We examine the long-run effects of forced migration for individuals who were displaced from Eastern Europe to Germany in the aftermath of World War II. Evidence suggests that displaced individuals were worse off economically, facing a considerably lower income and a higher unemployment risk than comparable nondisplaced Germans, even 20 years after being expelled. We extend this literature by investigating mortality outcomes. Using social security records that document the exact date of death and a proxy for pre-retirement lifetime earnings, we estimate a significantly and considerably higher mortality risk among forced migrants compared with nondisplaced West Germans. The adverse displacement effect persists throughout the earnings distribution except for the top quintile. Although forced migrants were generally worse off regarding mortality outcomes, those with successful labor market histories seem to have overcome the long-lasting negative consequences of flight and expulsion.
In the aftermath of World War II, approximately 8 million Germans were forcibly displaced from Eastern Europe to West Germany. This mass expulsion is one of the largest population movements in history, leaving the bulk of forced migrants to arrive in a country that suffered from major destruction. Against this background and the fact that displaced individuals had to restart with little or no possessions, the integration of this large inflow of migrants was a difficult task. Even though the German government took tremendous efforts to reduce the burden of war-related financial losses, the integration of forced migrants remained a societal challenge. Empirical research has shown that these migrants exhibited considerably lower incomes and a higher unemployment risk even 20 years after the mass exodus (Bauer et al. 2013; Lüttinger 1986). Emphasizing the role of the local environment, recent evidence further indicates that counties with high industrialization and low inflows were much more successful in integrating forced migrants within two decades after the war (Braun and Dwenger 2017). An open question that remains is whether flight and expulsion have long-run consequences across the entire life cycle.
This article aims to fill this gap by investigating lifetime outcomes of forced migrants. In particular, we analyze the effect of displacement on mortality, comparing displaced persons predominantly from the former Eastern territories of the German Reich with nondisplaced West Germans. We further examine effect heterogeneity across completed labor market biographies—that is, pre-retirement lifetime earnings—to learn more about mechanisms that transmit the effects of displacement over the life course.
The identification of the displacement effect is based on exogenous variation from the large-scale inflow of forced migrants into postwar Germany between 1944 and 1950. This migration flow involved several features that allow us to isolate the effect of flight and expulsion on the mortality of forced migrants relative to the indigenous West German population.
First, the predominant part of the displaced population under study had no choice to stay in their home region and migrated in a short period, which allows our analysis to avoid the self-selection problems and cohort effects inherent to many empirical studies on the integration of migrants (Borjas 1985, 1987, 1999). Second, selective return migration (for a discussion, see Lubotsky 2007) was virtually impossible because of the Iron Curtain until 1989. Even after the fall of the wall, the elderly population under study was unlikely to have re-migrated after settling in West Germany for more than 40 years. Third, language abilities and transferability of skills should not play a role for the group of migrants analyzed in this study given that most of the displaced persons were German speakers, received their education at German schools, and grew up under the same institutional setting. Finally, prewar characteristics of displaced and nondisplaced persons were largely similar in terms of socioeconomic characteristics (Bauer et al. 2013), providing confidence that the indigenous West German population is a valid control group. To avoid biased estimates due to prewar regional differences potentially affecting mortality outcomes among displaced and nondisplaced, we use information on regional disparities in mortality rates at the time of birth. Note, however, that we might not identify the average treatment effect on the treated (ATT) because the huge influx of forced migrants into postwar West Germany could have had general equilibrium effects that may also affect the mortality rates of our control group.
Studying the elderly population at risk from age 68 onward, we find that displaced individuals exhibit a considerably and significantly higher mortality risk than comparable nondisplaced persons. Our estimates of the mortality hazard ratio indicate that the mortality risk of displaced men is 15 % to 27 % higher than for nondisplaced men and 4 % to 11 % higher for displaced women than for nondisplaced women. Using a measure of pre-retirement lifetime earnings, we further show that the adverse displacement effect persists through large parts of the earnings distribution. Within the top quintile, not only does the adverse effect disappear, but our estimates even indicate a slightly lower mortality risk for those migrants. This suggests that displaced individuals with exceptionally successful labor market biographies are able to overcome the long-lasting negative consequences of flight and expulsion.
Although more than 65 million people were forcibly displaced worldwide in 2015 (United Nations High Commissioner for Refugees 2016), little is known about the consequences of forced migration.1 A growing literature has examined the effects of forced migration on host countries’ labor markets (Ruiz and Vargas-Silva 2015) or its impact on economic outcomes and integration of displaced persons (e.g., Bauer et al. 2013; Braun and Dwenger 2017; Falck et al. 2012; Fiala 2015; Sarvimäki et al. 2009), finding predominantly negative effects2 of forced migration in the short and medium term. Our contribution to this literature is to examine the effects of forced migration across the entire life cycle, focusing on mortality as an objective health outcome.
