: A legacy of the Holocaust in Russia." The Quarterly Journal of
Economics 126 (2): 895-946.
Akbulut-Yuksel, Mevlude, and Mutlu Yuksel. 2015. "The long-term direct and external
effects of Jewish Expulsions in Nazi Germany." American Economic Journal:
Economic Policy 7 (3): 58-85.
Alesina, Alberto, and Paola Giuliano. 2015. "Culture and institutions." Journal of Economic
Literature 53 (4): 898-944.
Alexander, Manfred. 2008. Kleine Geschichte der böhmischen Länder. Reclam.
Becker, Sascha O, Katrin Boeckh, Christa Hainz, and Ludger Woessmann. 2015. "The
empire is dead, long live the empire! Long-run persistence of trust and corruption in
the bureaucracy." The Economic Journal 126 (590): 40-74.
Becker, Sascha O, and Andreas Ferrara. 2019. "Consequences of forced migration: A
survey of recent findings." Labour Economics 59:1-16.
Becker, Sascha O, Irena Grosfeld, Pauline Grosjean, Nico Voigtlander, and Ekaterina
Zhuravskaya. 2020. "Forced migration and human capital: Evidence from post-WWII
population transfers." American Economic Review 110 (5): 1430-63.
Belot, Michele, and John Ermisch. 2009. "Friendship ties and geographical mobility:
evidence from Great Britain." Journal of the Royal Statistical Society: Series A
(Statistics in Society) 172 (2): 427^142.
Bharadwaj, Prashant, Asim Khwaja, and Atif Mian. 2008. "The big march: migratory flows
after the partition of India." Economic and Political Weekly, 39-49.
Brakman, Steven, Harry Garretsen, and Marc Schramm. 2004. "The strategic bombing
of German cities during World War II and its impact on city growth." Journal of
Economic Geography 4 (2): 201-218.
Bräuninger, Michael, and Andreia Tolciu. 2011. "Should I stay or should I go? Regional
mobility and social capital." Journal of Institutional and Theoretical Economics JITE
167 (3): 434-444.
38
Büchel, Konstantin, Maximilian V Ehrlich, Diego Puga, and Elisabet Viladecans-Marsal.
2020. "Calling from the outside: The role of networks in residential mobility." Journal
of urban economics 119:103277.
Čapka, František, Lubomír Slezák, and Jaroslav Vaculík. 2005. Nové osídlení pohraničí
českých zemí po druhé světové válce. Akademické nakladatelství C E R M .
Charnysh, Volha. 2019. "Diversity, institutions, and economic outcomes: Post-WWII
displacement in Poland." American Political Science Review 113 (2): 423-441.
Costa, Dora L, Matthew E Kahn, Christopher Roudiez, and Sven Wilson. 2018. "Persistent
social networks: Civil war veterans who fought together co-locate in later life."
Regional science and urban economics 70:289-299.
Cunningham, Scott. 2021. Causal inference: The mixtape. Yale University Press.
CZSO. 2014. Národní struktura obyvatel. Technical report, https://www.czso.cz/documents/
10180/20551765/170223-14.pdf. Prague: Czech Statistical Office.
Daněk, Petr. 1995. "Towards cultural regionalisation of the Czech Lands: Sudeten half a
century after the transfer." Scripta Fac. Brun 25:41-60.
David, Quentin, Alexandre Janiak, and Etienne Wasmer. 2010. "Local social capital and
geographical mobility." Journal of Urban Economics 68 (2): 191-204.
Dell, Melissa. 2010. "The persistent effects of Peru's mining mita." Econometrica 78 (6):
1863-1903.
Dvořák, Tomáš. 2013. Vnitřní odsun 1947-1953: Závěrečná fáze očisty pohraničí v
politických a společenských souvislostech poválečného Československa. Brno: Matice
Moravská.
Egger, Peter H , and Andrea Lassmann. 2015. "The causal impact of common native
language on international trade: Evidence from a spatial regression discontinuity
design." The Economic Journal 125 (584): 699-745.
Gerlach, David. 2010. "Beyond Expulsion: The Emergence of "Unwanted Elements" in the
Postwar Czech Borderlands, 1945-1950." East European Politics and Societies 24
(2): 269-293.
. 2017. The Economy of Ethnic Cleansing: The Transformation of the German-Czech
Borderlands after World War II. Cambridge University Press.
39
Glaeser, Edward L , David Laibson, and Bruce Sacerdote. 2002. " A n economic approach to
social capital." The economic journal 112 (483): F437-F458.
Glassheim, Eagle. 2000. "National mythologies and ethnic cleansing: The expulsion of
Czechoslovak Germans in 1945." Central European History 33 (4): 463-486.
. 2006. "Ethnic cleansing, communism, and environmental devastation in Czechoslovakia's
Borderlands, 1945-1989." The Journal of Modern History 78 (1): 65-92.
Greenwood, Michael J., Gary L . Hunt, Dan S. Rickman, and George I. Treyz. 1991.
"Migration, Regional Equilibrium, and the Estimation of Compensating Differentials."
The American Economic Review 81 (5): 1382-1390. ISSN: 00028282. http://www.
jstor. org/stable/2006927.
Grosfeld, Irena, Alexander Rodnyansky, and Ekaterina Zhuravskaya. 2013. "Persistent
antimarket culture: a legacy of the pale of settlement after the holocaust." American
Economic Journal: Economic Policy 5 (3): 189-226.
Helliwell, John F, and Robert D Putnam. 1999. Education and social capital. Technical
report. National Bureau of Economic Research.
Hotchkiss, Julie L , and Anil Rupasingha. 2018. "Individual Social Captial and Migration."
Kan, Kamhon. 2007. "Residential mobility and social capital." Journal of Urban Economics
61 (3): 436-457.
Kelly, Morgan. 2019. "The standard errors of persistence."
. 2020. "Understanding persistence."
Knack, Stephen. 1992. "Civic norms, social sanctions, and voter turnout." Rationality and
Society 4 (2): 133-156.
Majerová, Věra, et al. 2003. Český venkov 2003, Situace před vstupem do EU. Prague: PEF
C Z U Praha.
. 2004. Český venkov 2004, Život mladých a starých lidí. Prague: PEF C Z U Praha.
Matějka, Ondřej. 2008. Social Capital in Contemporary Sudetenland, edited by Joseph D
Lewandowski and Milan Znoj, 277-293. Newcastle, U K : Cambridge Scholars.
Meixner, Rudolf. 1988. Geschichte der Sudetendeutschen. Nürnberg: Helmut Preussler
Verlag.
40
Oto-Peralías, Daniel, and Diego Romero-Ávila. 2017. "Historical frontiers and the rise of
inequality: the case of the frontier of Granada." Journal of the European Economic
Association 15 (1): 54-98.
Pascali, Luigi. 2016. "Banks and development: Jewish communities in the Italian Renaissance
and current economic performance." Review of Economics and Statistics 98 (1):
140-158.
Putnam, Robert D. 1995. "Tuning in, tuning out: The strange disappearance of social capital
in America." PS: Political science & politics 28 (4): 664-684.
Radvanovský, Zdeněk. 2001. "Integrationsprobleme bei der Wiederbesiedlung der deutschen
Siedlungsgebiete in den böhmischen Ländern nach 1945." In Vorträge der Tagungen
des Collegium Carolinum in Bad Wiessee vom 20. bis 22. November 1992 und 19. bis
21. November 1993, 143-161. München, November.
Schumann, Abel. 2014. "Persistence of population shocks: Evidence from the occupation of
West Germany after World War II." American Economic Journal: Applied Economics
6(3): 189-205.
Školí, Jaroslav. 1983. Nové osídlení okresu Břeclav po roce 1945. Mikulov: Okresní archív
Břeclav.
Spurný, Matěj. 2011. Nejsou jako my: Česká společnost a menšiny v pohraničí 1945-1960.
Praha: Antikomplex.
Staněk, Tomáš. 1991. Odsun Němců z Československa, 1945-1947. Academia.
Testa, Patrick A . 2020. "The Economic Legacy of Expulsion: Lessons from Postwar
Czechoslovakia." Ueaal32, The Economic Journal (December), ISSN: 0013-0133.
https://doi.org/10.1093/ej/ueaal32. eprint: https://academic.oup.com/ej/advancearticle-pdf/doi/10.1093/ej/ueaal32/34702029/ueaal32.pdf.
https://doi.org/10.1093/
ej/ueaal32.
Urbánní a regionální laboratoř (URRlab) U K . 2015. Historická data v GIS. http://www.
historickygis.cz/.
Vaněk, Miroslav. 1996. Nedalo se tady dýchat: Ekologie v českých zemích v letech 1968 až
1989. Praha: Ústav pro soudobé dějiny A V ČR.
Waldinger, Fabian. 2010. "Quality matters: The expulsion of professors and the consequences
for PhD student outcomes in Nazi Germany." Journal of Political Economy 118 (4):
787-831.
41
Waldinger, Fabian. 2011. "Peer effects in science: Evidence from the dismissal of scientists
in Nazi Germany." The Review of Economic Studies 79 (2): 838-861.
. 2016. "Bombs, brains, and science: The role of human and physical capital for
the creation of scientific knowledge." Review of Economics and Statistics 98 (5):
811-831.
Wiedemann, Andreas. 2016. "Pojďs námi budovat pohraničí!": osídlování a proměna
obyvatelstva bývalých Sudet 1945-1952. Prague: Prostor.
Wilson, Margaret FJ, Brian O'Connell, Colin Brown, Janine C Guinan, and Anthony J
Grehan. 2007. "Multiscale terrain analysis of multibeam bathymetry data for habitat
mapping on the continental slope." Marine Geodesy 30 (1-2): 3-35.
42
The Long-term impact of the resettlement of the
Sudetenland on residential migration
Appendix
43
A Appendix: Additional Figures and Tables
44
Table A . l : Share of employees on working age population (%) by economic sector in 1930
Judicial districts by share
of ethnic Germans (%) F-test Difference
(90,100] (10,90] [0,10] (F-statistic) (3)-(l)
(1) (2) (3) (4) (5)
Auxiliary retail services 0.837 0.560 0.482 1.908 -0.355Ť
(0.167) (0.124) (0.099)
Construction 8.311 10.033 9.612 1.306 1.301
(0.501) (0.786) (0.557)
Crop and animal production 19.335 29.878 40.839 8,494*** 21.504***
(4.636) (4.791) (2.758)
Education 1.452 2.082 1.577 13.654"* 0.126
(0.076) (0.137) (0.049)
Financial and insurance activities 0.639 0.532 0.732 0.275 0.093
(0.084) (0.088) (0.185)
Forestry, hunting and fishing 1.444 1.608 1.284 0.212 -0.160
(0.367) (0.393) (0.290)
Gastronomy 2.348 1.591 1.485 8.366*** -0.863***
(0.152) (0.126) (0.128)
Health services 0.719 0.672 0.809 0.410 0.090
(0.042) (0.084) (0.106)
Manufacture of clothes 3.531 3.708 4.240 1.227 0.709
(0.287) (0.331) (0.308)
Manufacture of food products 4.388 4.614 6.076 2.166 1.688
(0.456) (0.623) (0.591)
Manufacture of footwear 1.819 1.591 2.153 0.496 0.334
(0.179) (0.283) (0.392)
Manufacture of glass 3.931 2.916 0.427 2.352 -3.504Ť
(2.915) (1.462) (0.163)
Manufacture of chemicals 0.571 0.688 0.411 0.246 -0.160
(0.294) (0.512) (0.182)
Manufacture of leather 0.564 0.622 0.672 0.097 0.108
(0.090) (0.169) (0.157)
Manufacture of machinery 3.080 2.343 3.355 0.569 0.275
(0.699) (0.374) (0.585)
Manufacture of metals 0.046 0.128 0.394 0.361 0.348
(0.036) (0.111) (0.314)
Manufacture of other products 0.233 0.166 0.070 9.355*** -0.163***
(0.055) (0.039) (0.011)
Number of districts 71 85 174
Note: Columns (1) to (3) contain means and standard errors (in parentheses). Column (4) contains F-test
for systematic differences across judicial districts groups and Column (5) contains differences in means
between districts with less than 10% and more than 90% of ethnic Germans: ' = p < 0.1, * = p < 0.05,
