A family affair: Evidence of chain migration during the mass emigration from the county of Halland in Sweden to the United States in the 1890s

Abstract

This paper examines the influence of individual and household factors on an individual’s propensity to emigrate from Halland, a region in south-west Sweden, to the United States during the era of mass migration in the late nineteenth century. The study has a case–control design, using individual-level longitudinal data for a group of emigrants (cases) and a group of non-emigrants (controls). Results indicate the importance of a family’s emigration history; individuals whose relatives had previously moved to the United States were more likely to emigrate themselves. In addition, the results also show how this impact varied between groups and how other factors relating to the individual’s life situation affected the migration decision. Thus, this paper shows how chain migration and migration networks play important roles during times of mass migration.

Keywords:

• chain migration
• family network
• migrant network
• migration decisions
• mass migration
• international migration
• case–control study
• longitudinal data
• Sweden
• late nineteenth century

Introduction

Migration is an unpredictable demographic process of great societal importance that has been thoroughly studied in various ways. Its complexity is reflected in the substantial number of well-established theories on the determinants and features of geographic mobility. Chain migration is a consistent feature of present-day migration and is defined as a social process by which potential migrants, within a family or in a community, are influenced by previously migrated family members or friends and eventually follow them to a new place of residence.

Just like present-day migration, large-scale emigration from Europe to North America in the nineteenth and early twentieth centuries had a comprehensive impact on society. One significant advantage of studying nineteenth-century migration is the lack of regulations in the United States (US) at that time regarding immigrants from countries such as Sweden. This facilitated emigration, since a majority of Swedish emigrants stood no risk of being refused entry to the US. This is clearly different from the current situation, where US immigration regulations are much more extensive and decisive regarding who can or cannot immigrate. Studying nineteenth-century migration can thus help us to understand the determining factors for emigration at the individual and household levels.

Although theories on the determinants of migration are essentially at the micro level, empirical studies of overseas migration in the nineteenth century have typically been based on macro-level data (e.g., O’Rourke and Williamson 1999; Hatton and Williamson 2005; Bohlin and Eurenius 2010) rather than micro-level (i.e., individual and household) data. Although over 1 million people emigrated from Sweden during the mass migration era, this was still rather few in terms of the overall population. Sweden’s average yearly emigration rate during the period 1880–1910 was about 6 per 1,000 inhabitants (Bidrag till Sveriges officiella statistik 1890; Sundbärg 1910). A micro-level study of the importance of possible determinants of such a rare event would thus require a very large ‘at risk’ population with most study designs; hence there are few such studies. The literature is full of best guesses based on theory, anecdotal evidence, and studies of the features of those who moved (without comparing them with those who did not); however, we lack knowledge about how individual characteristics and household context actually influenced emigration risk.

This paper aims to fill this gap by examining the importance of individual characteristics and household structure in individuals’ decisions to emigrate in the late nineteenth century, using micro-level data on individuals and families. Most importantly, I investigate chain migration by testing whether having any close relatives who had previously emigrated increased the probability of a person emigrating themselves. To overcome the problem of needing a very large ‘at risk’ population, I used a case–control design for the data. This method is common within epidemiological studies, but rarely used in social sciences, and is a novel approach to studying historical migration. I created a longitudinal data set with individual- and household-level data for people who emigrated in 1891 from the county of Halland in south-west Sweden to the US and for a control group of people who remained. Specifically, I used a random sample of 250 emigrants or ‘movers’ (cases) and 500 ‘stayers’ (controls), ranging in age from 14 to 34 years in 1890, and compared them across a number of potential determinants. Of particular interest is family emigration history, but I also included characteristics of individuals and their parental households, as potential confounders.

Differences in economic conditions between Sweden and the US were a major cause of emigration in the study period. Factors such as real wage differentials and better employment opportunities in the New World had a strong impact, and variations in business cycles were reflected in emigration rates (Bohlin and Eurenius 2010). Given such general drivers of emigration, the focus of this study is on the influence of family and household context on individual migration decisions.

The key result is that family emigration history strongly impacted the decision to emigrate, and this was especially important to young people. The propensity to emigrate increased significantly with increasing numbers of previous emigrants in the family. In addition, the results show that other factors related to an individual’s life situation had an impact on the decision to emigrate and affected different groups in different ways.

Theoretical framework and previous research

Chain migration is a well-known phenomenon in migration research. It can be defined as a process by which prospective migrants learn of opportunities, are provided with transportation, and have initial accommodation and employment arranged by means of social relationships with previous migrants (MacDonald and MacDonald 1964). Valuable information and various forms of support are transferred from those who have already migrated, facilitating the migration decisions of relatives and friends back home and initiating the chain migration process. Throughout history, chain migration has often played an important and evident role in the development of migration patterns. Numerous studies on more recent migration streams have shown the presence and significance of chain migration in different regions: Venezuela (Levy and Wadycki 1973), Hungary (Epstein and Gang 2006), immigration to the US (Borjas and Bronars 1991), and Mexican migration to the US (Massey et al. 1994; Palloni et al. 2001).

Theoretical framework

Establishing the determinants of out-migration (i.e., what makes individuals leave one place of residence for another) is complex (Klabunde et al. 2017; Kley 2017) and has long been a major theme in migration studies. From a neoclassical, microeconomic point of view, the decision to move or not is a rational choice by an individual (Todaro 1969; Borjas 1989). It is based on a calculation of the best option once the benefits (e.g., higher earnings) and costs (e.g., financial costs) of moving have been evaluated and compared. If the expected net gain from migration is positive, the individual decides to move; if negative, they decide to stay. In this way, migration can be seen as an investment in higher net earnings in the future. Factors influencing the decision to move are based on an individual’s human capital, including age, schooling, skills, marital status, and experience (Sjaastad 1962), but also on differences in wages, expected future income, and the probability of future employment (Todaro 1969; Hatton and Williamson 2005). Thus, neoclassical theory focuses primarily on individual and labour market conditions, explaining migration as mainly a result of individual behaviour, but not providing a complete explanation of the concept of chain migration.