Investigating the long-run consequences of forced migration also contributes to the demographic literature regarding the impact of forced migration on mortality (e.g., Haukka et al. 2017; Saarela and Elo 2016; Saarela and Finnäs 2009). These studies investigated a specific forced migration episode in Finland, where some areas were ceded to the Soviet Union in the aftermath of World War II. In terms of historical timing and territorial losses, the Finish case is very similar to the German one. Our central finding—that forced migration increases long-run mortality outcomes even decades after being displaced—is consistent with this literature. Although our estimates of the mortality risk ratio (internally displaced to nondisplaced) are somewhat larger, this seems highly plausible when taking into account that postwar displacement in Finland was rather well organized and generous in terms of compensations (see, e.g., Haukka et al. 2017), whereas the displacement from Eastern Europe to postwar Germany took place under arguably harsher conditions.
Furthermore, our study relates to the mortality literature that largely agrees on conditions and experiences early in life to have profound consequences later in life.3 Because the population that we study was predominantly aged 18–45 by the time of expulsion, our results closely reflect what has recently been termed the cumulative disadvantage over the life course (Case and Deaton 2015, 2017). Although Case and Deaton referred to a different study group,4 the mechanisms behind the cumulative disadvantage are strikingly close to what we find with respect to mortality outcomes of forced migrants in Germany: adverse conditions at the outset of their living in West Germany accumulated to disadvantages over the entire life cycle. Estimating higher mortality rates in the elderly displaced population indicates that the experience of expulsion has negative effects that are largely irreversible.
Finally, the vanishing displacement effect at the upper margin of the earnings distribution indicates an inverse relationship between lifetime earnings on the one hand and the mortality risk difference between displaced and nondisplaced persons on the other hand. This insight is in line with the inverse relationship between income and health (or mortality) as one of the most robust empirical findings in economics and medical science,5 and it is highly policy-relevant because it highlights the importance of successful economic integration.
West Germany experienced a major population inflow in the aftermath of World War II. Between 1944 and 1950, approximately 7.9 million displaced persons—or more than 16 % of the total West German population at that time—entered the country (Reichling 1958).6 Figure 1 shows the regions of origin of these migrants. Among them were 4.4 million National Germans (Reichsdeutsche) who used to live in areas that were located east of the Oder-Neisse rivers, such as Silesia, East Prussia, or Pomerania. Additionally, individuals of ethnic German origin (Volksdeutsche) who lived as minorities in foreign countries made up 3.5 million migrants, who mainly came from Sudetenland, which was located in Czechoslovakia close to the German border (Connor 2007). The map also includes territories that were already lost after World War I but included individuals who were displaced after World War II and thus belong to the primary study group in this study.
Expulsion of Germans from their homelands took place in three phases, starting in October 1944, when advancing Soviet troops entered through the eastern border of the German Reich. The cruelty of the approaching Red Army caused Germans to flee westward. These treks were very risky: forced migrants were exposed to extreme weather conditions (e.g., strong winter), malnutrition, and air strikes by the Allied troops. Many did not survive this exhausting flight. By the end of the war in May 1945 and the unconditional surrender of Nazi Germany, the second phase of displacement, the so-called wild expulsions conducted by Polish and Czechoslovakian authorities started. Those affected by these wild expulsions were forced to leave their homes and personal belongings behind and were put into internment camps, where they had to stay before being transferred to Germany. The Potsdam Treaty from August 1945 started the third phase of expulsion. As a result of this treaty, the eastern border of Germany was shifted westward to the Oder-Neisse rivers, and Germany was divided into four zones of occupation. Furthermore, the treaty legalized resettlement of the remaining German population in Poland and Czechoslovakia, which lasted until 1950.
Until 1950, and especially during the early phases of expulsion, self-selection of the displaced was arguably a minor issue. By 1946, approximately 6.2 million displaced persons were counted in West Germany, which amounted to almost 80 % of the 7.9 million who eventually arrived until 1950 (Reichling 1958:15).7 This indicates that the majority of forced migrants were displaced in the early phases under wild expulsions. For this group, displacement was almost universal and largely involuntarily.8
Many displaced persons who finally arrived in West Germany had to start their new lives without any financial means or a social network because larger groups of forced migrants from the same home region were often distributed geographically (Connor 2007). They entered into a country that was earmarked by vast destruction of the war accompanied by severe food and housing shortage. Tensions between the nondisplaced and the newcomers arose quickly, especially in rural areas, where the nondisplaced were less open to outsiders, experienced less war damage, and were less knowledgeable about the actual circumstances of the expulsion. In contrast, displaced persons in larger cities were soon recognized as precious economic resources for war reconstruction. For forced migrants, who had to rebuild their lives from scratch, this often meant working as unskilled laborers rather than as self-employed farmers, which many did before being displaced. Religious differences between the nondisplaced and newcomers produced additional tensions,9 and fear of foreign influence among the nondisplaced and local politicians did little to foster integration (Connor 2007).