** =p< 0.01, ***=/?< 0.001.
45
Table A.2: Share of employees on working age population (%) by economic sector in 1930
Judicial districts by share
of ethnic Germans (%) F-test Difference
(90,100] (10,90] [0,10] (F-statistic) (3)-(l)
(1) (2) (3) (4) (5)
Manufacture of paper 1.076 0.321 0.287 4.262* -0.789*
(0.453) (0.130) (0.078)
Manufacture of textiles 35.176 12.910 5.037 10.811*** -30.139***
(8.931) (5.523) (2.193)
Metals processing 6.720 4.794 5.687 0.607 -1.033
(1.227) (0.721) (0.856)
Military 0.620 4.313 4.332 1.048 3.712
(0.303) (2.067) (1.670)
Mining 2.186 5.147 0.297 2.799f
-1.889t
(1.741) (3.397) (0.128)
Other occupations 0.853 0.681 0.618 0.403 -0.235
(0.120) (0.137) (0.168)
Other public services 0.430 0.480 0.408 0.312 -0.022
(0.097) (0.082) (0.042)
Other transport services 2.303 1.396 1.391 4.580* -0.913*
(0.454) (0.147) (0.124)
Postal services 0.771 0.741 0.821 0.190 0.050
(0.092) (0.099) (0.082)
Printing and artistic creation 0.852 0.432 0.458 1.923 -0.394t
(0.169) (0.141) (0.123)
Production of electricity, gas, water, etc. 0.454 0.248 0.292 1.268 -0.162
(0.092) (0.060) (0.070)
Public administration 1.372 1.574 1.549 0.170 0.177
(0.127) (0.175) (0.215)
Quarrying 2.628 3.069 3.226 0.215 0.597
(0.545) (0.674) (0.542)
Railway transport 2.897 3.191 2.804 0.119 -0.093
(0.759) (0.746) (0.379)
Timber processing 4.900 4.140 5.018 0.224 0.118
(0.419) (0.396) (0.891)
Wholesale and retail trade 5.724 5.418 5.462 0.066 -0.261
(0.463) (0.615) (0.465)
Number of districts 71 85 174
Note: Columns (1) to (3) contain means and standard errors (in parentheses). Column (4) contains F-test
for systematic differences across judicial districts groups and Column (5) contains differences in means
between districts with less than 10% and more than 90% of ethnic Germans: ' = p < 0.1, * = p < 0.05,
**=/?< 0.01, ***=/?< 0.001.
46
Figure A . l : Share of ethnic Germans injudicial districts in 1930
Source: CZSO, Urbánní a regionální laboratoř (URRlab) UK (2015).
Note: The ethnic German population is defined according to the primarily spoken language.
Table A.3: Resettlement process in Břeclav area: Number and share of settlers by the
distance of origin
Municipality Settlers from Czech Republic by the distance of origin Total Total
< 5km < 10km < 30km < 100km Settlers Population
(1) (2) (3) (4) (5) (6)
Břeclav 195 (4.4%) 875 (19.5%) 1,719 (38.4%) 2,930 (65.5%) 4,476 11,010
Valtice 112(5.8%) 290(15.1%) 712 (37.2%) 956 (49.9%) 1,915 3,000
Hustopeče 70 (5.0%) 349 (25.0%) 756 (54.2%) 922(66.1%) 1,395 2,652
Dolní Dunajovice 0 (0.0%) 0 (0.0%) 302 (25.4%) 654 (55.0%) 1,190 1,689
Lednice 126 (10.7%) 299 (25.4%) 584 (49.5%) 792 (67.2%) 1,179 1,740
Pohořelice 98 (10.0%) 326 (33.4%) 728 (74.6%) 924 (94.7%) 976 2,961
Drnholec 0 (0.0%) 0 (0.0%) 27 (2.9%) 528 (56.4%) 937 1,484
Mikulov 0 (0.0%) 0 (0.0%) 8 (0.9%) 444 (47.4%) 937 5,337
Novosedly 0 (0.0%) 0 (0.0%) 25 (2.7%) 387 (41.3%) 936 1,062
Zaječí 213 (23.3%) 389 (42.6%) 660 (72.3%) 724 (79.3%) 913 1,214
Popice 35 (3.9%) 146 (16.3%) 640 (71.3%) 674(75.1%) 897 979
Vlasatice 13 (1.6%) 54 (6.5%) 107 (12.8%) 457 (54.6%) 837 988
Pouzdřany 138 (18.7%) 177 (24.0%) 276 (37.4%) 505 (68.4%) 738 1,067
Sedlec 0 (0.0%) 34 (5.2%) 164 (25.3%) 405 (62.5%) 648 803
Strachotín 65 (10.0%) 65 (10.0%) 402 (62.1%) 451 (69.7%) 647 789
Jevišovka 0 (0.0%) 0 (0.0%) 23 (3.7%) 72(11.5%) 627 808
Perná 0 (0.0%) 0 (0.0%) 34 (5.6%) 111 (18.3%) 608 776
Starovice 0 (0.0%) 327 (55.0%) 453 (76.1%) 468 (78.7%) 595 742
Přítluky 329 (56.9%) 419 (72.5%) 477 (82.5%) 502 (86.9%) 578 745
Bulhary 6 (1.1%) 63 (11.2%) 204 (36.1%) 437 (77.3%) 565 708
Horní Věstonice 0 (0.0%) 0 (0.0%) 161 (31.1%) 348 (67.3%) 517 549
Mušov 27 (5.4%) 33 (6.5%) 129 (25.6%) 356 (70.6%) 504 559
Březí 0 (0.0%) 0 (0.0%) 54(11.4%) 272 (57.4%) 474 1,208
Pavlov 0 (0.0%) 0 (0.0%) 69(15.2%) 245 (53.8%) 455 658
Brod nad Dyjí 0 (0.0%) 0 (0.0%) 103 (23.4%) 275 (62.5%) 440 587
Poštorná 84 (19.2%) 122 (27.9%) 159 (36.3%) 243 (55.5%) 438 3,073
Nový Přerov 5 (1.2%) 5 (1.2%) 18 (4.3%) 23 (5.4%) 423 510
Cvrčovice 42 (11.1%) 42 (11.1%) 97 (25.7%) 138 (36.5%) 378 591
Úvaly 0 (0.0%) 0 (0.0%) 89 (25.6%) 207 (59.7%) 347 n.a.
Milovice 16 (4.7%) 21 (6.2%) 114 (33.7%) 237 (70.1%) 338 425
Klentnice 0 (0.0%) 0 (0.0%) 54(16.4%) 245 (74.5%) 329 373
Bavory 0 (0.0%) 0 (0.0%) 31 (10.7%) 145 (49.8%) 291 342
Dolní Věstonice 0 (0.0%) 0 (0.0%) 99 (40.4%) 149 (60.8%) 245 412
Dobré Pole 0 (0.0%) 0 (0.0%) 36(15.7%) 73 (31.9%) 229 460
Charvatská Nová Ves 76 (34.2%) 132 (59.5%) 144 (64.9%) 153 (68.9%) 222 1,725
Nová Ves 120 (55.6%) 153 (70.8%) 216(100.0%) 279 (129.2%) 216 269
Smolín 5 (2.4%) 19 (9.3%) 31 (15.1%) 102(49.8%) 205 304
Nej dek 8 (4.5%) 37 (20.7%) 121 (67.6%) 136 (76.0%) 179 212
Pasohlávky 0 (0.0%) 0 (0.0%) 31 (21.8%) 54 (38.0%) 142 282
Nové Mlýny 4 (4.3%) 8 (8.6%) 57 (61.3%) 57 (61.3%) 93 143
Hlohovec 3 (5.6%) 19 (35.2%) 21 (38.9%) 25 (46.3%) 54 1,307
Kurdějov 15 (60.0%) 16 (64.0%) 21 (84.0%) 21 (84.0%) 25 524
Total 1,805 (6.4%) 4,420 (15.7%) 10,156 (36.1%) 17,126 (60.9%) 28,138 55,067
Source: Školí (1983), own elaboration.
48
Table A.4: Resettlement process in Břeclav area: Number and share of settlers from 4 most
important municipalities of origin within Czech Republic
Municipality Settlers from n most important origins Total Total
n = 1 n = 2 n = 3 n = 4 Settlers Population
( D (2) (3) (4) (5) (6)
Břeclav 416 (9.3%) 677 (15.1%) 935 (20.9%) 1,091 (24.4%) 4,476 11,010
Valtice 125 (6.5%) 224(11.7%) 301 (15.7%) 367 (19.2%) 1,915 3,000
Hustopeče 85 (6.1%) 159 (11.4%) 229 (16.4%) 294(21.1%) 1,395 2,652
Dolní Dunajovice 122 (10.3%) 203 (17.0%) 249 (20.9%) 291 (24.4%) 1,190 1,689
Lednice 70 (5.9%) 120 (10.2%) 170 (14.4%) 218(18.5%) 1,179 1,740
Pohořelice 131 (13.4%) 262 (26.8%) 296 (30.3%) 330 (33.8%) 976 2,961
Drnholec 99(10.6%) 195 (20.8%) 237 (25.3%) 272 (29.0%) 937 1,484
Mikulov 54 (5.7%) 92 (9.8%) 130 (13.9%) 161 (17.2%) 937 5,337
Novosedly 88 (9.4%) 137 (14.6%) 177 (18.9%) 216(23.1%) 936 1,062
Zaječí 161 (17.6%) 250 (27.4%) 302 (33.1%) 350 (38.3%) 913 1,214
Popice 123 (13.7%) 184 (20.5%) 239 (26.6%) 293 (32.7%) 897 979
Vlasatice 48 (5.7%) 87(10.4%) 119 (14.2%) 150(17.9%) 837 988
Pouzdřany 95(12.9%) 128 (17.3%) 161 (21.8%) 191 (25.9%) 738 1,067
Sedlec 51 (7.9%) 91 (14.0%) 120(18.5%) 144 (22.2%) 648 803
Strachotín 100 (15.5%) 177 (27.4%) 242 (37.4%) 277 (42.8%) 647 789
Jevišovka 11 (1.8%) 21 (3.3%) 29 (4.6%) 37 (5.9%) 627 808
Perná 176 (28.9%) 210 (34.6%) 233 (38.4%) 256 (42.2%) 608 776
Starovice 74(12.4%) 146 (24.5%) 208 (35.0%) 269 (45.2%) 595 742
Přítluky 329 (56.9%) 370 (64.0%) 393 (68.0%) 413 (71.5%) 578 745
Bulhary 41 (7.3%) 82(14.5%) 117 (20.7%) 152 (26.9%) 565 708
Horní Věstonice 107 (20.7%) 207 (40.0%) 241 (46.6%) 268 (51.8%) 517 549
Mušov 38 (7.5%) 69(13.7%) 92 (18.3%) 114 (22.6%) 504 559
Březí 34 (7.3%) 69(14.5%) 99 (21.0%) 130 (27.4%) 474 1,208
Pavlov 57(12.6%) 99 (21.9%) 119(26.1%) 134 (29.4%) 455 658
Brod nad Dyjí 46 (10.4%) 88 (20.0%) 130 (29.6%) 157 (35.7%) 440 587
Poštorná 51 (11.6%) 84(19.2%) 109 (24.9%) 128 (29.2%) 438 3,073
Nový Přerov 21 (5.0%) 26 (6.1%) 31 (7.3%) 36 (8.5%) 423 510
Cvrčovice 49(13.0%) 72(19.0%) 86 (22.8%) 95 (25.1%) 378 591
Úvaly 29 (8.4%) 54(15.6%) 78 (22.5%) 98 (28.2%) 347 n.a.