This neoclassical approach can be extended to include a larger group of people in the decision-making process: typically families or households, the intermediate meso level between the individual micro and the societal macro levels (e.g., Harbison 1981; Faist 1997). Meso factors studied in this literature include household size, demographic structure, life cycle stage, and the family’s socio-economic aspirations. The family acts as the actual decision-making unit to allocate its resources in the most productive way. This is crucial in the ‘new economics of migration’, which emphasizes the aim of migration as not only maximizing net gain but also minimizing risks (Stark and Bloom 1985). According to this theory, households have opportunities—which individuals do not have—to allocate resources to various sources of income and thereby secure their livelihoods. By allowing some family members to move and some to remain at home, the household can access markets with different economic conditions, such as financial markets and social insurance systems not available in the home country. Thus, migration can be seen as a form of social insurance for the household (Findley 1987; Massey 1990; Borjas and Bronars 1991). This extension of the neoclassical approach partly explains the presence of chain migration. However, to proceed further, we need to broaden the theoretical framework even more.

A well-known dilemma regarding the neoclassical model is the discrepancy between theory and reality. In many places, migration is absent, despite economic inequalities between regions or countries, while elsewhere migration continues, despite equalization of economic disparities between areas (e.g., Moretti 1999). Another problem regarding the rational choice approach to migration is that it does not explain how structural opportunities are translated to individual action (Faist 1997). Individuals make their own decisions but are, at the same time, part of a social context at a meso level and influenced by structural, economic, cultural, and political circumstances at a macro level (Massey 1990; Piché and Dutreuilh 2013). To provide an adequate explanation of this link within the migration decision process, we need to complement the neoclassical theoretical approach with concepts that also include social relations (Massey 1990; Radu 2008).

The concept of social capital has become increasingly important in migration research (Massey et al. 2008). This theoretical approach focuses on the relationships between people and the resources created as a result. Social capital can be described as the intangible resources within families or communities that help promote social development. An important characteristic is its convertibility to other forms of capital (e.g., to financial capital through remittances from previous migrants) (Palloni et al. 2001). Furthermore, as shown by Coleman (1988), social capital within families facilitates the increase in human capital of the younger generation. Thus, social capital is created in social relations at the meso level and serves as the link between societal conditions and political, economic, and cultural structures at the macro level and individual migration decision-making at the micro level (Faist 1997).

Considering social capital as a resource to enable action also emphasizes its importance in migration processes. Migrant networks constitute a form of social structure in which social capital is created and accumulated, and have been defined as ‘sets of interpersonal ties that link migrants, former migrants and non-migrants in origin and destination areas by ties of kinship, friendship and shared community origin’ (Massey 1990, p. 7). Such ties promote further migration by lowering costs and risks, thereby increasing the expected net gain of the move. Migration costs include financial and psychological costs, but also opportunity costs (i.e., earnings forgone while moving and establishing a new life at the destination). Relatives and friends who have already emigrated can help reduce these costs and support newcomers in the integration phase by providing temporary accommodation or contacts with authorities and potential employers. Migrant networks within families or kinship groups have proven to be particularly strong, and research has shown they play a key role in migration development. Numerous studies on recent migration streams have shown the influence of migration networks (e.g., Choldin 1973; Boyd 1989; Palloni et al. 2001; Haug 2008; see also Krissman 2005).

To summarize, this study investigates chain migration within families, taking into account the economic and social contexts within which people live; its theoretical starting point thus aligns with the traditions of both meso-level theory and migration network theory.

Previous research on chain migration

Studies on chain migration have often been based on aggregate data, and have focused on migrant stock in the receiving countries or cumulative numbers of emigrants from the regions of origin (e.g., MacDonald and MacDonald 1964; Carlsson 1976; Dunlevy and Gemery 1977; Tilly 1978; Gjerde 1991; Moretti 1999; Hatton and Williamson 2005). Chain migration during the nineteenth century has also been explored indirectly, using the amount of remittance sent back to the home countries (Magee and Thompson 2006). Others have considered it in terms of the approximately 50 per cent of Swedish emigrants during the 1880s who travelled with tickets paid for by previous migrants (Tilly 1978; O’Rourke and Williamson 1999).

Micro-level studies on chain migration in the same period are few. However, in a study on German emigration to the US during the mid-nineteenth century, Wegge (1998) tested the presence of chain migration. The data were derived from more than 50,000 emigrants from the German principality of Hesse–Cassel and provided information on individual factors such as age, gender, occupation, and the amount of cash emigrants brought. To determine the presence of chain migration, Wegge sorted the emigrants into groups of networked and non-networked individuals, defining a network as emigrants from the same village with the same surname. The analysis showed that previous migrants had a significant and positive influence on the migration decisions of later migrants from the same village with the same surname, leading to a rise in migration rates. The results also showed one clear effect of being part of a network: ‘networked’ emigrants brought less cash, which indicates they could rely on some support from previous migrants. In a follow-up study, Wegge (2008) investigated the structure and development of chain migration, showing the importance and persistence of migration networks.

Bras and Neven (2007) investigated the influence of household structure on internal migration patterns for women during the early nineteenth and early twentieth centuries. Their study used longitudinal microdata from population registers in two rural areas in Belgium and the Netherlands, and explored to what extent different aspects of household structure impacted women’s migration patterns. Above all, the analysis showed the impact of siblings, especially sisters, on women’s migration behaviour. Having sisters who had previously migrated not only increased the likelihood of moving but also affected the destination, indicating the clear presence of chain migration.