The postwar West German government, however, implemented several measures to support the integration of newly arrived Germans from the former Eastern territories of the German Reich and other Eastern European countries. The Law of Equalization of Burden (Lastenausgleichgesetz) from 1952 compensated forced migrants as well as the indigenous German population for war-related losses of property or savings. Another important law was the Law for Foreign Pensions, which acknowledged specific periods of work for the pension claims of forced migrants and thus reduced the burden for forced migrants. This law is crucial for the following empirical analysis because it allows us to identify displaced individuals in our data.
Data and Empirical Strategy
Data and Variables
The empirical analysis is based on large-scale administrative data from German pension insurance accounts. Specifically, we use individual records that document the termination of individual pensions due to death (Rentenwegfall) for the universe of deceased individuals who formerly contributed to the German public pension system. The pension shortfall is available for the entire German population covering the years 1994–2013,10 accessible via remote computing provided by the German federal pension insurance (Deutsche Rentenversicherung).11
Using these social security records is advantageous for answering the questions at hand. First, all data points are generated within the administrative process and document the exact dates of birth and death. Second, pension shortfall records provide an almost universal picture on mortality in Germany because they include everyone who has ever been registered for an insurance account and received a pension at some point in time. Because we study a sample of the deceased population from age 68 onward, it is almost certain that eligible individuals received pension benefits. In fact, according to official mortality statistics, the data source covers 82 % of all death cases and 96 % and 75 % of deaths among men and women, respectively.12 People who were never actively registered in the German public pension system did not have a pension insurance account and thus do not appear in the data. Men exhibited high labor force participation rates, and thus most of them had pension insurance accounts.13 The coverage rate was smaller for women because some of them never worked. In 1960, when women in our sample were at prime working age, women’s labor force participation rate was only 47 %, compared with 90 % among men (Federal Statistical Office 2018). Lower labor force participation rates among women translate into lower registration rates in the public pension system. For women who never worked, the only way of becoming registered in the public pension system would be the accumulation of periods of child-rearing.
Because systematic differences in mortality rates between the country of origin (former Eastern Territories of the German Reich) and the destination country (West Germany) would compromise our results, we supplement the analysis with historical mortality data from statistical yearbooks of the German Reich.14 Variation in regional mortality patterns is possible, for example, because of the effects of maternal nutrition and health status on mortality (see Barker 1990, 1995). Health and mortality outcomes could also be influenced by general living standards and region-specific industries that may involve physically demanding work. In fact, while Silesia and Bavaria had quite low life expectancies, regions in northwest Germany reported higher life expectancies (Kibele et al. 2015). Finally, differential fertility rates across regions could affect cohort sizes of the displaced population differently than for the nondisplaced. To encounter these challenges, we calculate the annual birth and death rates by relating the number of births and deaths, respectively, to the number of residents within each region in a given year. The corresponding variation across region and time as shown in Fig. 2 is used to account for regional disparities in fertility and mortality patterns during the relevant pretreatment period.
Separating the Displaced From the Nondisplaced
We identify displaced persons based on the legal framework of the Law for Foreign Pensions (LFP) (Fremdrentengesetz) enacted in 1960. The goal of the LFP was to acknowledge work periods or work-related periods (e.g., unemployment, illness, pregnancy, and child-rearing) accumulated previous to expulsion from Eastern Europe for the pension claims of forced migrants in the West German public pension system.
The average pre-displacement earnings that determined pension claims are based on the displaced person’s educational level and occupational group, referring to average earnings of these groups in Germany. For each person with LFP-related pension claims, the data document the region of origin, the amount of pre-displacement earnings, and the length of the pre-displacement work period in months that accounted for later pension claims.
Based on this information, we define forced migrants as those persons who entered West Germany between 1944 and 1950, encompassing the period of mass expulsion, who obtained pension claims referring to the LFP, and who were expelled from the former Eastern Territories or one of the Eastern European countries affected by the mass expulsion. Unique identification of forced migrants via LFP pension claims requires the existence of such claims. The LFP provided a generous program that aimed at treating forced migrants similar to the West German population in terms of pension claims. Because the LFP accounted not only for working times but also for other labor market–related periods, it is unlikely that displaced persons were not covered or did not call on these claims because their old-age incomes otherwise would deteriorate. Nondisplaced persons are defined to be individuals with German citizenship and no recent migration history who were living in West Germany during the study period. We drop all observations that satisfy neither the definition of the displaced nor the one of the nondisplaced.
Definition of Mortality
Throughout this article, we define mortality as the age at death. From the observed monthly dates of birth and death (the latter of which is documented from pension shortfall due to death), we calculate the age at death for each individual in the data. We then use this mortality measure to model the mortality hazard given by the probability of dying conditional on having survived until the respective age.