Milovice 33 (9.8%) 55 (16.3%) 75 (22.2%) 93 (27.5%) 338 425
Klentnice 69 (20.9%) 103 (31.4%) 134 (40.7%) 165 (50.0%) 329 373
Bavory 73 (25.0%) 107 (36.8%) 134 (46.0%) 153 (52.6%) 291 342
Dolní Věstonice 42(17.2%) 77 (31.2%) 92 (37.5%) 103 (42.2%) 245 412
Dobré Pole 40(17.5%) 65 (28.4%) 77 (33.6%) 88 (38.4%) 229 460
Charvatská Nová Ves 55 (24.8%) 82 (36.9%) 103 (46.4%) 122 (55.0%) 222 1,725
Nová Ves 17 (7.9%) 34(15.7%) 51 (23.6%) 67 (31.0%) 216 269
Smolín 28(13.7%) 40(19.5%) 50 (24.4%) 57 (27.8%) 205 304
Nejdek 33(18.4%) 53 (29.6%) 68 (38.0%) 82 (45.8%) 179 212
Pasohlávky 19(13.5%) 34 (24.3%) 46 (32.3%) 54 (37.7%) 142 282
Nové Mlýny 29 (31.2%) 37 (39.8%) 44 (47.3%) 49 (52.7%) 93 143
Hlohovec 10(18.5%) 14 (25.9%) 17 (31.5%) 20 (37.0%) 54 1,307
Kurdějov 8 (32.0%) 16 (62.6%) 20 (78.6%) 24 (94.6%) 25 524
Total 3,288(11.7%) 5,300 (18.8%) 6,783 (24.1%) 8,007 (28.5%) 28,138 55,067
Source: Školí (1983), own elaboration.
49
Table A.5: Resettlement process in Břeclav area: Number and share of settlers by country
of origin
Municipality Czech
Republic
Bulgaria Slovakia Soviet
Union
Yugoslavia Austria Other Total
Settlers
Total
Population
a) (2) (3) (4) (5) (6) (7) (8) (9)
Břeclav 4,181 (93.4%) 3 (0.1%) 0 (0.0%) 17 (0.4%) 7 (0.2%) 62(1.4%) 27 (0.6%) 4,476 11,010
Valtice 1,486 (77.6%) 7 (0.4%) 88 (4.6%) 3 (0.2%) 17 (0.9%) 55 (2.9%) 39 (2.0%) 1,915 3,000
Hustopeče 1,323 (94.8%) 1 (0.1%) 4 (0.3%) 10 (0.7%) 6 (0.4%) 15 (1.1%) 8 (0.6%) 1,395 2,652
Dolní Dunajovice 1,102 (92.6%) 0 (0.0%) 8 (0.7%) 13 (1.1%) 5 (0.4%) 2 (0.2%) 1 (0.1%) 1,190 1,689
Lednice 992(84.1%) 7 (0.6%) 27 (2.3%) 0 (0.0%) 50 (4.2%) 25 (2.1%) 9 (0.8%) 1,179 1,740
Pohořelice 809 (82.9%) 4 (0.4%) 48 (4.9%) 61 (6.2%) 3 (0.3%) 12(1.2%) 11 (1.1%) 976 2,961
Drnholec 815 (87.0%) 11 (1.2%) 77 (8.2%) 0 (0.0%) 0 (0.0%) 31 (3.3%) 4 (0.4%) 937 1,484
Mikulov 796 (85.0%) 2 (0.2%) 19 (2.0%) 2 (0.2%) 20(2.1%) 1 (0.1%) 7 (0.7%) 937 5,337
Novosedly 785 (83.9%) 81 (8.7%) 18(1.9%) 0 (0.0%) 33 (3.5%) 3 (0.3%) 4 (0.4%) 936 1,062
Zaječí 860 (94.2%) 0 (0.0%) 15 (1.6%) 4 (0.4%) 0 (0.0%) 9(1.0%) 14 (1.5%) 913 1,214
Popice 873 (97.3%) 0 (0.0%) 0 (0.0%) 19 (2.1%) 0 (0.0%) 2 (0.2%) 2 (0.2%) 897 979
Vlasatice 676 (80.8%) 22 (2.6%) 59 (7.0%) 32 (3.8%) 8(1.0%) 2 (0.2%) 3 (0.4%) 837 988
Pouzdřany 707 (95.8%) 0 (0.0%) 4 (0.5%) 11 (1.5%) 0 (0.0%) 0 (0.0%) 3 (0.4%) 738 1,067
Sedlec 564 (87.0%) 0 (0.0%) 17 (2.6%) 0 (0.0%) 46(7.1%) 2 (0.3%) 6 (0.9%) 648 803
Strachotín 588 (90.9%) 0 (0.0%) 12(1.9%) 47 (7.3%) 0 (0.0%) 1 (0.2%) 0 (0.0%) 647 789
Jevišovka 158 (25.2%) 445 (71.0%) 5 (0.8%) 0 (0.0%) 13 (2.1%) 0 (0.0%) 2 (0.3%) 627 808
Perná 513 (84.4%) 0 (0.0%) 4 (0.7%) 1 (0.2%) 2 (0.3%) 1 (0.2%) 1 (0.2%) 608 776
Starovice 589 (99.0%) 0 (0.0%) 0 (0.0%) 4 (0.7%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 595 742
Přítluky 561 (97.1%) 0 (0.0%) 6(1.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 2 (0.3%) 578 745
Bulhary 544 (96.3%) 0 (0.0%) 5 (0.9%) 1 (0.2%) 0 (0.0%) 3 (0.5%) 3 (0.5%) 565 708
Horní Věstonice 497(96.1%) 0 (0.0%) 8(1.5%) 0 (0.0%) 0 (0.0%) 4 (0.8%) 4 (0.8%) 517 549
Mušov 470 (93.3%) 10 (2.0%) 8(1.6%) 0 (0.0%) 2 (0.4%) 2 (0.4%) 0 (0.0%) 504 559
Březí 398 (84.0%) 0 (0.0%) 4 (0.8%) 1 (0.2%) 18 (3.8%) 0 (0.0%) 0 (0.0%) 474 1,208
Pavlov 398 (87.5%) 0 (0.0%) 15 (3.3%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 4 (0.9%) 455 658
Brod nad Dyjí 421 (95.7%) 4 (0.9%) 15 (3.4%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 440 587
Poštorná 400(91.3%) 0 (0.0%) 6(1.4%) 4 (0.9%) 0 (0.0%) 6(1.4%) 1 (0.2%) 438 3,073
Nový Přerov 91 (21.5%) 327 (77.3%) 1 (0.2%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 1 (0.2%) 423 510
Cvrčovice 248 (65.6%) 0 (0.0%) 23 (6.1%) 83 (22.0%) 11 (2.9%) 1 (0.3%) 2 (0.5%) 378 591
Úvaly 333 (96.0%) 0 (0.0%) 14 (4.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 347 n.a.
Milovice 335 (99.1%) 0 (0.0%) 3 (0.9%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 338 425
Klentnice 321 (97.6%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 2 (0.6%) 0 (0.0%) 0 (0.0%) 329 373
Bavory 272 (93.5%) 0 (0.0%) 0 (0.0%) 11 (3.8%) 8 (2.7%) 0 (0.0%) 0 (0.0%) 291 342
Dolní Věstonice 272(111.0%) 0 (0.0%) 4(1.6%) 0 (0.0%) 4(1.6%) 0 (0.0%) 4(1.6%) 245 412
Dobré Pole 176 (76.9%) 0 (0.0%) 1 (0.4%) 0 (0.0%) 20 (8.7%) 14(6.1%) 0 (0.0%) 229 460
Charvatská Nová Ves 215 (96.8%) 0 (0.0%) 4(1.8%) 0 (0.0%) 0 (0.0%) 3 (1.4%) 0 (0.0%) 222 1,725
Nová Ves 158 (73.1%) 6 (2.8%) 17 (7.9%) 7 (3.2%) 0 (0.0%) 0 (0.0%) 20 (9.3%) 216 269
Smolín 158 (77.1%) 0 (0.0%) 0 (0.0%) 47 (22.9%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 205 304
Nejdek 161 (89.9%) 0 (0.0%) 7 (3.9%) 4 (2.2%) 4 (2.2%) 0 (0.0%) 3 (1.7%) 179 212
Pasohlávky 103 (72.5%) 0 (0.0%) 27 (19.0%) 0 (0.0%) 0 (0.0%) 11 (7.7%) 0 (0.0%) 142 282
Nové Mlýny 85 (91.4%) 0 (0.0%) 5 (5.4%) 0 (0.0%) 0 (0.0%) 1 (1.1%) 2 (2.2%) 93 143
Hlohovec 43 (79.6%) 0 (0.0%) 1 (1.9%) 1 (1.9%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 54 1,307
Kurdějov 117(468.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 8 (32.0%) 0 (0.0%) 25 524
Total 24,594 (90.3%) 930 (3.4%) 579 (2.1%) 383 (1.4%) 279(1.0%) 276(1.0%) 187 (0.7%) 28,138 55,067
Source: Školí (1983), own elaboration. Note: Category "other" covers settlers from Poland, Germany,
Romania, France, Hungary, Argentina, United Kingdom, Italy, Norway, Canada, Israel, Ecuador and United
States.
50
Figure A.2: Origins of settlers in Břeclav area
Source: Školí (1983), own elaboration.
Figure A. 3: The ethnic border and district borders
^=t'lfi]iiLb
u n k r
52
10.0-
7.5-
5.0-
2.5-
0.0-
12.5g
10.0-
u
S 7.5-
e
•2 5.0-
ea
00
0.0-
10-
5-
0-
5 -
1 0 Figure
A.4: Migration rates by population
Population below median Population above median
EP"
Not resettled
municipalities
Resettled
municipalities
Not resettled
municipalities
Resettled
municipalities
h H Not resettled municipalities l - H Resettled municipalities
53
Table A.6: Impact of the resettlement on migration rates (3 order RD-polynomial)
Panel estimates Cross-sectional estimates using data
aggregated by period
1971-2015 1971-1989 1990-2015
(1) (2) (3) (4)
Dependent variable: Emigration rate (%)
Resettled municipality (= 1) 0.570*** 0.551*** 0.990*** 0.290*
(0.133) (0.143) (0.223) (0.133)
Adjusted R 2
0.169 0.443 0.454 0.359
Observations 82423 2104 1882 2104
Dependent variable: Immigration rate (%)
Resettled municipality (==1) 0.405** 0.464** 0.778* 0.271
(0.148) (0.167) (0.311) (0.176)
Adjusted R 2
0.091 0.319 0.268 0.314
Observations 82423 2104 1882 2104
Dependent variable: Net immigration rate (%)
Resettled municipality (==1) -0.166 -0.087 -0.212 -0.019
(0.131) (0.165) (0.286) (0.176)
Adjusted R 2
0.096 0.101 0.236 0.124
Observations 82423 2104 1882 2104
Note: Estimates from Equation (1) in the main text with 3 order RD-polynomial. Only municipalities
within a 15 km band around the ethnic border are considered. Results control for year fixed effects, an
RD-polynomial, region fixed effects, population fixed effect, log of altitude, terrain roughness, log of the
distance to the country border, longitude, latitude, and squared longitude and latitude squared. Symbols
represent statistical significance of differences in means between resettled and not resettled municipalities
with ' = p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.001. Values in brackets are standard errors
clustered by municipality.
54
Table A.7: Descriptive statistics for turnout in municipal elections
Elections Municipalities Difference
(2)-(3)A l l Resettled Not resettled
Difference
(2)-(3)
(1) (2) (3) (4)
1990 85.72 78.22 87.59 9 37***
(0.14) (0.36) (0.15)
1994 78.18 70.05 80.36 -10.31***
(0.13) (0.40) (0.13)
1998 67.37 61.13 69.36 8 23***
(0.17) (0.50) (0.18)
2002 66.17 62.44 67.59
(0.16) (0.46) (0.17)
2006 63.27 59.93 64.53 -4.59***
(0.16) (0.48) (0.17)
2010 63.41 61.34 64.42 -3.08***
(0.15) (0.46) (0.17)
2014 60.62 58.37 61.65 3 2Y***
(0.16) (0.47) (0.17)
Note: Columns (1) to (3) contain means and standard errors (in parentheses). Columns (4) contains difference
in means with ' = p < 0.1, ** = p < 0.05, * * * = p < 0.01, * * * = p < 0.001.