Still, to get an even better understanding of the migration decision process and the mechanisms of chain migration during the nineteenth century, there is a need for further studies at the micro level, hence this study.

Context, data, design, and model

Context

It has been estimated that Sweden lost just over 1 million people to emigration during the late nineteenth and early twentieth centuries—approximately one-fifth of its average total population. An overwhelming proportion (around 90 per cent) of Swedish emigrants moved to the US (Carlsson 1976). Sweden was experiencing ongoing societal and demographic change, including a growing population of young adults and increasing competition over land and jobs in the countryside. For many people, emigration thus became a chance for a better future in the New World. Swedish emigration rates fluctuated considerably during this period, as did those in other western European countries, corresponding to long swings in the Atlantic economy. Downswings in the European economy corresponded to upswings in the US economy, which increased European emigration rates, and vice versa. The most intensive emigration period in Sweden was 1879–93, when approximately 500,000 Swedes emigrated (Carlsson 1976; Bohlin and Eurenius 2010).

Approximately 80 per cent of the Swedish population still lived in rural areas during this era (Bidrag till Sveriges officiella statistik 1890). The dominant type of agriculture in Halland was small farms cultivated by the owner and other family members. A rise in natural increase in this region during the mid-nineteenth century resulted in rapid population growth in the decades preceding the emigration era. Children were often considered as an economic asset contributing to the labour force in small farms, and large families were fairly common. Several scholars have stressed the prevailing differences in land ownership structure and demographic behaviour between Swedish regions (Wohlin 1909; Sundbärg 1910; Winberg 1975; Lundh 2013). Both demographic change, resulting in increasing competition over income possibilities, and the still tiny industrial sector in Halland, meaning few job opportunities outside agriculture, spurred emigration among young adults. As mentioned earlier, the average yearly emigration rate for Sweden during 1880–1910 was about 6 per 1,000 inhabitants, while the corresponding figure for Halland was around 10 per 1,000, making Halland the Swedish county with the highest emigration rate (Bidrag till Sveriges officiella statistik 1890; Sundbärg 1910).

Data and sources

Sample and collection of information. The study population includes 42,382 men and women born in 1856–76, who were living in the countryside in Halland in 1890 (Swedish 1890 Census; see ‘Data sources’). I chose a rural study population because a vast majority of the population in Halland (and Sweden) lived in the countryside and, consequently, just over 80 per cent of emigrants from Halland came from rural areas, as did about 80 per cent of Swedish emigrants (Bidrag till Sveriges officiella statistik 1890).

The starting year of this study is 1890, with movers emigrating to the US in 1891 and stayers not emigrating in 1891. The choice of the 1890 Census as a starting point resulted mainly from factors related to the data sources. First, the 1884 Emigration Ordinance made data from church records more reliable. This ordinance prohibited emigration agents employed by Atlantic shipping companies from conveying any travellers abroad if they could not present proper migration certificates for all emigrants to the police authority. Another reason is that older church records tend to be less complete. Furthermore, the age structure among emigrants was virtually constant during the 1879–93 emigration boom, with 15–35 year-olds representing approximately 70 per cent of all Swedish emigrants during the period (Bidrag till Sveriges officiella statistik 1880–1893). Altogether, this made the Census year 1890 suitable as the starting point.

To study chain migration and the effect of family emigration history, longitudinal data on families and the individuals within them are needed, with information on both movers and stayers. Two samples were created: 250 movers (the cases) and 500 stayers (the controls). The sample of movers was created by first identifying all emigrants from the Halland countryside in 1891. This was done by searching parish migration registers for 1891 in each of the 88 countryside parishes, resulting in 1,500 emigrants. I then selected all emigrants born in 1856–76 (i.e., aged 15–35 in 1891) with the US as their destination, which left 1,150 individuals (just over 75 per cent of all emigrants). Finally, a random sample of 250 individuals was drawn from this population by picking individuals at a fixed interval, giving a random sample of individual movers for whom the potential drivers of emigration could be studied. Note that the phenomenon of emigrants travelling alongside family members cannot be studied using this sample; a different data set and study design would be required. (About 10 per cent of the emigrants in this study travelled with another family member, but in my sample such family members were not included due to the fixed interval sampling procedure.)

The sampling of stayers was conducted in a similar way. First, a list of the total birth cohort for 1856–76 in Halland was extracted from the 1890 Census. Then, a random sample of 500 individuals was selected by a fixed interval calculated to attain the intended sample size. Some sampled individuals had to be excluded and replaced by the succeeding individual in the list, for example, non-rural individuals and those who eventually emigrated in 1891. Both samples include individuals who may have previously moved within Sweden, because mobility among young people in the countryside was significant during the late nineteenth century, as changing employment often meant changing place of residence.

To determine what influenced the decision whether to emigrate, a substantial amount of longitudinal information was extracted from church records for every individual (both cases and controls). The goal was to track each individual back to their place of birth and date of their parents’ marriage, in an attempt to reconstruct the family of origin. This was a very time-consuming process and young people’s frequent moves sometimes made the tracing rather complicated. During this process, information on close family members (siblings, parents, or spouses) was also extracted. However, it was not possible to follow all family members, especially all siblings, throughout their lives.

Just under 80 per cent of the 1,500 parents (of the 750 sampled individuals) were still alive in 1890, with a known place of residence in Sweden, while just over 18 per cent had died. Approximately 2 per cent had previously emigrated, which means that complete information for this group was lacking. The total number of siblings found in the original families was 3,874, of which I was able to trace 2,663 siblings (approximately 70 per cent). This provided information on whether they were deceased or alive, and whether they were still living in Sweden in 1890 or had emigrated. In the analysis, I assumed the non-traced siblings were all alive in 1890, because all siblings who died while living at home were recorded and because most of the mortality below age 40 (thus covering most siblings in the study) occurred in childhood. I also assumed that no non-traced siblings had previously emigrated. This is unlikely to be completely accurate but is still the most reasonable assumption (because emigration was a rare event in statistical terms). The end of the ‘Results and discussion’ subsection discusses results where I assumed instead that all non-traced siblings had previously emigrated, as a robustness check.