Approximating Lifetime Earnings
To examine mortality differences across the entire earnings distribution, we approximate pre-retirement lifetime earnings using a variable that includes the individual sum of earnings points (EP). This measure documents pension claims that predominantly consist of labor earnings. One EP from labor earnings of individual i in year t is defined as , where yit are labor earnings of individual i in year t, and are the average labor earnings of all contributors of the public pension system in that year. Intuitively, one EP reflects the relative earnings position of each individual in a given year. For example, an employee receives exactly one EP per year if she contributes at average earnings and two EP if she contributes at twice the average earnings. For each individual, our data include the sum of annual earnings points given by , where T is the last year of gainful employment before an individual retires. Although EP are not a perfect measure of earnings because they also include creditable periods—for example, from education or child-rearing (see section A of the online appendix for a detailed description of its limitations)—we argue that they are a fairly reasonable proxy of pre-retirement lifetime earnings.
The actual distribution of earnings is shown in Fig. 3, reflecting the typical labor force patterns of men and women within the observed birth cohorts: whereas men overwhelmingly worked in full-time jobs, often with working biographies of 40 or more years (sample mean = 44 EP), women worked either part-time or not at all (sample mean = 20 EP). The patterns in Fig. 3 summarize these labor market histories by showing a remarkable spike for women at low EP values (between 8 and 12) due to creditable periods of child-rearing (maximum of 2 EP per child until 1992), and another spike between 20 and 30 EP that reflects gainful employment dominated by part-time work. Among men, the patterns are much more homogenous, with the highest densities between 40 and 60 EP, reflecting typical full-time work biographies.
Table 1 further documents substantial gender differences in mortality patterns and labor market histories. Women strongly outlived their male counterparts and were less active in the labor market, having collected a lower number of earnings points and obtained fewer contribution months to the pension insurance.
Observational Plan and Estimation Sample
The final sample documents cases of death between 1994 and 2013 for the West German population. The observational plan is depicted in a lexis diagram (Fig. 4), spanning the space of sampled birth cohorts (baseline: 1885–1925) over age and years. Based on this observational plan, the final estimation sample consists of 4.98 million observations (2,985,403 females and 1,992,963 males).
The particular cohort choice is motivated by identifying forced migrants according to their pre-migration pension claims. For this purpose, we need to ensure that individuals have reached working age after they migrated. Taking into account the beginning of mass expulsion in 1944, we set the minimum age at migration to 18 by restricting the sample to birth cohorts born in 1925 or earlier. Choosing younger cohorts (born after 1925) increases the likelihood that displaced individuals are mistakenly sampled as nondisplaced Germans because they did not have the opportunity to collect creditable pension claims previous to migration. The lower margin of the cohort distribution (i.e., older persons) is unrestricted and establishes only that individuals born in 1885 are the oldest persons to reach the observation period (starting in 1994) in terms of survival.15 Our baseline cohort choice balances the sample at two margins: it allows for coverage of a large number of birth years and ensures that the displaced are uniquely distinguished from the nondisplaced. Later, we relax this restriction by also including younger cohorts up to 1929 to show the sensitivity of our baseline estimates against this choice.
Sampling of Deceased Individuals and Age Selection
The sample consists of deceased individuals only, which is a direct consequence of how the administrative records document pension shortfall due to death. The fact that individuals enter the sample only if they die within the observation period introduces a specific type of selection. The population at risk of dying, however, includes everyone who receives any type of pension.16
The observation period (calendar years 1994–2013) and the cohort selection (birth years 1885–1925) imply that all persons in the sample have reached at least their 68th birthday. Thus, the sample is positively selected in age: all results and interpretations refer to the elderly population at the upper margin of the age distribution. This means that we study a population that is predominantly retired (more than 99 % claimed a pension by the age of 68), allowing us to observe the entire pre-retirement earnings history in terms of completed labor market biographies. Comparing earnings biographies is more feasible if they are completed by the date of retirement, which holds for our sample.17 This allows us to estimate the mortality effect of displacement across the lifetime earnings distribution to reveal heterogeneity in connection with labor market integration.
By the end of the observation period in 2013, the youngest cohort (1925) had reached age 88, implying that we face only minor selection issues concerning the unknown population of those who are not sampled because they were still alive.
Positive Selection of Forced Migrants
An important question regarding the influx of forced migrants into West Germany is who eventually arrived in West Germany after the expulsion. This comprises concerns not only about the prevalence of death during the flight but also about self-selection into East and West Germany. The crucial point about all these sources of selection is that they potentially induce a positively selected pool of displaced persons. Some of the selection mechanisms and pathways are not well documented or even unobservable. However, based on aggregate statistics, we substantiate that the displaced population was arguably a positively selected group upon arrival in West Germany, implying that our estimates constitute a lower bound of the effect of forced migration on mortality.
First, in 1950, the displaced were overrepresented at younger ages (up to age 40) and underrepresented in older age groups (Reichling 1958:54–56). One explanation for the different age composition of displaced persons is that missing individuals among the displaced perished in the process of occupation and expulsion (Reichling 1958). Even if one were to argue that young and middle-aged persons should be subject to war-specific losses, this should apply to the nondisplaced German population similarly and therefore would not confound our comparisons of the displaced with the nondisplaced. Thus, for those who eventually arrived in Germany, the overrepresentation of displaced individuals at younger ages suggests a type of positive selection: especially younger people are thought to be healthier and more resistant to the strenuous flight.