55
References
Školí, Jaroslav. 1983. Nové osídlení okresu Břeclav po roce 1945. Mikulov: Okresní archív
Břeclav.
Urbánní a regionální laboratoř (URRlab) U K . 2015. Historická data v GIS. http://www.
historickygis.cz/.
56
B Appendix: Description of data set and merging definitions
of communities
B.l Migration data
Data on permanent residence changes within a year was obtained from the website of the
Czech Statistical Office (CZSO, https://www.czso.cz/csu/czso/databaze-demografickychudaju-za-obce-cr,
last accessed April 4t h
2019). These data are compiled from administrative
historical records on annual residence changes from 1971 to 2015 by the C Z S O . Issues
related to changing definitions of borders and territories of municipalities are not applicable
as CZSO matches them to a consistent data set throughout. The data report the number of
people who move in and out of a municipality per year and provide information on the
population of a municipality on January 1s t
of a year. The latter are imputed by C Z S O
using the last census and annual migration, mortality and natality statistics. The emigration
and immigration rates as well as the net immigration rates are used as dependent variables
and are defined as the number of residential moves across municipality borders over a year
in percent of the population on January 1s t
of the year.
The data are not a balanced panel because some records were lost prior to their
digitization. The number of missing observations is highest in the late 1980s. In this period
observations on only 66% of the municipalities are available. Data availability swiftly
improves after the fall of the communist regime in 1989. As of the year 1992 96.7% of the
data are observed and this ratio has grown ever since. Throughout the main part of the
paper we use the maximum number of observations possible and thus focus on unbalanced
panel data. We address issues relating to missing observations in additional robustness
checks, where we present results based on a balanced panel (see Appendix C).
Furthermore, in processing the data we remove outliers by removing the top 1% of
observations with respect to emigration and immigration rates that may be due to large
region-specific shocks (such as e.g. the destruction of a number of municipalities due to
open pit lignite mining in the Northwest of the Czech Republic - see Vaněk 1996). In
our robustness checks in Appendix C we show that this has only minor impacts on the
estimated causal effects and if anything increases them. This implies that by omitting
outliers, estimates are more conservative.
B.2 Census Data
The key source of historical demographic data of municipalities is the publicly available
database "Historický lexikon obcí" (available at ast accessed April 4 t h
2019). The database
57
is compiled and maintained by the CZSO. It contains municipality level data from general
censuses on the total population going back to 1869 using the definition of municipalities
in 2011. This data is our source of information on the population and the number of houses
in a municipality in 1930.
Unfortunately, data on the ethnic Germans in a municipality, which was contained in the
1930 census and is based on a question on the language spoken in the household was, not
digitized. We therefore inputted this data from the printed statistical overviews of the 1930
census published in the 1930s for the 11,797 municipalities reported in the 1930 census.
These municipalities do not accord with the definition of the 6,258 municipalities in 2011,
that are the observational units for all other data. To derive consistent data we therefore
applied the same aggregation rules as used by the CZSO in the construction of time series for
the "Historický lexikon obcí" and aggregated pre-WWII data to the 2011 municipalities.3 2
Using these aggregation rules we succeeded in matching 6168 municipalities (98.6%)
with their 1930 counterparts. 4,210 of these municipalities were matched 1 to 1, and
865 municipalities were matched 2 to 1. Only 307 current municipalities consist of more
than five 1930 municipalities with the maximum number of municipalities being found in
today's Prague, the country capital, which consisted of 60 historical municipalities in 1930.
In our robustness checks we also use data from the decennial population censuses held
since 1980. These are the only source on the age and highest completed education of the
population of the municipalities. The post-WWII general censuses were held in 1950,1961,
1970, 1980, 1991, 2001, and 2011. Data on the age and education structure at municipality
level are available only for 1980 onward because non-digitized records from the previous
censuses were lost to a flood in 2002. We therefore obtained the 2011 census data at the
CZSO website (https://www.czso.cz/csu/czso/otevrena_data_pro_vysledky_scitani_lidu_
domu_a_bytu_201 l_sldb_2011, last accessed April 4 t h
2019), while data from the 1980,
1991 and 2001 censuses aggregated on the level of 2011 municipalities were provided to us
by the C Z S O upon request.
B.3 Data on election turnout
Data on election turnout for all municipal elections from 2006 to 2014, was downloaded
from the official website for election results run by CZSO (https://www.volby.cz/opendata/
opendata.htm), last accessed April 4 t h
2019). For earlier elections the data were provided
by C Z S O upon request. These data contain the number of eligible voters as well as the
32. The necessary aggregation rules were kindly provided by the CZSO. They are based on tracking
individual neighborhoods. For example, if a municipality was divided and merged with multiple municipalities
the aggregation rules associate the historical records of the municipality with a municipality which absorbed
the largest neighborhood. By applying these rules the CZSO can construct long-term time-series on population
for current municipalities.
58
number of ballots cast in all the free municipal elections with voluntary voting held in
today's Czech Republic since 1990 for each municipality. They can therefore be used to
calculate turnout rates. These are defined as the share of votes cast in percent of the eligible
voters for each election. This measure is highly reliable, as voting outside the community
in which voters are registered as residents is not possible in municipal elections in the
Czech Republic.
B.4 Data on the resettlement of Břeclav area
There is no comprehensive data set available on the resettlement process due to its often
chaotic nature. To our knowledge the study by Školí (1983)) on the resettlement of the
Břeclav area, South Moravia, is the only data source covering a compact region. This study
presents data on the resettlement collected from two principal sources: individual 1950
census questionnaires, and records on the land redistribution. Školí's study contains basic
information on origins of settlers. For the settlers from today's Czech Republic he lists
individual municipalities of the origin while for settlers from Slovakia and foreign countries
the study contains only districts or country of origin. The presented data sometimes report
only information on the number of families that moved and sometimes only the number of
persons, while in some instances (12% of all cases) it contains both families and persons.
We use these overlaps to impute the number of persons moving based on the average family
size (3.8 individuals) to estimate the number of moving persons for all records where this
information is missing in the original data.
B.5 Survey data on social capital
The paper also uses two surveys on the social values and social capital in the Czech
Republic. While several surveys in the Czech Republic include questions on local social
capital and various values, to the best of our knowledge, only these two provide geolocated
respondents at a municipality level and thus allow to determine whether respondents lived
in resettled or not resettled municipalities.
B.5.1 Survey "Český venkov 2004: život mladých a starých lidí"
Data on social values is taken from the survey "Český venkov 2004: život mladých a
starých lidí" (Majerová et al. 2004), which was conducted among young (aged 18-29) and
old (aged 60-74) respondents in small municipalities (with less than 2000 inhabitants) in
2004 using a standardized questionnaire. Data are available from the Czech Social Science
Data Archive (http://archiv.soc.cas.cz/en, last accessed April 4t h
2019). They contain
observations on 751 young respondents and on 767 old respondents.
59
We exclude respondents living in municipalities with a population in excess of 2,000
inhabitants according to the 2001 census (12 young respondents and 7 old respondents)
and with a self-reported age outside the range declared by the survey definition (4 young
respondents older than 29 and 17 old respondents younger than 60) from the sample. This
leaves us with 735 young and 738 old respondents. Young respondents are drawn from 275
and old respondents from 278 different municipalities. The geographical distribution of
the municipalities and municipality types covered are reported in Figure B . l .
The questions we use asked respondents on their on their values. These were elicited
through the battery of questions listed in Table B . l . As respondents evaluated all listed
values independently the answers tend to be strongly skewed towards positive evaluations
(i.e. "Very important" and "Rather important") with "Very important" being chosen
in 51% cases by young and in 52% by the old group (see Figure B.2). The importance
of spiritual values is the only significant exception from the pattern as only 7% of young
respondents consider them being "Very important" (see Table 6). Therefore, in the main
part of the paper these variables are encoded as an indicator variable that takes on a
value of 1 if a respondent answered a question with "Very important" and zero else. We,
however, conducted robustness checks with linear regressions using the full set of values for
responses. These robustness checks suggest that our choice of coding has few consequences
for the qualitative results (see Appendix C below).
60
Table B.l: Questions used from survey "Český venkov 2004: život mladých a starých lidí"
Wording of the question
English Czech (original)
In the following questions, we willfocus on your
personal values and goals.
People in their lives value some things more and
some things less. How important arefollowing
things in your life? Please, assign to each
of listed values an importance for your life
nowadays. Please, try to use the whole scale
from 1 to 4from the CARD while evaluating.
CARD:
1. Very important
2. Rather important
3. Rather unimportant
4. Not important at all
V následujících otázkách se zaměříme na Vaše
Životní hodnoty a cíle.
Lidé si ve svém životě některých věcí cení více,
některých méně. Jak důležité jsou následující
věci pro Váš osobní život? Přiřaďte prosím
každé z vyjmenovaných hodnot důležitost, kterou
jim ve svém současném životě přikládáte. Při
hodnocení podle KARTY se, prosím, snažte
využívat celou stupnici odjednička do čtyřky.
KARTA:
1. Velmi důležité
2. Spíše důležité
3. Spíše nedůležité
4. Zcela nedůležité
VALUES:
1. Nature, environment
2. Job, occupation
3. Relationships
4. Faith, spiritual values
5. Interests, hobbies
6. Housing
7. Friendship
8. Health status
9. Family life, children
10. Material conditions
HODNOTY:
1. Příroda, životní prostředí
2. Práce, zaměstnání
3. Mezilidské vztahy
4. Víra, duchovní hodnoty
5. Zájmy, koníčky
6. Bydlení
7. Přátelství, parta
8. Zdravotní stav
9. Rodinný život, děti
10. Materiální podmínky
Source: "Český venkov 2004: život mladých a starých lidí" (Majerová et al. 2004)
61
Figure B . l : Location of respondents by municipality type in "Český venkov 2004: život
mladých a starých lidí"
(a) Young respondents
(b) Old respondents
Source: "Český venkov 2004: život mladých a starých lidí" (Majerová et al. 2004), own calculations
62
B.5.2 Survey "Český venkov 2003: situace před vstupem do EU"
The second survey used is the survey "Český venkov 2003: situace před vstupem do
EU" (Majerová et al. 2003). It too was conducted in municipalities with less than 2,000
inhabitants in 2003. In its first wave this survey asked mayors of municipalities on the
number of events organized by local clubs and green activities in their municipality, while in
the second wave individuals were asked on whether they are member of a club, participated
in an event, have made donation or have committed to conduct voluntary work. We use
this survey to evaluate the stock of local social capital in the municipalities of the Czech
Republic. Data from both waves of the survey can be downloaded from the Czech Social
Science Data Archive (http://archiv.soc.cas.cz/en, last accessed April 4t h
2019).
The first wave was a correspondence expert opinion survey. In this 2,000 mayors of small
municipalities (with up to 2000 inhabitants) were asked to assess various characteristics
of their municipality. 1,324 (66%) of the mayors participated. From these we exclude
37 municipalities with a population above the declared threshold of 2,000 inhabitants
according to the 2001 population census. This leaves us with 1,287 usable observations
(see top panel of Figure B.3 for their spatial distribution and municipality types).
We focus on two questions asked in this first wave. The first is the reported number of
events organized by local clubs; the second is the number of "green activities". For the
number of events, mayors were asked to report the number of events organized by rather
broad categories of clubs: Voluntary firefighters, sport clubs, hunters, breeders, Red cross,
women union, and others. The survey used the following question:
"Označte spolky a zájmová sdružení, které aktivně působí ve Vaší obci a
napište počet akcí, které daný spolek pořádal pro veřejnost za poslední rok."
"Choose clubs and voluntary organizations, which are active in your village and
fill in the number of public events the club organized in the last year."