Sources

Information on all 750 individuals and their families was found in the catechetical examination records (husförhörslängderna; see ‘Data sources’), in which the parish minister listed household members each year. The primary purpose of this was to maintain the population register and also to examine the biblical knowledge and reading skills of each household member. The records contained lists with basic information on all household members and information on changes in the household since the previous record, such as births, deaths, and migration. The minister recorded when each move occurred and where the individual moved to and from. In a specific migration register (flyttlängderna), they also recorded all moves into and out of the parish, including both domestic and overseas migration, and every person who intended to move had to request a change of address certificate. We can assume a certain underestimation of emigration rates in these migration registers, as ministers could not always separate emigration and domestic migration, although the 1884 Emigration Ordinance had made the data more reliable.

The 1890 Census was based on the information in catechetical examination records. It contains data on every individual registered in each parish on 31 December 1890. Information is available for every household and all its members, which, in addition to the family, includes any servants, lodgers, or members of the (grand)parental generation living in the household.

Method

The purpose of this study is to identify and evaluate factors that may have had an impact on the propensity of individuals to emigrate, by comparing groups of individuals that emigrated (the cases) with groups that did not (the controls). This case–control method is common in epidemiological studies where, for example, individuals with a disease are compared with those without, regarding earlier exposure to different factors. The method is much less common in social science, but is highly useful for studying rare events (see, e.g., Van Poppel 1997). Emigration from Halland was a very rare event; on average, approximately 1 per cent of the population emigrated each year between 1880 and 1910. Also, all sampled individuals in the study were identified at the start as either emigrants or non-emigrants, and information about individuals was then collected by tracing them back in time, making the study retrospective. Finally, the sample size was somewhat limited due to the time-consuming process of tracing and extracting information. Together, these circumstances made the implementation of a case–control study suitable (Hennekens and Buring 1987; Woodward 1999).

Case–control studies tend to be more susceptible to bias than other epidemiological analytical designs (Schulz and Grimes 2002). Common sources of bias are the sampling procedure for controls and possible differences in the quality of information from the different samples. In this study, the cases and the controls were sampled in the same random way from the same study population and thus the controls experienced the same risk of exposure as the cases. Moreover, the information for both cases and controls was collected in the same way from the same sources, so any risk of bias related to this concern was avoided.

In case–control studies, the odds ratio is used as an estimate of the relative risk of an event occurring. This is because the individuals in a study are sampled on the basis of their case–control status, which leads to proportions of cases and controls in the sample that are inconsistent with their actual proportions in the study population. However, as long as the event in the study is rare, the odds ratio is still a good approximation of relative risk (Woodward 1999; Grimes and Schulz 2008). Thus, to measure the extent to which there is an association between emigration and the different independent variables, odds ratios are used (see Appendix). To estimate them, I used logistic regressions, and in the paper present significance levels based on robust standard errors for each coefficient. I also present the significance of each variable group based on likelihood ratio tests, excluding each category separately (drop1 test).

Analytical model

Based on migration theory and previous research, I used a model with three types of migration determinant for the multivariate analysis. The first is a family emigration history variable capturing chain migration, the focus of this study. The second and third include variables indicating household and individual characteristics, respectively. All covariates are categorical. The outcome variable is binary, indicating emigration in 1891 (or not).

Family emigration history

I estimated the chain migration effect directly at the micro level on the individual’s likelihood of emigration, by regressing the number of close family members (siblings, parents, or spouses) who had previously emigrated. The variable consists of three categories: those with no family history of previous emigration, those with one close family member who had previously emigrated, and those with two or more close family members who had previously emigrated. I expect the existence of previous migrants in the family to have increased the chance of emigrating.

Household characteristics

These variables reflect the structure and socio-economic status (SES) of the parental household, as parents’ social status and the individual’s position in the family could set crucial boundaries for children’s futures.

Two variables in this group relate to the size and structure of the family of origin. The first reflects the effect of number of siblings alive in 1890, with two categories: ‘fewer than three’ or ‘three or more’ siblings alive in 1890. It could be argued that only siblings actually living in the household in 1890 should be included, because only these siblings would have influenced the individual’s circumstances. However, siblings living in other households could also have influenced such matters as future inheritance or may have been involved in other ways in family affairs. Moreover, it is difficult to determine how close to the original home the siblings lived. I expect the propensity to emigrate to have increased with the number of siblings alive in 1890.

The second household variable reflects the impact of not having both parents alive in 1890. This could affect the amount of support available when making the crucial decision of whether to emigrate. Having both parents alive could mean possible economic contributions and could also eliminate the potential pressure to take care of a widowed parent or siblings still living at home. By contrast, having only one parent still alive could mean that an inheritance is expected in the not too distant future. Overall, I expect having both parents alive in 1890 to have increased the propensity to emigrate.

A third household variable is the father’s occupation at the individual’s birth, which is intended to capture the impact of the parental household’s SES on the individual’s likelihood of emigration later in life. Occupational titles were coded using the HISCO and HISCLASS system (van Leeuwen et al. 2002; van Leeuwen and Maas 2011). To achieve sufficient group sizes, occupations were merged into three categories: ‘skilled’ occupations, including blacksmiths, carpenters, teachers, and tailors (HISCLASS 1–7); ‘farmers’ and landholders, including peasants, freeholders, and tenants (HISCLASS 8); and unskilled ‘workers’ and the landless, including crofters, cottagers, maids, farmhands, and farmworkers (HISCLASS 9–12). I expect individuals from families with a lower SES to have been more likely to emigrate.