Second, sorting from East to West Germany amounted to a total of 500,000 displaced individuals who migrated from East to West between 1950 and 1955 (Reichling 1958), accounting for approximately 5 % of the displaced persons in West Germany. At the same time, only 50,000 displaced persons moved into the opposite direction, from West to East Germany. This asymmetric movement toward West Germany can be explained historically by a much poorer economic development in East Germany already toward the mid-1950s. To prevent further out-migration, the wall was eventually built in 1961. The dominance of East-to-West movements speaks in favor of positive selection if those with higher motivation or abilities are actually attracted to invest in moving toward a more promising economic environment in West Germany. In addition, changes in the initial distribution of displaced individuals across Germany were rather unlikely. After being assigned to a specific locality in West Germany, strict moving restrictions prevented displaced persons from moving (for details, see Müller and Simon 1959). These restrictions were relaxed in 1949 but were rather strict thereafter.
The share of forced migrants in our sample, at approximately 0.7 %, seems low considering that forced migrants represented no less than 16 % of the West German population just after World War II. However, because of the birth cohort restriction, the forced migrants in our sample were predominantly aged 25–45 at the time of entering the country (see Fig. B.1 in the online appendix). Only 30 % of the total inflow of the displaced in West Germany belonged to this age group (Reichling 1958). Moreover, our sample contains only those migrants who had reached at least their 68th birthday (i.e., had not died before 1994). Finally, having collected any earnings points that count for pension claims is a prerequisite to enter the sample. Despite these requirements and the age restriction, our final sample still includes more than 33,000 displaced persons (12,072 men and 21,031 women).
The nature of the sampling and the cohort-age restrictions partly explain the small number of displaced persons in the final sample. Although this reduces concerns of endogenous self-selection, different mortality patterns between the displaced and nondisplaced before they reach age 68 remain relevant for the unbiased estimation of the displacement effect. A quantification of these different mortality rates is possible based on census statistics from 1950 (Reichling 1958). To scale down the initial population share, we use the coverage rate of the pension records relative to official mortality statistics of 82 % of the total population.18 The attrition rate calculated from 1950 to 1993 is 81 % among the displaced and 58 % among the nondisplaced.19 A higher attrition rate among the displaced before age 68 coincides with a more positively selected pool at the beginning of the observation period in 1994. In terms of the measured mortality outcomes, this is plausible to the extent that healthier people tend to grow older. As for the positive selection in age, this would imply that we estimate a lower bound of the true displacement effect.
with extreme-value distributed survival spells (complementary log-log model) for i individuals. The survival time a reflects age measured in years. The indicator Di takes the value 1 for displaced individuals from the former Eastern territories and 0 for nondisplaced West German individuals. Our primary interest lies in the displacement effect, measured by the parameter α. We control for time-invariant observable characteristics that may affect the mortality hazard in vector xi, which includes, in all specifications, birth cohort dummy variables to balance out differences in birth cohort representation between displaced and nondisplaced persons. In some specifications, xi further includes earnings points (EP) as a proxy of lifetime earnings as well as region-specific birth and death rates to account for initial regional disparities in fertility and mortality patterns that may relate to mortality outcomes of the displaced and the nondisplaced later in life.
The sampling structure implies that individuals are at risk beginning at age 68. For any subsequent age, we include duration indicator variables in the model, estimating the parameters δa.20 This procedure is possible because of the large data set and is particularly advantageous because it leaves the baseline hazard in its most flexible version. All parameters in Eq. (1) are estimated separately for men and women.
Comparing the age at death distribution at the sample mean yields a first indication that nondisplaced West Germans outlive their displaced counterparts considerably (Table 1). The gap is statistically significant and amounts to 0.9 years among men and 1.6 years among women. How this translates into differences in the mortality distribution between displaced and nondisplaced persons is illustrated in Fig. 5.
Two findings are noteworthy. First, nondisplaced Germans lived considerably longer than displaced persons. Second, the mortality difference is larger among women. For example, observing individuals at risk beginning at age 68, Fig. 5 documents an 11 percentage point gap between displaced and nondisplaced women at age 88; that is, at age 88, 73 % of displaced women but only 62 % of nondisplaced women were deceased. Among men, the difference in the mortality distribution is smaller but still considerable. At age 88, we measure a 6 percentage point difference in mortality: 80 % of displaced men and only 74 % of nondisplaced men were deceased by that age.