We use the reported number of events as a dependent variable.
For "green activities" — i.e. collective activities aimed to improve the environment and
living conditions in the municipality (e.g. "spring cleaning", football field reconstruction,
etc.) with the participation of locals — mayors were asked:
"Pořádají se u Vás akce pro zvelebování obce, kterých se účastní obyvatelé
obce?"
"Are there green activities in your village, which locals participate in?"
If respondents answered this question affirmatively they were also required to list the
number of events held. We used this number as the dependent variable in the estimates.
63
Figure B.2: Distribution of the responses to value questions in "Český venkov 2004: život
mladých a starých lidí"
(a) Young respondents
Nature, environment
200-
200-
200-
s
I
200-
Relationships
Friendship
Family life and children
400Job,
occupation
Faith, spiritual values
•I. I.
Material conditions
l._ IIVery
important
Not Very
important important
at all
Not
important
at all
(b) Old respondents
Nature, environment
I.Relationships
Job, occupation
Faith, spiritual values
I. IlFriendship
Family life and children Material conditions
400
I IVery
important
Not Very
important important
at all
Not
important
at all
Source: "Český venkov 2004: život mladých a starých lidí" (Majerová et al. 2004), own calculations
64
Figure B.3: Location of respondents by municipality type in "Český venkov 2003: situace
před vstupem do EU"
(a) First Wave (mayors)
lities
(b) Second Wave (residents)
lities
Source: "Český venkov 2003: situace před vstupem do EU" (Majerová et al. 2003), own calculations
65
The second wave of the survey focused on individuals living in a selection of small
municipalities in the Czech Republic. Interviews were conducted among 1,634 respondents.
We exclude 27 respondents from municipalities with a population above the declared
threshold (of 2,000) in the 2001 census. This leaves us with 1,617 observations living in
223 different municipalities (see bottom panel of Figure B.3 for their spatial distribution
and municipalities).
The first question we use, addresses the "use" of voluntary organizations and explicitly
focuses on the municipality of residence by the respondent and members of his/her
household:
"Využíváte vy nebo jiní členové domácnosti - spolky, zájmová sdružení v
obci?"
"Do you, or any member of your household, use clubs or voluntary organizations
in your village?"
The response to this question could be "Yes" or "No". The verb "využívat" (to use)
in the original Czech wording of the question is not commonly used in this context.
We interpret it as a broader term covering not only a membership but also not binding
interactions - such as playing football on a club field without being an organized club
member. It, therefore, addresses the nature of social capital as it captures interactions with
fellow villagers. We encoded positive responses with a value of 1 and negative ones with a
value of zero.
The other questions used, by contrast, focus on general charitable and social activities
not necessarily related to the municipality of residence. They aim at eliciting information
on general activities such as social events attendance, donation on charitable activities
and volunteering. In addition, the question on volunteering explicitly asks respondents to
take into account activities outside the municipality of residence. The wording of these
questions is the following:
"Využíváte vy nebo jiní členové domácnosti - společenské akce (disco, plesy,
zábavy)?"
"Do you, or any member of your household, attend social events (festivals, balls,
etc.)?"
"Zapojil(a) jste se v roce 2002 jako dobrovolník do nějaké akce (nejen v obci)?"
"Did you volunteer for an event in 2002 (not only in your village) ? "
"Poskytl(a) jste v roce 2002 dar na dobročinné účely?"
"Did you make a donation on charitable activities in 2002?"
66
The first two questions could be answered by "Yes" or "No" only and were thus encoded
by an indicator variable that takes on the value of 1 for positive and zero for negative
responses. The last question could be answered by "Yes, in kind and money", "Yes, in
kind", "Yes in money" and "No". Here we encoded all positive responses (irrespective of
whether they were in kind, in money or both) with one and a negative response with zero.
B.5.3 Representativity of Questionnaires
These questionnaires focus on a rather specific sample of municipalities (municipalities with
2,000 or less inhabitants). This raises obvious concerns about the representativity of this
questionnaire. While we have no way to deal with this issue, we checked on the distribution
of municipality size in 2001 among our matched municipalities, the municipalitiesin
the proximity of the ethnic border, and our overall sample. This check suggests that
municipalities with 2,000 inhabitants or less account for 90% of the municipalities in the
Czech Republic and host 26% of the population. Similarly, such municipalities account for
98% of the matched municipalities in the Version 3 of the matching estimates and 90% of
their population. Furthermore small municipalities account for 92% of the resettled and
not resettled municipalities within 15 kilometers band around the ethnic border used in
RD-estimates and 45% of the population of these municipalities. Thus, while clearly not
representative of the Czech Republic, the Sudetenland or the matched population, the data
is at least representative of a large share of the affected area.
67
References
Majerová, Věra, et al. 2003. Český venkov 2003, Situace před vstupem do EU. Prague: PEF
C Z U Praha.
. 2004. Český venkov 2004, Život mladých a starých lidí. Prague: PEF C Z U Praha.
Školí, Jaroslav. 1983. Nové osídlení okresu Břeclav po roce 1945. Mikulov: Okresní archív
Břeclav.
Vaněk, Miroslav. 1996. Nedalo se tady dýchat: Ekologie v českých zemích v letech 1968 až
1989. Praha: Ústav pro soudobé dějiny A V ČR.
68
C Appendix: Robustness tests
This appendix presents the results of robustness tests presented in Table 5 as well as the
additional robustness tests mentioned in the main part of the paper and an additional
robustness test for regressions in Table 6. In these robustness tests we
1. Follow a recent suggestion by Kelly (2020) to calculate standard errors corrected for
spatial auto-correlation and (as an alternative approach)
2. base inference on permutation tests rather than on clustered standard errors used in
the baseline specification.
3. Consider specifications in which different functional forms are allowed for on different
sides of the border.
4. Provide estimation results using local linear regression based on several optimal
bandwidth selection methods.
5. Use an identification strategy based on spatial matching, that avoids having to define
a border between resettled and not resettled municipalities and thus does not require
defining this border based on the 50% population share.
6. We conduct the RD-analysis and define resettled municipalities as municipalities with
pre-WWII a share of ethnic Germans of 50% or more and all others as municipalities
that were not subjected to resettlement.
7. We exclude municipalities close to Bavaria (West Germany) from the estimation
sample.
8. We control for potentially divergent regional trends by inclusion of linear trends for
each region and period (before/after 1990) into the baseline specification.
9. We exclude the municipalities located in the northeast of the Czech Republic
characterized by a large number of ethnic border irregularities.
10. We use a balanced panel dataset.
11. We include outliers omitted in the previous analysis.
12. We use the 1950 population as a control variable.
13. We estimate the baseline specification separately for three parts of the country.
14. We Estimate linear regressions of the social value data with this encoded on a scale
from 1 to 4 to ensure that our change of coding has no effects on results
69
C. 1 Robustness test: Standard errors adjusted for spatial auto-correlation
In a recent paper Kelly (2019) provides evidence that RD-design identification estimates
based on geographic data may be severely biased in the presence of Spatial auto-correlation.
He shows that standard errors could be severely affected by the presence of spatial
autocorrelation in cases where the the Moran's Z scores are significant (i.e. exceed 2). In
our application (as shown in Table C . l ) this is the case for emigration and immigration
rates.3 3
Kelly (2020) proposes this concern by explicitly considering spatial auto-correlation
in the estimation of such RD-models. We therefore estimate standard errors using his
suggested method using his R-code.3 4
In doing so, we closely follow the suggestions made by Kelly (2020): We iterate the
procedure for ranges between 10 and 260 kilometers (10, 60, 110, 160, 210, 260) and
smoothness parameters between 0.1 and 1 (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1).
To stay on the conservative side we report the largest standard errors estimated in this
parameter space.
Table C. 1 shows that resulting standard errors are virtually identical for emigration
rate and only marginally larger for immigration (by 0.006-0.013) and net immigration (by
0.006-0.016) rates.
C.2 Robustness test: Permutation test
Additionally, we test the significance of our point-estimates in a permutation test as
an alternative way to treat potential spatial auto-correlation. In this test we obtain an
empirical distribution of placebo estimates by artificially shifting ethnic border and use
this distribution to test the statistical significance of baseline point-estimates at the actual
border. Such a procedure should account for the spatial autocorrelation structure if this is
identical at ethnic border and the nearby regions used in the permutation test.
To obtain the placebo estimates we use a sample of not resettled municipalities located
from 5 to 50 kilometers from ethnic border toward inland regions. We then artificially shift
ethnic border by approx. 15 meters at a time between 20 and 35 km away from the actual
one. This ensures that the minimum band along the placebo border is 15 km - as it is in our
estimation sample. Resulting distribution of 1,000 placebo estimates is used for testing.
One limitation of this permutation test is that it is based only on placebo shifts away from
33. This table uses a Moran's I that is calculated by taking into account the five closest municipalities (as
equally weighted neighbors).
34. For the R code developed by Kelly (2020) see https://github.com/morganwkelly/persistence (last
accessed on March 15, 2021).
70
Table C . l : Standard errors adjusted for spatial autocorrelation
RD-polynomial
1s t
order 2 order 3 order
(1) (2) (3)
Dependent variable:
Emigration rate (%)
Resettled municipality (== D 0.735*** 0.690*** 0.570***
(0.062) (0.091) (0.131)
Moran's Z 3.203 3.238 3.249
Smoothness 0.200 0.200 0.200
Dependent variable:
Immigration rate (%)
Resettled municipality (== D 0.504*** 0.465*** 0.405*
(0.076) (0.112) (0.161)
Moran's Z 2.456 2.470 2.480
Smoothness 0.400 0.200 0.200
Dependent variable:
Net Immigration rate (%)
Resettled municipality (== D -0.231*** -0.225* -0.166
(0.068) (0.101) (0.147)
Moran's Z 0.888 0.889 0.882
Smoothness 0.200 0.200 0.200
Note: Panel estimates from Equation (1) in the main text. Only municipalities within a 15 km band around
the ethnic border are considered. Results control for year fixed effects, an RD-polynomial, region fixed
effects, population fixed effect, log of altitude, terrain roughness, log of the distance to the country border,
and 2n d
order polynomial of longitude and latitude. Symbols represent statistical significance of differences
in means between resettled and not resettled municipalities with ' = p < 0.1, * = p < 0.05, ** = p < 0.01,
*** = p < 0.001. Values in brackets are spatial autocorrelation-adjusted standard errors. Moran's Z scores
are calculated by taking into account the five closest municipalities (as equally weighted neighbors).
71
the border toward the interior of the Czech Republic as the distance between the ethnic and
country border is often too narrow to allow for shifting towards the country border.
Results reported in Table C.2 show that in the permutation test the effect on the
net immigration rate is significant at 0.1% level for the specification with 2 n d
order
RD-polynomial. Other results are identical to baseline estimates.
Table C.2: Permutation test
RD-polynomial
1s t
order 2 n d
order 3 r d
order
(1) (2) (3)
Dependent variable:
Emigration rate (%)
Resettled municipality (==D 0.735*** 0.690*** 0.570***
(0.021) (0.033) (0.055)
Dependent variable:
Immigration rate (%)
Resettled municipality (==D 0.504*** 0.465*** 0.405***
(0.069) (0.069) (0.085)
Dependent variable:
Net Immigration rate (%)
Resettled municipality (= -0.231*** -0.225*** -0.166
(0.068) (0.078) (0.089)
Note: Panel estimates from Equation (1) in the main text. Only municipalities within a 15 km band around
the ethnic border are considered. Results control for year fixed effects, an RD-polynomial, region fixed
effects, population fixed effect, log of altitude, terrain roughness, log of the distance to the country border, and
2n d
order polynomial of longitude and latitude. Values in brackets are standard deviations of the empirical
distributions obtained in permutation tests. Symbols represent statistical significance of differences in
means between resettled and not resettled municipalities derived in a permutation test with t
= p < 0.1,
* = p < 0.05, **=/?< 0.01, ***=/?< 0.001.