Individual characteristics

These variables reflect the individual characteristics of potential emigrants, such as their gender. In general, emigration rates during the study period were higher for men than for women (Donato and Gabaccia 2015). This was particularly evident in Sweden when emigration rates were high, so one possible explanation is that women did not react as swiftly as men to fluctuations in business cycles. A typical Swedish female worked in the household sector (e.g., as a maid), which seems to have been less vulnerable to business cycle fluctuations (Carlsson 1976; Bohlin and Eurenius 2010). Consequently, I expect men to have been more likely to emigrate than women.

The second individual variable is age; just over 75 per cent of all emigrants from Halland in 1891 were aged 15–35. This corresponds well with figures for the total number of emigrants from Sweden and is why only individuals in this age range were included in the study. To investigate any differences in propensity to emigrate within this age range, the sample was divided into four groups, based on individuals’ ages in 1890: 14–18, 19–23, 24–28, and 29–34. I expect younger individuals to have been more prone to emigrate than older individuals.

The third variable is marital status. Being married likely indicated more stable employment opportunities or access to land, which should have had a curbing effect on the propensity to emigrate. It also meant the decision to emigrate was likely a joint decision and more difficult to make than if one was unmarried. Thus, I expect the propensity to emigrate to have been lower among married people.

Ideally occupation would be included as a covariate, as employment and livelihood are clearly crucial in migration decisions. However, for unmarried people in the countryside at that time, the variation in occupational titles was rather limited, and 96 per cent of individuals in the two youngest age groups were unmarried. The vast majority were maids and farmworkers—not really permanent occupations—while more established occupational titles (e.g., peasant or crofter) were reserved for married people. Therefore, as the available information does not provide sufficient description of individuals’ occupations, the variable was dropped from the analysis.

Analysis

Descriptive statistics

Table 1 presents the descriptive statistics for the sample and shows how the groups of stayers and movers differ from each other. It also reports the results from chi² tests, which indicate whether the differences in proportions are statistically significant. First, there is a clear difference between the two groups in terms of family emigration history. The share of those with siblings or parents who had previously emigrated is considerably higher among movers than among stayers. Among men, approximately 38 per cent of movers had a family history of emigration compared with 19 per cent of stayers. This pattern is the same for women: approximately 48 per cent of the movers had family members who had previously emigrated compared with just 24 per cent of the stayers. The most significant difference is for those with two or more family members who had previously emigrated. I remind the reader that the samples of movers and stayers are random and only conditioned on age. The design does not completely remove the possibility of systematic differences between cases and controls, but makes it most likely that these differences in proportions are the result of a real association between family emigration history and the decision to emigrate.

Table 1  Descriptive statistics on the analysed samples: stayers and movers in the county of Halland, Sweden, 1890

	Males			Females
	Stayers		Movers	Stayers		Movers
	Percentage		Percentage	Percentage		Percentage
Family emigration history		***			***
No	81.1		62.4	76.2		52.3
Yes, one person	14.1		19.7	17.0		27.3
Yes, two or more	4.8		17.9	6.8		20.4
Household variables
Siblings alive in 1890		***			***
Fewer than three	29.6		17.3	25.9		12.5
Three or more	70.4		82.7	74.1		87.5
Father’s occupation		n.s.			n.s.
Farmer	50.0		49.4	65.0		54.6
Skilled	8.3		6.8	7.8		9.0
Worker	41.7		43.8	27.2		36.4
Both parents alive		n.s.			n.s.
Yes	62.6		67.3	65.0		67.1
No	37.4		32.7	35.0		32.9
Individual variables
Age		***			***
14–18	35.9		31.5	27.6		42.0
19–23	14.0		35.2	26.5		33.0
24–28	18.5		21.0	25.5		12.5
29–34	31.6		12.3	20.4		12.5
Marital status		**			***
Married	19.9		11.1	25.2		8.0
Unmarried	80.1		88.9	74.8		92.0
N	206		162	294		88

Note: The statistical significance of the differences in proportions between stayers and movers for each variable was tested using a chi² test.

n.s. (non-significant) p ≥ 0.10, *p < 0.10, **p < 0.05, ***p < 0.01.

Source: Author’s calculations based on data from Swedish 1890 Census and Swedish catechetical records (National Archive of Sweden, Riksarkivet).

Second, looking at the household variables, the two groups differ in some respects; for example, 83 per cent of the male movers had three or more siblings alive in 1890, compared with 70 per cent of male stayers. Among females, the corresponding shares were 88 per cent of movers and 74 per cent of stayers. As to father’s occupation, there is no substantial difference between movers and stayers among males. For females, the proportion of ‘farmer’ origin was larger among stayers (65 per cent, compared with 55 per cent among movers), while the proportion of ‘worker’ origin was larger among movers (36 per cent, compared with 27 per cent among stayers). Having both parents alive was almost equally common within the whole sample, with the proportion among both male and female stayers and movers amounting to about 65 per cent.