To facilitate the interpretation of our estimates from Eq. (1), we present estimated mortality hazard ratios instead of coefficients or marginal effects. Our estimates of the mortality hazard ratio of displaced to nondisplaced men range from 1.15 to 1.27 (Table 2), indicating that the mortality risk of displaced men is 15 % to 27 % higher than for nondisplaced German men. Similarly, for women, the estimated mortality hazards translate to a 4 % to 11 % higher mortality risk of displaced women compared with nondisplaced women.21
The smallest estimates of the effect of displacement on the mortality risk are obtained from specifications that include only the displacement indicator, a set of duration dummy variables, and birth cohort dummy variables (columns 1 and 5 of Table 2). To make the two populations under consideration as homogeneous as possible, we also control for lifetime earnings22 and prewar disparities in regional birth and death rates. Although including lifetime earnings alone changes the results very little (columns 2 and 6 of Table 2), the difference in estimating the displacement effect on mortality increases when we control for regional birth and death rates (columns 3 and 7 of Table 2).
Differential Mortality Across the Earnings Distribution
The baseline estimates suggest that pre-retirement lifetime earnings (EP) do not influence differential mortality risks between displaced and nondisplaced persons. Estimates presented to this point, however, are measured at the mean of the earnings distribution.
Estimating the displacement effect separately by earnings quintile reveals effect heterogeneity across different margins of the distribution (see Table 3). For the first three quintiles of the earnings distribution, the mortality hazard ratio is considerably larger than 1, indicating that displaced individuals face a higher mortality risk. However, this ratio declines when moving toward the upper margin of the earnings distribution (for a graphical presentation, see Fig. B.2 in the online appendix). The measured effect is significantly smaller for the fourth quintile than for the third quintile. In the top quintile, the displacement effect significantly falls below 1. These principal patterns hold for both men and women and are robust across different specifications.
These results indicate that the adverse displacement effect, as previously measured at the sample mean, appears to be driven by individuals in the lower parts of the earnings distribution, while displaced individuals with exceptionally successful labor market biographies manage to overcome the long-lasting negative consequences of flight and expulsion. However, analyzing the displacement effect across the entire earnings distribution may introduce another layer of selection. In particular, if a displaced and a nondisplaced person have identical lifetime earnings, they are also likely to differ in some other unobservable characteristics. If, for example, forced migration reduced lifetime earnings, on average, then a displaced person with given lifetime earnings should have more favorable unobserved characteristics than a nondisplaced person with identical lifetime earnings.23 Despite these concerns about the bad control variable and connected issues of selection, heterogeneity of the displacement effect across the earnings distribution is particularly informative about the long-run transmission channel of the effect that is arguably mediated through poor integration.
In a first sensitivity check, we consider only the nondisplaced population living in the federal state of Bavaria in order to make the nondisplaced population more comparable with forced migrants, for two reasons. First, region-specific birth and mortality rates in Bavaria, measured at time of birth, are much more similar to those of the former Eastern Territories than to those of West Germany overall (Fig. 2). Second, Bavaria exhibited by far the largest inflow of displaced persons in the aftermath of World War II because of its geographical location in the southeast of Germany.24 The spatial distribution of displaced persons also did not change much over time—a fact documented by Schumann (2014), who found that population shocks in the aftermath of World War II are highly persistent. Table 3 shows that our main estimation results change very little for the full earnings point distribution as well as for the different quintiles if we compare only nondisplaced Bavarians with the displaced rather than using the full West German nondisplaced sample.
To identify displaced persons in the data, we base our core estimates on a strong age restriction that allows individuals to enter the sample only if they were born in 1925 or earlier. Columns 2–3 of Table 4 show the estimation results when this cohort restriction is relaxed to include younger cohorts as well. The adverse displacement effect appears to increase slightly when the baseline restriction is relaxed (born in 1925 or earlier) by further including the cohorts up to 1927 or 1929. As discussed earlier, adding younger cohorts to our sample increases the likelihood of misclassifying displaced persons as nondisplaced. One would expect the estimated mortality difference to become smaller as an increasing number of displaced with worse mortality expectations erroneously moves to the comparison group of nondisplaced. However, this is not entirely clear because of changes of the relative representation of the displaced across birth cohorts. Nevertheless, the overall changes are fairly small, indicating that our baseline estimates produce the most conservative results of the displacement effect on mortality.
A final concern is that heterogeneity between displaced and nondisplaced persons may arise from the length of individual working biographies. For example, individuals who start their working careers early may work in manual jobs rather than obtain a university degree and follow an academic track. Because this type of selection may correlate with health impairments accumulated over time, we take a subsample of individuals who contributed to the German public pension system at least for 40 years and thus must have started their working biography at young ages. Whereas the results for males are unchanged, the displacement effect strongly changes and becomes negative for females (column 4 in Table 4). Contributing to the pension scheme for at least 40 years is, however, a tough restriction that is fulfilled by only 60 % of the men and 10 % of the women in the sample. This indicates that the female sample in particular is positively selected in terms of labor force participation. In contrast, the results for the restricted male sample (who have at least 40 years of contribution) are almost unchanged compared with the baseline estimates. This finding is not surprising: it is much more common for men to contribute 40 or more years, and thus the sample composition remains rather similar. The results for long contribution periods are in line with the findings for the top earnings quintile (Table 3), suggesting that the adverse displacement effect disappears or even becomes positive among displaced individuals who are successfully integrated into the labor market.