C.3 Robustness test: RD-polynomials
The baseline specification (1) uses an RD-polynomial of the same degree on both sides of
ethnic border. A s a robustness check we relax this restriction and allow the order of an
RD-polynomial to differ on either side of the border as such a specifications allows for a
higher flexibility in modelling the running variable. Results reported in C.3 are in the line
with the baseline results. As in our baseline results the effect varies between 0.6 and 0.7
for the emigration rate, between 0.4 and 0.5 for immigration rate, and it is around 0.2 for
the net immigration rate.
72
Table C.3: RD-polynomials
RD-polynomial
(Not resettled municipalities)
RD-polynomial 1s t
order 2 n d
order 3 r d
order
(Resettled municipalities) (1) (2) (3)
Dependent variable:
Emigration rate (%)
1s t
order 0.731*** 0.720***
(0.068) (0.078)
2 n d
order 0.694*** 0.679***
(0.088) (0.099)
3r d
order 0.585*** 0.581***
(0.126) (0.129)
Dependent variable:
Immigration rate (%)
1s t
order 0.495*** 0.528***
(0.077) (0.091)
2 n d
order 0.473*** 0.498***
(0.098) (0.113)
3r d
order 0.380** 0.372**
(0.137) (0.140)
Dependent variable:
Net Immigration rate (%)
1s t
order -0.236*** -0.192*
(0.068) (0.080)
2 n d
order -0.221* -0.182f
(0.088) (0.101)
3r d
order -0.205f
-0.209f
(0.122) (0.124)
Note: Panel estimates from a specification equivalent to Equation (1) in the main text allowing for different
RD-polynomials on each side of the cutoff. Only municipalities within a 15 km band around the ethnic border
are considered. Results control for year fixed effects, an RD-polynomial, region fixed effects, population fixed
effect, log of altitude, terrain roughness, log of the distance to the country border, and 2n d
order polynomial
of longitude and latitude. Symbols represent statistical significance of differences in means between resettled
and not resettled municipalities with ' = p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.001. Values in
brackets are standard errors clustered by municipality.
73
C.4 Robustness test: Local linear regressions
Further to relax our assumptions on the functional form even more we estimate local
linear regressions with the optimum bandwidth selected by Mean Square Error (MSE) and
Coverage Error Rate (CER) selectors implemented by Calonico et al. (2020). Selectors are
used to choose one bandwidth for both sides of the border or for each side of the border
individually.
Estimates for the full sample of observations reported in columns (1) to (3) in Table C.4
diverge from our baseline estimates. However, these results are affected by the definition of
ethnic border which follows the administrative borders of municipalities. The municipality
reference points, that are used for the calculation of the distance to ethnic border, are only
rarely located at administrative borders (see an illustrative example in Figure C.l). As a
result there are very few observations in the close proximity of ethnic border (see Figure
6) which makes non-parametric estimates unstable and sensitive to outliers. When we
exclude 0.5 km band along the ethnic border, which contains 10 not resettled and 7 resettled
municipalities, and adjust distances accordingly, the non-parametric estimates in columns
(4) to (6) converge to parametric ones. Similarly doubling the bandwidth in columns (7) to
(9), that reduces sensitivity to sparse observations in the proximity of ethnic border, leads
to point estimates that are close to parametric ones.
74
Figure C . l : Calculation of distance to ethnic border (illustrative example)
75
Table C.4: Non-parametric (local linear) estimates
Full sample, optimum bandwidth Exclusion of 0.5 km band Double of optimum bandwidth
Dependent variable Dependent variable Dependent variable
Emigration Immigration Net Emigration Immigration Net Emigration Immigration Net
rate (%) rate (%) immigration rate (%) rate (%) immigration
rate (%) rate (%)
rate (%) rate (%) immigration
rate (%)
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Bandwidth selection: MSE-optimal bandwidth
Resettled municipality (=1) 0.209t
0.398*" -0.156f
0.778*** 0.519*** -0.285*** 0.568*** 0.464*** -0.182"
(0.107) (0.093) (0.091) (0.060) (0.064) (0.068) (0.061) (0.054) (0.056)
Bandiwdth (km) 1.843 2.765 3.624 2.511 2.855 3.650 3.687 5.529 7.249
Bandwidth selection: MSE-optimal bandwidths (below and above the cutoff)
Resettled municipality (=1) 0.400*" 0.387*** _0.144t
0.823*** 0.526*** -0.288*** 0.649*** 0.478*** -0.184***
(0.084) (0.089) (0.086) (0.058) (0.067) (0.065) (0.050) (0.053) (0.054)
Bandiwdth, Not resettled municipalities (km) 2.525 3.630 3.124 5.532 3.669 3.865 5.050 7.260 6.248
Bandiwdth, Resettled municipalities (km) 2.410 2.753 4.197 2.424 2.532 4.122
Bandwidth selection: CER-optimal bandwidth
4.820 5.506 8.394
Resettled municipality (=1) -0.557" 0.471** 0.136 0.789*** 0.627*** -0.218* 0.309** 0.389*** -0.194*
(0.202) (0.161) (0.149) (0.081) (0.088) (0.094) (0.097) (0.084) (0.082)
Bandiwdth (km) 1.047 1.570 2.058 1.426 1.622 2.073 2.093 3.139 4.116
Bandwidth selection: CER-optimal bandwidths (below and above the cutoff)
Resettled municipality (=1) -0.088 0.427** 0.152 0.795*** 0.584*** -0.228* 0.457*** 0.411*** -0.185*
(0.145) (0.151) (0.144) (0.077) (0.088) (0.089) (0.076) (0.080) (0.078)
Bandiwdth, Not resettled municipalities (km) 1.434 2.061 1.774 3.142 2.084 2.195 2.867 4.122 3.548
Bandiwdth, Resettled municipalities (km) 1.368 1.563 2.383 1.377 1.438 2.341 2.737 3.127 4.766
Note: Panel estimates from local linear regressions. Only municipalities within a 15 km band around the ethnic border are considered. Results control for year fixed effects,
an RD-polynomial, region fixed effects, population fixed effect, log of altitude, terrain roughness, and log of the distance to the country border. Symbols represent statistical
significance of differences in means between resettled and not resettled municipalities with ' = p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.001. Values in brackets
are robust standard errors.
C.5 Robustness test: Spatial matching strategy
The RD-strategy requires us to define the ethnic border, that never existed as a border of,
for instance, an administrative unit. Every possible definition of ethnic border is, therefore,
to some extent arbitrary and could potentially drive our results. Therefore, we also present
estimates based on spatial matching strategy, which does not require any definition of
ethnic border.
The matching estimates apply on nearest neighbor matching (with replacement) based
on a Euclidean distance measure3 5
using population in 1930 and geographic distance as
matching variables. Population in 1930 is included as a pretreatment variable to control for
important unobservable man-made amenities of agglomerations (e.g., provision of public
services, infrastructure, hospitals, cinemas, or sports stadiums and others) that may affect
migration. Distance, by contrast, is used to mitigate the effect of unobserved municipality
characteristics such as the distance to regional capital, economic conditions, natural and
man-made amenities, which also influence migration decisions.
We conduct three versions of the matching procedure. These differ in the maximum
distance and population differences allowed for. In a first Version these maxima are 5
kilometers and 500 inhabitants. In a second Version they are 10 kilometers and 250
inhabitants. In the third and least restrictive Version they are 10 kilometers and 500
inhabitants. Depending on these restrictions the matching procedure provides between 46
and 179 matched pairs (Figure C.2).
For each matching Version we estimate the following specification by ordinary least
squares:
where yit is one of the three outcome variables (emigration, immigration, and net
immigration rates) defined for municipality i in year t, <pp is a match-specific intercept, R M ,
is an indicator variable for a resettled municipality, and Z,- includes geographical variables
(log of altitude, the terrain roughness, log of shortest distance to the Czech border, and
2 n d
order polynomial of longitude and latitude). The year fixed effects <pt are included to
control for changes in institutions and economic developments affecting all municipalities
alike, and rjit is the error term.
35. The standardized Euclidean distance d between two municipalities is calculated as J a b = ^8AB+
P\B>
where g is the geographical distance between municipalities adjusted for the earth's curvature, and p is
the population difference. Both g and p are normalized variables with zero mean and unit variance across
all municipality pairs eligible for matching. A resettled municipality is matched with an not resettled
municipality that minimizes d.
yn = <t>p + y R M j + pZi + (f>t + 7]it (2)
77
Figure C.2: Matched pairs of municipalities by maximum permitted geographical distance
and population differences (First part)
78
Figure C.3: Matched pairs of municipalities by maximum permitted geographical distance
and population differences (Continued)
(c) Version 3
Source: CZSO, own calculations.
Note: In Version 1 municipalities are matched only if they are located less than 5 km from each other and
their population difference is less than 500 inhabitants. In Versions 2 and 3 these maxima are 10 km and 250
inhabitants, and 10 km and 500 inhabitants, respectively.
79
Table C.5: Balance tests for matched pairs
Matching version
Version 1 Version 2 Version 3
Not resettled Resettled Not resettled Resettled Not resettled Resettled
municipalities municipalities municipalities municipalities municipalities municipalities
(1) (2) (3) (4) (5) (6)
Altitude (meters) 348.304 353.598 362.827 385.405* 376.866 395.824
(19.691) (18.840) (10.604) (10.904) (10.107) (10.159)
Terrain roughness 2.699 2.574 2.679 2.800 2.798 2.910
(0.188) (0.203) (0.112) (0.112) (0.099) (0.102)
Distance to the country border (km) 27.280 25.901** 27.692 24.968*** 26.622 24.005***
(2.332) (2.256) (1.270) (1.202) (1.168) (1.097)
Population3
832.718 831.254 821.441 819.232 882.305 921.772
(9.914) (12.011) (7.527) (7.809) (7.746) (8.351)
Maximum geographical distance (meters) 5000 10000 10000
Maximum difference in population 500 250 500
Mean geographical distance (meters) 4135.2 8112.6 7909.4
Mean difference in population 206.4 118.6 214.6
Number of matched pairs 46 147 179
Note: In Version 1 municipalities are matched only if they are located less than 5 km from each other and their population difference is less than 500 inhabitants. In Versions
2 and 3 these maxima are 10 km and 250 inhabitants, and 10 km and 500 inhabitants, respectively. Table 2 reports means (and robust standard errors in parentheses) with
statistical significance of differences in means between resettled and not resettled municipalities with ' = p < 0.1,* = p < 0.05, ** = p < 0.01, *** = p < 0.001. For
geographical variables tests for differences in means control for matched pair fixed effects. Tests for difference in population additionally control for the log of altitude,
terrain roughness, and log of distance to the country border. P-values are calculated using robust standard errors.
Table C.6 presents the baseline estimates of the parameter of interest (7) of Equation (2)
from all three versions of matching using standard errors that are clustered by municipality.
Table C.6: Robustness test: Spatial matching strategy
Matching
1s t
version 2 n d
version 3 r d
versionrd
(1) (2) (3)
Resettled municipality (=1)
Observations
Adjusted R 2
Resettled municipality (=1)
Observations
Adjusted R 2
Resettled municipality (=1)
Observations
Adjusted R 2
Dependent variable:
Emigration rate (%)
0.620*** 0.782*** 0.816***
(0.088) (0.053) (0.040)
3342 10404 12788
0.178 0.219 0.219
Dependent variable:
Immigration rate (%)
0.516*** 0.607*** 0.546***
(0.085) (0.061) (0.058)
3342 10404 12788
0.119 0.126 0.136
Dependent variable:
Net immigration rate (%)
-0.104* -0.175** -0.271***
(0.051) (0.053) (0.055)
3342 10404 12788
0.119 0.136 0.140
Note: Estimates from Equation (2). In Version 1 settlements are matched only if they are located less than 5
km from each other and their population difference is less than 500 inhabitants. In Versions 2 and 3 these
maxima are 10 km and 250 inhabitants, and 10 km and 500 inhabitants, respectively. All estimates control
for pair fixed effects, year fixed effects, log of altitude, terrain roughness, and log of the distance to the
country border. Symbols represent statistical significance of differences in means between resettled and not
resettled municipalities with ' = p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.001. Values in brackets
are standard errors clustered by municipality.