Turning to the individual variables, there are some substantial differences between stayers and movers. The samples were chosen at random among those aged 14–34 in 1890, but were not stratified within this age range, therefore differences may exist between the samples in the age structure within this range. Among the movers, there is a clear majority in the two youngest groups: 67 per cent of male movers were in the two youngest age groups, 14–18 and 19–23, compared with 75 per cent of female movers. Among female stayers, age is quite evenly distributed over the four age groups, with 20–28 per cent of the sample in each age group, while it is more varied among the male stayers. Only 14 per cent of male stayers were in the 19–23 age group and just above 18 per cent in the 24–28 age group. These figures correspond with official population statistics for the study period, which show the proportion of men aged 20–30 in Halland to have been among the smallest in Sweden (Bidrag till Sveriges officiella statistik 1890). This reflects the over-representation of men and young adults among migrants and the fact that Halland was a county of both internal and external net out-migration (Sundbärg 1910). Finally, being married is not as common among movers as stayers, with the largest difference among women: only 8 per cent of emigrating women were married compared with just over 25 per cent of those who stayed. We would, of course, expect a much lower share of marriage among the movers, given their lower average age. The average age at marriage in rural Halland communities in 1890 was 30.5 for men and 28.5 for women (Lundh 2013).

Results and discussion

To estimate the effect of chain migration on an individual’s propensity to emigrate in 1891, I performed a set of logistic regressions for males and females separately. Table 2 shows the results of testing the family emigration history variable alone. Tables 3–5 show estimates of the effect from chain migration adjusted for possible confounders. Table 3 shows results for the whole sample, while Tables 4 and 5 include individuals only in the younger or older age groups, respectively.

Table 2  The effect of family emigration history on the likelihood of emigration from Halland, Sweden, in 1891

	Males	Females
	Odds ratio	Odds ratio
Family emigration history	(***)	(***)
No (Ref.)	1.00	1.00
Yes, one person	1.82**	2.34***
Yes, two or more	4.80***	4.38***
N	368	382
Likelihood ratio chi^2, p-value	0.000	0.000

Notes: Robust standard errors were used for the coefficients. The joint statistical significance of both indicators for migration history was estimated using a likelihood ratio test after dropping both the ‘Yes, one person’ and ‘Yes, two or more’ indicator variables (drop1 test). The outcome is presented in parentheses.

n.s. (non-significant) p ≥ 0.10, *p < 0.10, **p < 0.05, ***p < 0.01. Ref. is the reference category.

Source: As for Table 1.

Table 3 The effect of family emigration history and other individual and household characteristics on the likelihood of emigration from Halland, Sweden, in 1891

	Males	Females
	Odds ratio	Odds ratio
Family emigration history	(***)	(***)
No (Ref.)	1.00	1.00
Yes, one person	1.82*	2.74***
Yes, two or more	4.76***	6.13***
Household variables
Siblings alive in 1890	(*)	(**)
Fewer than three (Ref.)	1.00	1.00
Three or more	1.60*	2.15**
Father’s occupation	(n.s.)	(**)
Farmer (Ref.)	1.00	1.00
Skilled	0.74	1.84
Worker	1.49	2.11**
Both parents alive	(n.s.)	(n.s.)
No (Ref.)	1.00	1.00
Yes	0.91	0.78
Individual variables
Age	(***)	(*)
14–18 (Ref.)	1.00	1.00
19–23	3.16***	1.17
24–28	1.51	0.43**
29–34	0.46*	0.71
Marital status	(n.s.)	(**)
Unmarried (Ref.)	1.00	1.00
Married	1.08	0.30**
Constant	0.32***	0.13***
N	368	382
Likelihood ratio chi^2, p-value	0.000	0.000

Notes: Robust standard errors were used for the coefficients. The joint statistical significance of each variable group was estimated using a likelihood ratio test after dropping each group separately (drop1 test). The outcome is presented in parentheses.

n.s. (non-significant) p ≥ 0.10, *p < 0.10, **p < 0.05, ***p < 0.01. Ref. is the reference category.

Source: As for Table 1.

Table 4 The effect of family emigration history and other individual^a and household characteristics on the likelihood of emigration from Halland, Sweden, in 1891: men and women aged 14–23

	Males	Females
	Odds ratio	Odds ratio
Family emigration history	(***)	(***)
No (Ref.)	1.00	1.00
Yes, one person	2.26*	2.41**
Yes, two or more	9.13***	10.43***
Household variables
Siblings alive in 1890	(n.s.)	(*)
Fewer than three (Ref.)	1.00	1.00
Three or more	1.31	2.12*
Father’s occupation	(n.s.)	(**)
Farmer (Ref.)	1.00	1.00
Skilled	0.93	2.60*
Worker	1.38	2.65***
Both parents alive	(n.s.)	(n.s.)
No (Ref.)	1.00	1.00
Yes	1.13	0.65
Individual variables
Age	(***)	(n.s.)
14–18 (Ref.)	1.00	1.00
19–23	3.27***	1.15
Constant	0.29**	0.13***
N	211	225
Likelihood ratio chi^2, p-value	0.000	0.000

^aMarital status is excluded from this table due to the small number of individuals married in this age group.

Notes: Robust standard errors were used for the coefficients. The joint statistical significance of each variable group was estimated using a likelihood ratio test after dropping each group separately (drop1 test). The outcome is presented in parentheses.

n.s. (non-significant) p ≥ 0.10, *p < 0.10, **p < 0.05, ***p < 0.01. Ref. is the reference category.

Source: As for Table 1.

Table 5 The effect of family emigration history and other individual and household characteristics on the likelihood of emigration from Halland, Sweden, in 1891: men and women aged 24–34

	Males	Females
	Odds ratio	Odds ratio
Family emigration history	(n.s.)	(**)
No (Ref.)	1.00	1.00
Yes, one person	1.30	3.52**
Yes, two or more	1.50	3.20**
Household variables
Siblings alive in 1890	(*)	(n.s.)
Fewer than three (Ref.)	1.00	1.00
Three or more	2.08*	2.22
Father’s occupation	(n.s.)	(n.s.)
Farmer (Ref.)	1.00	1.00
Skilled	0.37	0.73
Worker	1.77	1.10
Both parents alive	(n.s.)	(n.s.)
No (Ref.)	1.00	1.00
Yes	0.72	1.29
Individual variables
Age	(***)	(n.s.)
24–28	3.10***	0.60
29–34 (Ref.)	1.00	1.00
Marital status	(n.s.)	(*)
Unmarried (Ref.)	1.00	1.00
Married	0.86	0.35*
Constant	0.16***	0.08***
N	157	157
Likelihood ratio chi^2, p-value	0.021	0.092

Notes: Robust standard errors were used for the coefficients. The joint statistical significance of each variable group was estimated using a likelihood ratio test after dropping each group separately (drop1 test). The outcome is presented in parentheses.

n.s. (non-significant) p ≥ 0.10, *p < 0.10, **p < 0.05, ***p < 0.01. Ref. is the reference category.