In this article, we analyze the long-run consequences of forced migration on mortality. In the aftermath of World War II, almost 8 million Germans who were expelled from the former Eastern territories of the German Reich and other Eastern European countries arrived in West Germany. We use this natural experiment to identify the displacement effect on differential mortality patterns of displaced and comparable nondisplaced persons.
Our results show that the mortality risk of displaced individuals is substantially higher than among comparable nondisplaced Germans. These estimates are robust across several specifications, but they are also larger than the ones documented for a similar forced migration episode in Finland, where some areas were ceded to the Soviet Union in the aftermath of World War II. However, this seems highly plausible when taking into account that postwar displacement in Finland was rather well organized and generous in terms of compensations (see, e.g., Haukka et al. 2017), whereas the displacement from Eastern Europe to postwar Germany took place under arguably harsher conditions, especially during the early phases of expulsion.
Although controlling for pre-retirement lifetime earnings at the mean seems to have little influence, estimating the displacement effect within each quintile of the earnings distribution reveals that the adverse displacement effect is driven by displaced individuals in the lower parts of the earnings distribution. At the upper margin of the distribution, the adverse displacement effect not only becomes smaller in magnitude but turns positive. This effect heterogeneity, equally measurable among men and women, provides evidence that displaced individuals with exceptionally successful labor market biographies manage to overcome the long-lasting negative consequences of flight and expulsion.
The documented mortality gap between displaced and nondisplaced persons may operate through two distinct channels. The first one refers to the direct and long-lasting link of the traumatizing event of displacement on health outcomes later in life and its correlation with mortality. The second channel reflects indirect effects of poor integration of forced migrants into the labor market and the society. Although the German legislation adopted a number of different laws that aimed at reducing the burden of war-related financial losses (e.g., Lastenausgleichsgesetz), the integration of forced migrants remained a societal challenge. Despite an enormous labor demand in postwar Germany, unemployment rates among the displaced remained higher than among the nondisplaced, even in the long run (Bauer et al. 2013; Lüttinger 1986; Reichling 1958). Also impeding their integration were resentments of nondisplaced Germans toward the newcomers, which had already started arising during the postwar period characterized by severe food and housing shortages due to major destruction (Connor 2007).
The cumulative disadvantage over life plausibly describes that the consequences of expulsion reached far beyond local displacement and the loss of material possessions. The atrocities that forced migrants experienced during flight and expulsion seem to have caused long-lasting wounds that did not heal by the end of the war. For some, this was the beginning of poor integration into society and the labor market that cumulatively contributed to long-run disadvantages. The cumulative disadvantage represents a meaningful explanation for the transmission mechanism over life, thus permitting attribution of differential mortality to forced migration even decades after the expulsion took place.
We are grateful to Ronald Bachmann, Gerard van den Berg, Julia Bredtmann, Bernd Fitzenberger, Joseph-Simon Görlach, Elke Jahn, Rainer Kotschy, and Regina Riphahn for helpful suggestions, and we thank the editors and three anonymous referees for their valuable comments that helped to improve this article. We also thank the participants of the meetings of the DFG-SPP 1764 Priority Program (Essen, 2016), RWI Research Seminar (2016), RGS Conference (Dortmund, 2017), ESPE Conference (Glasgow, 2017) and the DFG-SPP 1764 International Conference (Nürnberg, 2018) for insightful discussions. We further thank the team of the research center of the German Federal Pension Insurance (FDZ-RV), in particular Ute Kirst-Budzak, Torsten Hammer, and Ingmar Hansen, for supporting the data processing. Fabian Dehos, Gökay Demir, and Jan Wergula provided excellent research assistance. Financial support from the German Research Foundation (DFG-SPP 1764) is gratefully acknowledged.
An extensive literature has examined voluntary migration. However, the process behind forced migration differs in many respects from voluntary migration decisions such that many of the policy implications and conclusions from studies on voluntary migrants may not be applicable to forced migrants (for an overview on the forced migration literature, see Ruiz and Vargas-Silva 2013).
Positive effects on long-term income of males in Finland have been argued to be due to an accelerated transition from traditional to modern occupations (Sarvimäki et al. 2009). A similar result was documented for those forced migrants in Germany who worked in the agricultural sector before migrating to West Germany (Bauer et al. 2013).
Individuals born in a recession die earlier than individuals born in times of economic prosperity (van den Berg et al. 2006). Recent evidence has suggested that entering the labor market in a recession increases mortality later in life even if initial losses in earnings have faded (Schwandt and von Wachter 2017).
Case and Deaton (2017) used the term cumulative disadvantage to describe recently increasing middle-age mortality rates among non-Hispanic whites without a college degree in the United States (1998–2015), triggered by increasingly poor labor market conditions at the time of labor market entry.
For recent empirical evidence on the link between income and health, see Chetty et al. (2016). What remains controversial is measuring the causal impact of income on health (or mortality) and vice versa. Low income could explain poor access to health care or less healthy nutrition. One could also think of poor health as reducing the ability to be gainfully employed or even as preventing people from investing in human capital. See Smith (1999) and Deaton (2003) for discussions.