These estimates suggest that resettlement increased emigration rates from the resettled
municipalities by 0.6 to 0.8 percentage points and immigration rates by approximately 0.5
to 0.6 percentage points on average in the years 1971 to 2015. In addition, according to the
results net immigration decreased by 0.1 to 0.3 percentage points per year on average in the
studied period. A l l estimates are significant at least at 5% level and are fully comparable
with baseline results obtained with RD-strategy. These results suggest that our baseline
results are not driven by a specific definition of ethnic border.
81
C.6 Robustness test: The 1950 population as a control variable
In the baseline specification we use 1930 population for population fixed effects. The 1930
population is clearly exogenous to the resettlement process, but it may be a bad proxy for
the development infrastructure and other man-made amenities as municipality sizes of
formerly German-dominated municipalities changed substantially between 1930 and the
post-resettlement period. To address this concern we re-estimate the baseline specification
(1) using 1950 population for population fixed effects.
In the estimates (Table C.7), coefficients for emigration and immigration as well as net
immigration rates change by less than 0.1 percentage points relative to the baseline results
reported in the main part of the paper and also maintain their significance levels.
C.7 Robustness test: Focusing on all municipalities
Another concern with the RD-strategy is that we focus only on municipalities which either
had a 90% German speaking majority in the German part of the country or a 90% Czech
speaking majority in the Czech speaking part of the country. This means we omit 466
(18%) municipalities from our data (see Figure 2 in the main part of the paper). This could
imply that estimates may be influenced by a few outliers located directly at the border.
To check on this issue, we added another robustness check where we define resettled
municipalities as municipalities with a majority (more than 50%) of ethnic Germans. The
rest of municipalities as not being resettled. The disadvantage of this approach is that
this may lead to a comparison of municipalities at the border, that were only marginally
differently affected by resettlement. The advantage is that now virtually all municipalities
in the 15 kilometer band around the ethnic border are included in the analysis.
The results of this robustness check (reported in Table C.8) once more indicate a high
consistency with the results reported in the main part of the paper. According to the results
the impact of resettlement on emigration and immigration rates as well as net immigration
rates changes by less than 0.2 throughout. The mild decline in the estimated effect is also
consistent with inclusion of less-treated municipalities into the estimation sample.
82
Table C.7: Robustness test: The 1950 population as a control variable
RD-polynomial
1s t
order 2 order 3 order
(1) (2) (3)
Dependent variable:
Emig;ration rate (%)
Resettled municipality (= 0.710*** 0.688*** 0.597***
(0.062) (0.089) (0.129)
Adjusted R 2
0.170 0.170 0.170
Observations 82388 82388 82388
Dependent variable:
Immi:gration rate (%]
Resettled municipality (==D 0.536*** 0.490*** 0.448**
(0.071) (0.103) (0.147)
Adjusted R 2
0.090 0.090 0.090
Observations 82388 82388 82388
Dependent variable:
Net Immigration rate (%)
Resettled municipality (= -0.173** -0.198* -0.150
(0.062) (0.089) (0.125)
Adjusted R 2
0.097 0.097 0.097
Observations 82388 82388 82388
Note: Panel estimates from Equation (1) in the main text. Only municipalities within a 15 km band around the
ethnic border are considered. Results control for year fixed effects, an RD-polynomial, region fixed effects,
population fixed effect based on 1950 population, log of altitude, terrain roughness, log of the distance to the
country border, and 2n d
order polynomial of longitude and latitude. Symbols represent statistical significance
of differences in means between resettled and not resettled municipalities with ' = p < 0.1,* = p < 0.05,
** = p < 0.01, *** = p < 0.001. Values in brackets are standard errors clustered by municipality.
83
Table C.8: Robustness test: Focusing on all municipalities
RD-polynomial
1s t
order 2 n d
order 3 r d
order
(1) (2) (3)
Dependent variable:
Emigration rate (%)
Resettled municipality (= 0.593*** 0.514*** 0.491***
(0.039) (0.057) (0.081)
Adjusted R 2
0.188 0.188 0.189
Observations 99109 99109 99109
Dependent variable:
Immigration rate (%)
Resettled municipality (==D 0.350*** 0.340*** 0.353***
(0.043) (0.062) (0.089)
Adjusted R 2
0.103 0.103 0.103
Observations 99109 99109 99109
Dependent variable:
Net Immigration rate (%)
Resettled municipality (= -0.242*** -0.175** -0.138f
(0.039) (0.057) (0.081)
Adjusted R 2
0.100 0.100 0.100
Observations 99109 99109 99109
Note: Panel estimates from Equation (1) in the main text with resettled municipalities being defined as
municipalities with German-speaking majority in 1930. All other municipalities are considered as not
being resettled. Only municipalities within a 15 km band around the ethnic border are considered. Results
control for year fixed effects, an RD-polynomial, region fixed effects, population fixed effect, log of altitude,
terrain roughness, log of the distance to the country border, and 2n d
order polynomial of longitude and
latitude. Symbols represent statistical significance of differences in means between resettled and not resettled
municipalities with ' = p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.001. Values in brackets are
standard errors clustered by municipality.
84
C.8 Robustness test: Exclusion of a part of ethnic border close to Bavaria
(West Germany)
The major institutional shift related to fall of the Iron curtain and the communist regime
in 1989 may have an asymmetric impact on municipalities in the estimation sample as it
may asymmetrically increase the attractiveness of the Czech-German border region as a
place of residence. In this robustness check we re-estimate regression (1) excluding the
municipalities located close to Bavaria (West Germany) - see Figure C.4. The results
reported in table C9 are once more highly consistent with those reported in the main text.
Figure C.4: Municipalities within 15 km to the ethnic border after omitting part of ethnic
border closest to Bavaria (West Germany)
Results reported in Table C.9 are practically identical to the baseline estimates confirming
high robustness of our results.
C.9 Robustness test: Exclusion of a part of ethnic border characterized
by a large number of irregularities
One of the issues with using the RD-strategy is that throughout history (with the exception
of the time from 1938 to 1945) a well-defined border between the Czech and German parts
of today's Czech Republic did not exist and that the ethnic border between the two parts of
the country is marked by a number of irregularities such as ethnic enclaves. To provide
some evidence of the potential impact of these irregularities, we omitted municipalities
85
Table C.9: Robustness test: Exclusion of the part of the ethnic border closest to Bavaria
(West Germany)
RD-polynomial
1s t
order 2 n d
order 3 r d
order
(1) (2) (3)
Resettled municipality (=1)
Adjusted R 2
Observations
Resettled municipality (=1)
Adjusted R 2
Observations
Resettled municipality (=1)
Adjusted R 2
Observations
Dependent variable:
Emigration rate (%)
0.659*** 0.605*** 0.572***
(0.064) (0.091) (0.131)
0.172 0.172 0.172
72006 72006 72006
Dependent variable:
Immigration rate (%)
0.456*** 0.409*** 0.400**
(0.072) (0.103) (0.149)
0.098 0.098 0.098
72006 72006 72006
Dependent variable:
Net immigration rate (%)
-0.203** -0.196* -0.172
(0.065) (0.094) (0.134)
0.101 0.101 0.101
72006 72006 72006
Note: Panel estimates from Equation (1) in the main text. Only municipalities within a 15 km band around the
ethnic border are considered. Municipalities on the part of the ethnic border closest to Bavaria (West Germany)
are excluded from the sample (see Figure C.4). Results control for year fixed effects, an RD-polynomial,
region fixed effects, population fixed effect, log of altitude, terrain roughness, log of the distance to the
country border, and 2n d
order polynomial of longitude and latitude. Symbols represent statistical significance
of differences in means between resettled and not resettled municipalities with ' = p < 0.1,* = p < 0.05,
** = p < 0.01, *** = p < 0.001. Values in brackets are standard errors clustered by municipality.
86
from the Northeast part of the Czech Republic characterized by a very ragged ethnic
border.3 6
Figure C.5: Municipalities within 15 km to the ethnic border after omitting Northeast part
of the Czech Republic
Share of ahnic
Germans (%)
•
l90L LOOj ReseLiled minid politic*
(50, ft&j
\t\ I Of Kol resetted municipuFiiic!Lllmic
burnt!tr
The exclusion of these municipalities from the sample has virtually no impact on the
estimated effects in RD-analysis (see Table C.10). In the RD-regression the estimated
impact of resettlement on the emigration rate is now marginally lower at 0.5 to 0.7
percentage points, as in the baseline specification, the estimated impact of the emigration
rate also remains unchanged at 0.4 to 0.5 percentage points. The impact on net immigration
is, however, insignificant for higher order RD-polynomials. This is in line with other
results in the main part of the paper and suggests that results for net immigration rates
are somewhat less robust (and limited only to the 1980s) than those pertaining to and net
immigration rates.
36. For the sample of municipalities, included in this robustness check, after the exclusion of Northern
Moravia see Figure C.5.
87
Table C I O : Robustness test: Exclusion of a part of ethnic border characterized by a large
number of irregularities
RD-polynomial
1s t
order 2 n d
order 3 r d
order
(1) (2) (3)
Dependent variable:
Emigration rate (%)
Resettled municipality (=1) 0.734*** 0.631*** 0.546**
(0.081) (0.123) (0.179)
Adjusted R 2
0.154 0.154 0.154
Observations 57763 57763 57763
Dependent variable:
Immigration rate (%)
Resettled municipality (=1) 0.516*** 0.440** 0.494*
(0.094) (0.140) (0.204)
Adjusted R 2
0.085 0.085 0.085
Observations 57763 57763 57763
Dependent variable:
Net immigration rate (%)
Resettled municipality (=1) -0.218** -0.191 -0.052
(0.080) (0.120) (0.170)
Adjusted R 2
0.092 0.092 0.092
Observations 57763 57763 57763
Note: Panel estimates from Equation (1) in the main text. Only municipalities within a 15 km band around
the ethnic border are considered. Municipalities on the part of the ethnic border characterized by a large
number of irregularities are excluded from the sample (see Figure C.5). Results control for year fixed
effects, an RD-polynomial, region fixed effects, population fixed effect, log of altitude, terrain roughness,
log of the distance to the country border, and 2n d
order polynomial of longitude and latitude. Symbols
represent statistical significance of differences in means between resettled and not resettled municipalities
with ' = p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.001. Values in brackets are standard errors
clustered by municipality.
88
CIO Robustness test: Divergent regional trends
One concern with our baseline specification may be that regions are following different
time trends in the outcome variable. These divergent trends may be a results of the new
settlers "settling in" to their new region of residence, could be due to longer run trends in
economic development over our rather long period of interest (1971-2015) or may result
from the many major institutional reforms in the period (such as e.g. transition) that could
impact on the attractiveness of certain regions analyzed. Since such could bias our baseline
results we also extended specification (1) to include linear time trends each region that are
allowed to differ for the period period before and after 1989, to account for the major shift
in the institutional environment and correspond to trends observable in outcome variables.
Results reported in Table C . l 1 are highly consistent with our baseline estimates as they
do differ from those by less than 0.01 percentage points.
C. 11 Robustness test: Using balanced panel data set
As noted in the main part of the paper and detailed in Appendix B , we use unbalanced
panel data on the municipalities to include the maximum number of observations possible.
Focusing on this unbalanced data may lead to biased results if there are systematic difference
in missing observations of resettled and not resettled municipalities. Since such selectivity
cannot be entirely precluded we repeat estimates using a balanced panel of municipalities.
This was constructed by using only those municipalities for which annual observations are
available for all years from 1971 to 2015.
Table C . l 2 shows the results for the causal impact of resettlement on residential
migration. These results are highly consistent to results reported in the main part of the
paper. The estimated treatment effects differ by less than 0.1 percentage points from those
in the main part of the paper.