Source: As for Table 1.

A key finding is the clear evidence for chain migration among both men and women in Halland emigrating to the US in 1891. Family emigration history had a strong impact on the propensity to emigrate (Table 2). It is also clear that the odds of emigrating became higher with increasing numbers of previous emigrants in the family. The results for the chain migration variable were almost unchanged for males but grew somewhat stronger for females when adjusted for possible confounders (Table 3). This means that the effect of family emigration history was not greatly affected by the other factors included in the model.

Even if the coefficients differed between men and women, the effect of family emigration history was not significantly different between the sexes (see interaction effects in Tables A1 and A2 in the supplementary material). This also applies to the results where older and younger individuals were separated (Tables A3 and A4 in the supplementary material).

It is quite clear that a family history of emigration had a strong impact on the decision to migrate. A close family member who had already emigrated had a significant influence on the presumptive emigrant back home, by sending frequent financial remittances, or serving as a link to a new social context and new community. In this manner, the newcomer could at least hope to get a smooth start in the new homeland, something that likely facilitated the emigration decision by helping to relieve psychological stress and potential homesickness. This kind of support could, of course, also be obtained from friends or other acquaintances from back home who had emigrated earlier. In this study, however, it was only possible to trace relatives of the sampled individuals.

The household and individual variables also show some significant results. The likelihood of emigration grew stronger with the number of siblings alive in 1890, especially among women (Table 3). A large family could mean difficulty in maintaining subsistence for everyone in bad times, but an emigrating family member could ease the burden of having many mouths to feed in the family and, if they succeeded abroad, could send remittances home. On the other hand, having many siblings could mean lower expectations of caring for parents, older relatives, or younger siblings, making it easier to emigrate. Number of siblings could also influence future possibilities for staying on the farm or affect future inheritance in other ways.

In terms of father’s occupation, women whose fathers were unskilled workers showed a higher propensity to emigrate than women whose fathers were farmers. In general, men emigrated regardless of the social background of their family. For most people, emigration meant an opportunity to improve or maintain their social status. This was likely an issue in most social classes in times of population growth combined with scarce working opportunities. For many young people who came from homes with poor circumstances, emigration was the only chance to make a living and obtain a better life. Social status could also be an issue for individuals coming from well-off families, perhaps with a farm and land of their own: depending on prospects for future inheritance, emigration could be a chance to achieve future prosperity or at least maintain social status. This indicates that emigrants may have been positively selected on traits such as ambition and initiative (Chiswick 1999). My clear evidence of chain migration shows, however, that emigration was also a possibility for many others.

Furthermore, whether both parents were still alive seems irrelevant to the decision to emigrate. Men and women were no more likely to emigrate if both parents were alive, which was not the expected result.

In terms of age, the results show that the likelihood of emigrating peaked in the 19–23 age group and then declined with age, particularly for women. A decline for men is also evident, but their likelihood of emigrating did not become significantly lower until age 29–34.

Marital status impacted differently on men and women’s propensities to emigrate. The results show that men’s decisions to emigrate were not affected by their marital status, while emigration was rather unusual among married women. Other research has shown this was a phenomenon evident throughout the country, particularly in Halland, as a large number of married men moved to America and left their wives (‘the America widows’) and children back home (Sundbärg 1910, p. 115; Carlsson 1976, p. 126). This was probably a result of the emigration stream from Sweden having developed by the late nineteenth century into more of a labour migration, mainly consisting of individuals who moved to get a job in the US. Diminishing opportunities to obtain land or a larger farm in the US also reduced the incentives to leave Sweden for good with the whole family. Instead, it appears that many married men moved for a limited period and then returned back home to their families. The official statistics for this period are also consistent with this phenomenon as they show the number of returning married men clearly exceeding the number of returning married women (Bidrag till Sveriges officiella statistik 1890).

A majority of the emigrants in the sample were in the younger two age groups. It is possible that family emigration history and other variables had different effects on the propensity to emigrate for younger and older individuals. To investigate this, I divided the sample into two groups by age and separately ran regressions on the younger and older age groups.

Tables 4 and 5 show the effect of all variables on the likelihood of emigrating in 1891 for men and women aged 14–23 and 24–34, respectively. I excluded the marital status indicator from Table 4 because of the small number of individuals in the younger age group who were married.

It is evident that family emigration history was more significant in the migration decision process for younger men and women than their older counterparts. Among the older group, family emigration history still had some impact for women, but among men it seems less relevant. Older individuals likely did not depend on other family members in the migration decision process; they perhaps had a wider social network outside the family.

We can also note that the household context affected the age groups in different ways. Primarily, father’s SES had a stronger influence on young women’s decisions than on those of older women or men in general. The number of siblings had a stronger impact on younger women and on older men, which shows, as discussed earlier, how family size could have different impacts on individuals’ decisions and future possibilities.