The historical details regarding this mass migration have been documented thoroughly. Douglas (2012) provided a detailed overview about the historical background of the mass migration, and Connor (2007) focused on the integration of forced migrants into postwar Germany. An alternative source from the Eastern European perspective is Eberhardt (2011), who reported the number of Germans who were displaced from Poland and territories incorporated into Poland. Finally, Lüttinger (1986) and Reichling (1958) summarized important aggregate statistics on forced migrants in Germany that were drawn from censuses of the postwar period.
A small share of displaced persons who were initially located in the Soviet occupation zone eventually arrived in West Germany. This type of east-west sorting amounted to a total of approximately 500,000 individuals who moved from East Germany (Soviet occupation zone including Berlin) to West Germany from 1950 to 1955 (Reichling 1958:357–358).
The expulsion was not entirely universal regarding some ethnic Germans who remained in Eastern Europe for several decades (Aussiedler/Spätaussiedler), also including German minorities (e.g., in Upper Silesia or Transylvania).
For example, some Catholic Sudeten Germans were placed in Protestant North Hesse and Franconia, and many Protestant migrants were settled to Catholic areas in Lower Bavaria (Connor 2007).
Precisely, the covered period is from December 1993 to November 2013. Because of the nature of the administrative process, each annual wave of pension shortfall records samples all cases of death from January to November of a given calendar year and adds those cases documented for December of the preceding year. For example, the 2012 wave includes all deaths that were documented from December 2011 to November 2012.
For a detailed description of the sampling design (in German) and a codebook, see Forschungsdatenzentrum Deutsche Rentenversicherung (2017).
We obtain these shares from relating the total number of deaths reported in the official mortality statistics for Germany (Federal Statistical Office 2016) to the number of deaths in the pension shortfall records that we use as primary data source.
Only 4 % of male cases of death documented in official mortality statistics are not covered by the pension shortfall. Probably the most plausible explanation is that a considerable share of these men worked as civil servants from the beginning of their employment biography. Pensions of civil servants are tax-financed and handled separately from the public (pay-as-you-go) system.
We use an electronically preprocessed version of the original print by Besser (2008).
The cohort distribution is depicted in Fig. B.1 (online appendix), showing the percentage of each cohort in the sample.
For example, the population at risk also includes persons who out-migrated. They need only to have accumulated pension claims in Germany at some time—for example, by employment that is subject to social security contributions or by periods of child-rearing. In this case, a persons’ death is documented in the shortfall records because pension payments are terminated.
Comparing earnings biographies is more feasible if they are completed by the date of retirement because this rules out false comparisons of earnings biographies that differ only because of age differences.
This is necessary to make these population counts comparable with the pension shortfall records that document only participants of the public pension system.
The mortality rates are higher among men (95 % for displaced and 75 % for nondisplaced men) than among women (74 % for displaced and 50 % for nondisplaced women). The calculation assumes that everyone has deceased in the observed birth cohorts by the end of the observation period. The calculations are available from the authors upon request.
Because of the small number of observations at the upper margin of the age-at-death distribution, we use 40 duration indicator variables for the ages a = 68, . . . , 107. Estimating δa for ages above 107 is difficult because we observe very few persons to survive this age (e.g., the maximum age reached by one single person in the sample is 111).
To further support that interregional variation reflected by the displacement indicator matters, we implemented a multilevel mixed-effects linear regression with regions of origin (Eastern Europe vs. West Germany) at the higher level and individuals at the lower level. From this exercise, we infer that the share of variation in mortality explained at the regional level ranges from 2.7 % to 7.7 % (men) and 0 % to 7 % (women); detailed results are available from the authors upon request. Although the random effects parameters indicate only a small variance contribution at the regional level for women after birth cohort dummy variables (close to 0) are included, the overall results indicate that interregional variation matters. A large fraction of variation is explained at the individual level; but given that mortality is determined by various—presumably individually driven—factors, the estimated variance shares at the regional level can still be considered as substantial.
Lifetime earnings may represent a “bad control” (for a discussion, see Angrist and Pischke 2009) because they may themselves be an outcome and thus be affected by displacement. In particular, comparing displaced with nondisplaced individuals at a given value of lifetime earnings may differ by some unobserved characteristics that compensate the initial disadvantage in the earnings potential of the displaced. This problem is arguably a minor one because including earnings in the baseline specifications does not considerably change the estimated hazard ratios (see Table 2, comparing columns 1 and 2 for men and columns 4 and 5 for women). Both of these variables covary positively with the mortality rate and the displacement indicator given that prewar birth and death rates were considerably higher in Eastern Europe than in West Germany (see Fig. 2).
Arguably, this would be the case because the displaced person ended up earning the same despite of being displaced.
A total of 1,937,297 displaced persons were registered in Bavaria by 1950. In relative terms, this amounted to 21 % of the Bavarian population at that time (Reichling 1958).