89
Table C. 11: Robustness test: Divergent regional trends
RD-polynomial
1s t
order 2 order 3 order
(1) (2) (3)
Dependent variable:
Emig;ration rate (%)
Resettled municipality (= 0.738*** 0.687*** 0.566***
(0.063) (0.092) (0.133)
Adjusted R 2
0.172 0.172 0.172
Observations 82423 82423 82423
Dependent variable:
Immigration rate (%]
Resettled municipality (==D 0.505*** 0.466*** 0.410**
(0.070) (0.103) (0.148)
Adjusted R 2
0.096 0.096 0.096
Observations 82423 82423 82423
Dependent variable:
Net immigration rate (%)
Resettled municipality (= -0.233*** -0.221* -0.156
(0.063) (0.092) (0.131)
Adjusted R 2
0.100 0.100 0.100
Observations 82423 82423 82423
Note: Panel estimates from extended Equation (1) in the main text. Only municipalities within a 15 km band
around the ethnic border are considered. Results control for year fixed effects, an RD-polynomial, region
fixed effects, population fixed effect, log of altitude, terrain roughness, log of the distance to the country
border, 2n d
order polynomial of longitude and latitude, and linear time trends for each region and period
before/after 1989. Symbols represent statistical significance of differences in means between resettled and
not resettled municipalities with ' = p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.001. Values in
brackets are standard errors clustered by municipality.
90
T a b l e d 2: Robustness test: Balanced panel dataset
RD-polynomial
1s t
order 2 n d
order 3 r d
order
(1) (2) (3)
Resettled municipality (=1)
Adjusted R 2
Observations
Resettled municipality (=1)
Adjusted R 2
Observations
Resettled municipality (=1)
Adjusted R 2
Observations
Dependent variable:
Emigration rate (%)
0.821*** 0.780*** 0.667***
(0.078) (0.113) (0.164)
0.214 0.214 0.214
57974 57974 57974
Dependent variable:
Immigration rate (%)
0.467*** 0.478*** 0.435**
(0.080) (0.113) (0.168)
0.098 0.097 0.098
57974 57974 57974
Dependent variable:
Net immigration rate (%)
-0.354*** -0.302** -0.232f
(0.068)
0.101
57974
(0.101)
0.101
57974
(0.140)
0.101
57974
Note: Panel estimates from Equation (1) in the main text. Only municipalities within a 15 km band
around the ethnic border and with full set of observations are considered. Results control for year fixed
effects, an RD-polynomial, region fixed effects, population fixed effect, log of altitude, terrain roughness,
log of the distance to the country border, and 2n d
order polynomial of longitude and latitude. Symbols
represent statistical significance of differences in means between resettled and not resettled municipalities
with ' = p < 0.1, * = p < 0.05, ** = p < 0.01, *** = p < 0.001. Values in brackets are standard errors
clustered by municipality.
91
C.12 Robustness test: Inclusion of outliers
In the main part of the paper we excluded the top 1% of all observations in terms of
emigration and immigration rates to avoid using implausible values caused by unobserved
shocks.3 7
Our motivation for this was that the RD-estimates (due to few observations
directly at the border) may be very sensitive to such outliers. We therefore estimated
RD-estimates with and without eliminating such outliers.
Results in Table C. 13 indicate that depending on the functional form of the RD-polynomial,
emigration rates were by 0.5 to 0.9 percentage points and statistically significantly higher
in resettled than in not resettled municipalities. This is comparable or only slightly higher
than the 0.6 to 0.7 percentage points in the baseline estimates. The results also suggest
that immigration rates to the resettled municipalities from other regions were by 0.4 to 0.7
percentage points and statistically significantly higher in all estimates than in not resettled
municipalities (rather than 0.4 to 0.5 percentage point in the baseline results). Finally, also
the estimated effect of resettlement on net immigration rates are negative but lack statistical
significance with the exception of the specification with first order RD-polynomial.
C. 13 Robustness test: Impact of the resettlement by part of the country
Since resettled municipalities are located across different parts of the border of today's
Czech the estimated overall impact of the resettlement could be driven by some unobserved
local-specific characteristics of different border segments. In one robustness test we
therefore split our estimation sample into three sub-samples by nearest foreign country of
the municipality (Poland, Germany, and Austria) that coincide with the northern, western,
and southern parts of the Czech Republic.
The descriptive evidence presented in Figure C.6 suggests that differences between
resettled municipalities have higher migration rates than not resettled ones in all parts
of the country. Results of the regression (1) estimated separately for each sub-sample
corroborate this observation (see Table C. 14). Overall the results follow the pattern of the
pooled baseline estimates. The effect of the resettlement on emigration rate is positive and
significant in all parts of the country. On the other hand the estimates for immigration and
net immigration rate tend to be less statistically significant. Differences between regions
also seem to minor and—if anything—suggest a moderately weaker impact if estimates are
based solely on regressions focusing on the Austrian border.
37. Vaněk (1996) discusses an example of such a when he states that 116 municipalities were destroyed
due to lignite mining in Northwest part of the Czech Republic.
92
Table C.13: Robustness test: Inclusion of outliers
RD-polynomial
1s t
order 2 order 3 order
(1) (2) (3)
Dependent variable:
Emig;ration rate (%)
Resettled municipality (= 0.878*** 0.655*** 0.543**
(0.083) (0.123) (0.175)
Adjusted R 2
0.161 0.161 0.161
Observations 83849 83849 83849
Dependent variable:
Immi:gration rate (%]
Resettled municipality (==D 0.701*** 0.530*** 0.441*
(0.100) (0.139) (0.197)
Adjusted R 2
0.069 0.069 0.069
Observations 83849 83849 83849
Dependent variable:
Net immigration rate (%)
Resettled municipality (= -0.177* -0.125 -0.102
(0.084) (0.120) (0.157)
Adjusted R 2
0.076 0.076 0.076
Observations 83849 83849 83849
Note: Panel estimates from Equation (1) in the main text. Only municipalities within a 15 km band around
the ethnic border are considered. Results control for year fixed effects, an RD-polynomial, region fixed
effects, population fixed effect, log of altitude, terrain roughness, log of the distance to the country border,
and 2n d
order polynomial of longitude and latitude. Symbols represent statistical significance of differences
in means between resettled and not resettled municipalities with ' = p < 0.1, * = p < 0.05, ** = p < 0.01,
*** = p < 0.001. Values in brackets are standard errors clustered by municipality.
93
Figure C.6: Migration rates by part of the country
Poland
10
o
-5H
-10H
£ 5-
u
3
c
.2 o-
3
blj
-5
-10H
Germany
Austria
io H
OH
-5
-10H
Emigration
rale
Immigration
rale
Net immigration
rate
h-j—1 Not resettled municipalities |—'—I Resettled municipalities
Source: CZSO, own calculations.
Note: Migration rates are reported for municipalities in the estimation sample by nearest foreign country.
94
Table C.14: Robustness test: Impact of the resettlement by part of the country
Part of the country
Municipalities close to Poland Municipalities close to Germany Municipalities close to Austria
RD-polynomial RD-polynomial RD-polynomial
1s t
order 2 n d
order 3r d
order 1s t
order 2 n d
order 3r d
order 1s t
order 2 n d
order 3r d
order
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Dependent variable: Emigration rate (%)
Resettled municipality (==D 0.780*** 0.802*** 0.761*:
** 0 992*** 1.063*** 0.767t
0.439*** 0.427** 0.417*
(0.083) (0.121) (0.170) (0.155) (0.278) (0.445) (0.111) (0.150) (0.199)
Adjusted R 2
0.179 0.179 0.180 0.149 0.149 0.149 0.153 0.154 0.154
Observations 40685 40685 40685 22897 22897 22897 18841 18841 18841
Dependent variable: Immigration rate (%)
Resettled municipality (==D 0.532*** 0.577*** 0.488*:
0.767*** 0.736* 0.659 0.339t
0.486
(0.087) (0.127) (0.182) (0.171) (0.301) (0.469) (0.148) (0.194) (0.301)
Adjusted R 2
0.091 0.091 0.091 0.079 0.079 0.079 0.082 0.082 0.082
Observations 40685 40685 40685 22897 22897 22897 18841 18841 18841
Dependent variable: Net immi£;ration rate (°/<
Resettled municipality (= -0.249** -0.224t
-0.274t
-0.224 -0.327 -0.108 -0.172 -0.089 0.068
(0.079) (0.117) (0.156) (0.156) (0.263) (0.381) (0.129) (0.163) (0.252)
Adjusted R 2
0.107 0.107 0.107 0.098 0.099 0.099 0.095 0.096 0.096
Observations 40685 40685 40685 22897 22897 22897 18841 18841 18841
Note: Panel estimates from Equation (1) in the main text. Only municipalities within a 15 km band around the ethnic border are considered. Results control for year
fixed effects, an RD-polynomial, region fixed effects, population fixed effect, log of altitude, terrain roughness, log of the distance to the country border, and 2n d
order
polynomial of longitude and latitude. Symbols represent statistical significance of differences in means between resettled and not resettled with ^ = p < 0.1, * = p < 0.05,
** = p < 0.01, *** = p < 0.001. Values in brackets are standard errors clustered by municipality.
C.14 Robustness Check 7: Estimation of values data using the full set of
values for responses
As a final robustness check we conducted a linear regressions analysis of the values data
used in the main part of the paper using the full set of values for responses. These robustness
checks suggest that our choice of coding has few consequences for the quantitative results
(see Table C.15). The estimated coefficients once more remain insignificant in most cases.
The only exceptions being hobbies, health and material conditions in the old cohort. As
also found in the main part of the paper the older cohort in resettled municipalities puts a
higher value on the importance of hobbies. Furthermore, in contrast to the results in the
main part of the paper, the same applies to the importance of material conditions, while the
importance given to health is lower in the resettled municipalities.
96
Table C.15: Estimation of values data using the full set of values for responses
Dependent variable:
Things and values very importantin personal life (Likert scale, 4 = "Very important", 1= "Not important at all")
Nature,
environment
Job, Relationships Faith,
occupation spiritual values
Hobbies Housing Friendship Health Family life
and children
Material
conditions
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Panel A: Young cohort (18-29)
Resettled municipality (== D 0.048 0.107 0.019 -0.120 0.130 0.058 0.039 -0.074 -0.150 0.143
(0.106) (0.091) (0.117) (0.163) (0.117) (0.079) (0.098) (0.113) (0.149) (0.102)
Adjusted R2 0.045 0.051 0.067 0.098 0.028 0.021 0.045 0.036 0.218 0.027
Observations 680 680 680 680 680 680 680 679 680 679
Panel A: Young cohort (18-29)
Resettled municipality (== D -0.041 0.112 -0.212 -0.154 0.320* -0.021 -0.030 -0.082t
-0.104 0.302*
(0.114) (0.235) (0.137) (0.181) (0.153) (0.099) (0.107) (0.046) (0.103) (0.141)
Adjusted R2 0.030 0.013 0.017 0.091 0.002 0.013 0.012 0.040 0.125 0.048
Observations 611 602 609 608 610 611 609 610 610 610
Table reports estimated coefficients on an indicator variable for resettled municipalities after controlling for municipality (log of altitude, terrain roughness, log of distance
to the country border, region fixed effect, and population fixed effect) and personal (age group, education, labor market status, marital status, household size, and for being
born in the municipality of residence) characteristics. Standard errors clustered by municipality are reported in parentheses are :t
= p < 0.1, ** = p < 0.05, *** = p < 0.01,
*** = p < 0.001. Descriptive statistics are the mean and the standard deviations (in parentheses).
References
Calonico, Sebastian, Matias D. Cattaneo, Max H . Farrell, and Rocio Titiunik. 2020. rdrobust:
Robust Data-Driven Statistical Inference in Regression-Discontinuity Designs. R
package version 0.99.9. https://CRAN.R-project.org/package=rdrobust.
Kelly, Morgan. 2019. "The standard errors of persistence."
. 2020. "Understanding persistence."
Vaněk, Miroslav. 1996. Nedalo se tady dýchat: Ekologie v českých zemích v letech 1968 až
1989. Praha: Ústav pro soudobé dějiny A V ČR.
98