Since I could not trace some siblings, I originally assumed that none of the non-traced siblings had emigrated. As a robustness check, I re-estimated the regressions assuming instead that all non-traced siblings had emigrated. This is a completely unrealistic assumption given that the average yearly emigration rate was around 10 per 1,000 inhabitants. The robustness check can thus be seen as a test of the direction in which the assumption that no non-traced siblings emigrated biased the coefficients. The results show that the coefficients for family emigration history in Tables 2–5 are likely somewhat upwardly biased. When assuming that all non-traced siblings emigrated, the coefficients were weaker and not always statistically significant (Tables A5–A8 in the supplementary material). The coefficients could also be upwardly biased because the migration history variable is a ‘family history’ variable, that is, it counts the number of previous cases among close relatives (Khoury and Flanders 1995). What is important about the results in Tables 2–5 is thus not the size of the coefficients but rather the direction of the effects and their statistical significance. The upward bias does not change the finding of strong evidence for a clear pattern of chain migration in the sampled nineteenth-century population.

Conclusion

Chain migration refers to the social process by which potential migrants are influenced by family members or friends who have previously migrated, and eventually follow them to a new place of residence. It has been a consistent feature of both historical and present-day migration and is relevant today, in terms of the extensive and difficult political issues surrounding the handling of current migration flows, not least family reunification. Most previous studies on overseas migration (in the nineteenth century) have been performed at the macro level, which does not fully explain the interaction between potential emigrants and their relatives and friends who had already moved. This study provides further knowledge by investigating the subject at the micro level, in order to establish the mechanism of chain migration more clearly and obtain a better understanding of the migration decision process.

The study is based on meso-level theory and migration network theory, using household- and individual-level data on emigration from Halland, a county in south-west Sweden, to the US in 1891. A benefit of studying chain migration during this period is that the absence of legal constraints allows a clearer picture of the micro- and meso-level determinants to be obtained. I introduce a new method to historical studies on chain migration: using a case–control design. This compares a random sample of emigrants with a random sample of non-emigrants across a number of potential determinants, such as individual characteristics, household structure, and most importantly, family emigration history. The study design and the use of micro-level data enable me to present results that formerly may have been assumed but not proven empirically.

Primarily, the results show a clear presence of chain migration from Halland to the US during the late nineteenth century. Family emigration history had a strong impact on the propensity to emigrate. In addition, I show that the chain migration effect differed between groups. Family emigration history was more important to younger than older people in general, and results also show gender differences, with the influence of family members who had emigrated earlier having a stronger impact on women’s than men’s decisions to emigrate. The gender difference is also noticeable for other covariates, with some circumstances of more importance to women in the emigration decision process than to men; these include household size and SES, but also individual characteristics, such as marital status. Living conditions in the countryside were similar throughout Sweden and the ongoing societal and demographic change affected emigration streams from all over the country, which could indicate that the proven effects for Halland fully apply to Sweden as a whole. However, circumstances specific to Halland, such as land ownership and family structure, suggest that studies on regions with other conditions might show different results.

Current migration processes constitute a significant feature of today’s globalized world. Many migrants choose to move voluntarily, but the rapidly increasing refugee flows around the world clearly show that too many people are forced to leave their homes. Although this paper examines emigration from one small area in Sweden in the late nineteenth century, we can see similarities to populations and ongoing emigration processes in sending countries today. Emigration from Sweden in the late nineteenth century was generally voluntary, but rapid population growth and hard economic conditions had led to difficulties for many young adults in finding a way to make a living, and emigration became a chance to obtain a better future. As a part of ongoing industrialization in the western world, the mass emigration was thus largely a relocation of people from rural regions with poor conditions to more urbanized and industrialized areas. Similar migration processes are also present in many societies today and this study contributes to the knowledge of what influences individuals’ migration decisions and the mechanisms of chain migration.

Data sources

National Archive of Sweden, Riksarkivet, The Swedish Census 1890. Available: https://sok.riksarkivet.se/folkrakningar (accessed: 16 October 2016).

National Archive of Sweden, Riksarkivet, Catechetical Records, Husförhörslängderna, and Migration Registers for the Years 1856–1891 for the Following Counties: Halland, Malmöhus, Kristianstad, Blekinge, Jönköping, Göteborg- och Bohuslän, Elfsborg and Skaraborg. Available: https://sok.riksarkivet.se/kyrkoarkiv?infosida=start (accessed: 16 October 2016).

Notes and acknowledgements

1 Anna-Maria Eurenius is in the Unit for Economic History, Department of Economy and Society, School of Business, Economics, and Law, University of Gothenburg. Please direct all correspondence to Anna-Maria Eurenius, PO Box 625, SE-405 30 Gothenburg, Sweden; or by E-mail: [email protected]

2 I appreciate the support from my supervisors, Christer Lundh and Stefan Öberg, and especially the latter for guiding me in the statistical method. I have benefited from comments and suggestions on earlier versions of this paper by participants of seminars in Gothenburg and Glasgow (European Social Science History Conference 2012). I also thank two anonymous referees for their valuable comments.

Appendix: Odds ratios

The odds of a certain event happening are the probability of that event occurring divided by the probability of the event not occurring: probability/(1 − probability).

The relative odds of the event occurring are expressed as the odds ratio (i.e., the odds for a specific category divided by the odds for the reference category). This can be shown in a 2 × 2 table as follows:

Table

	Cases	Controls	Total
Exposed	a	b	a + b
Unexposed	c	d	c + d
Total	a + c	b + d	a + b + c + d

The odds ratio (OR) can be calculated as: OR = (a/c)/(b/d) = ad/bc (Hennekens and Buring 1987). The sizes of the two groups do not matter because the odds ratio compares the relative frequency of exposed individuals within the two groups: cases and controls. Because the odds of the reference category are set to one, an odds ratio greater than one indicates a larger risk for the category in question than for the reference category, while an odds ratio less than one indicates a smaller risk.