Paper for session Migration at the Swedish Economic History Meeting, Gothenburg 25-27 August 2011 Movers and stayers. Household context and emigration from Western Sweden to America in the 1890s Anna-Maria Eurenius, Dept. of Economic History, University of Gothenburg, Sweden. anna-maria.eurenius@econhist.gu.se First draft, please do not quote Abstract Emigration from Europe to North America in the 19th and early 20th century was of great societal importance and has inspired many studies at the aggregate level. We know less however about the individual and household level characteristics of migrants and the extent to which migration was selective. This paper deals with the importance of household structure and socioeconomic status for individual decisions to migrate. The design of the study implies a case control study of longitudinal individual level data for a random sample of emigrants in 1891 compared with a control group of stayers. Data are from the Swedish census 1890 and parish records for the period 1855-1891 for Halland county in Western Sweden. The indicator variable is emigration in 1891 (or not) and the explanatory variables describe individual characteristics (sex, age, marital status, occupation), household features (fathers occupation, survival of parents) and the individual s position in the household (birth rank in relation to surviving siblings). Also information about previous emigration in the family is available for analysis. 1
1. Introduction Establishing the determinants of out-migration, i.e. what makes individuals leaving one place of residence for another, has for long been one of the major themes in migration studies. From an economic theory point of view the decision whether to move or not is a rational choice of the best option when the benefits (e.g. higher earnings) and costs (e.g. moving costs or psychic costs) of a migratory move have been calculated and compared. If the net gain from migration is positive, the individual decides to move; if it is negative, the decision is to stay. In this way, migration is an investment in higher net earnings in the future. Factors influencing the decision to move in this strand of literature are sex, age, marital status, human capital, earnings and employment (Sjaastad, 1962; Todaro 1969). For younger persons not only the own employment and earnings could be assumed to be important, also the possible support from the parental household and the competition over family resources with siblings could be assumed to count. Theory of chain migration predicts that the tendency to migrate is influenced by previous cohorts of migrants from the family or location (Carlsson, 1976). Although theory on migration determinants is basically at the micro level, studies of oversees migration in the 19 th century typically deals with macro data (Bohlin and Eurenius, 2010; Hatton and Williamson, 1993, 1998). The reason for this is simply lack of individual level data. Emigration was a rare event, even in the 19 th century. At the most 1 % of the population moved during a year, usually substantially fewer. Calculating based on individual level data the importance of possible determinants of a rare outcome requires a large risk population. Modern Swedish register data allows such calculations since they could be made for the entire population or regional parts of it. For the period before the 1970s such data is not available though, and historians have to use databases of local populations of much smaller size, which makes it difficult to study the determinants of emigration. This study takes another approach, and attaches to an epidemiological method used in medical studies of rare outcomes: the case-control design. It implies that a group of emigrants is compared with a control group of non-emigrants across a number of potential determinants. The methods requires that the two groups could be randomly drawn from the populations of movers and stayers, which is possible using the census of 1890 and the migration register of 1891. 2
2. Data 2.1 Sources This study is performed on data from the county of Halland in the southwest of Sweden, which was the county with highest emigration rates during the period 1880 1910. The yearly average emigration rate in Halland was approximately 10 persons per thousand inhabitants. The same rate for the whole nation during that period was just below 6 persons. The study is based on individual data, which makes it possible to analyze the impact of factors reflecting human capital, household structures and socioeconomic status had on the decision to migrate. The sources are the 1890 census and church records such as catechetical examination records and migration registers. In the catechetical examination records the household members were listed by the parish priest every year. The purpose was to examine the biblical knowledge and reading skills of those living in the household. The records included basic lists of all household members and their birth years and birth parishes. Information on changes in the household since the previous record such as births, deaths and migration were also noted in the records. The catechetical examination records contained information on migration both into and out from the parish, and also within the parish. The priest recorded, when, where to and from where someone moved. When a person moved into or out from the parish he or she had to take out a change of address-certificate. It was issued by the parish priest who also recorded all moves during the year in the migration register. One can assume a certain underestimation of the emigration rates in the registers since the priests not always could separate emigration and domestic migration. The 1884 Emigration Ordinance made the data more reliable. Emigrant agents were then prohibited to convey any travels abroad if they could not present proper migration certificates for all emigrants to the police authority. The source material has made it possible to collect the necessary information to get as complete picture as possible of the living conditions of every individual in this study from birth until the year 1890. 3
2.2 The study design and sample Even though Halland showed the highest emigration rates in Sweden, the overall probability to emigrate still was very small. On average one percent of the population in Halland emigrated each year during 1880-1910, which means that the necessary size of a random sample reflecting the real circumstances would be far too extensive. A suitable method for this study is therefore to perform a case-control study. The purpose of such a study is to identify and evaluate factors that may have an impact on an event by comparing groups of individuals that will execute the event (the cases ) with groups that will not execute the event (the controls ) but are otherwise similar. This method is often used in epidemiological studies where individuals who have, for example, a disease are compared with a group that don t have the disease regarding earlier exposure to different factors. Thus, the design of this study is as follows: Event: Emigrate to North America 1891 Study population: Men and women living in the countryside in the county of Halland, born 1856 1876, amounted to 42 382 persons. Sample of movers (the cases ): As a first step all emigrants from Halland 1891 living in the countryside were identified. This was done by searching the migration registers for 1891 in every one of the 88 countryside parishes in Halland. It proved to be 1 500 persons emigrating. Then, all emigrants born 1856 1876, i.e. in ages 15-35, with North America as destination were sorted out and summed up to 1 150 individuals. At last a random sample was conducted by extracting every tenth of that group, which finally resulted in 115 emigrants constituting the cases in the study. Sample of stayers (the controls ): The group of controls was made as a random sample of 200 individuals from the study population in the same birth cohorts as the sample of movers. To access all relevant persons the sample was based on the census 1890. 4
The starting point for this study is 1890. The cases emigrated to North America in 1891 while the controls did not. To determine what influenced the decision on whether to move or not a rather significant amount of information has been collected for every individual. This has been done starting with the census 1890 and then by extracting data from church records. The aim has been to track every individual back to their place of birth and the family of origin. Mobility among young people in the countryside was still large during the latter part of the 1800s and changing employments often also meant changing place of residence. This has sometimes made the tracing rather complicated. The number of observations in the study may seem rather limited in this first draft. This is due to the extensive time required to locate each individual and to excerpt all necessary data. 2.3 The variables The independent variables in this study can be divided into two groups. The first group consists of variables that reflect the individual s human capital. A general view within migration research is that young unmarried people with no land of their own were the most anxious to emigrate. They were considered to have had more to gain from migration than other groups and less ties to the home country. The variables sex, age, marital status and the individual s own profession represent this. The other group of variables reflects household structure and socioeconomic status. The assumption here is that the propensity to emigrate was dependent on the parents social status and the individual s position in the family. This could for instance have an impact on the future prospects to be able to stay on the family farm or in other ways being favored with regard to inheritance. Variables representing such aspects are father s occupation, whether one or both parents were alive in1890, and the number of older siblings alive in 1890. The importance of previous emigration for emigration rates is well known, and lots of emigrants in the latter part of the 1800s travelled with prepaid tickets or had received crucial information from previous emigrants to the New World. Therefore, the final variable captures the family history of previous migration, i.e. the existence of previously migrated siblings or parents. 5
Table 1 shows the variables reflecting individual characteristics. The first one is the individual s sex. The distribution between men and women is virtually equal for the total sample, which correspond to the population in Halland as a whole. The second variable is the individual s age. Just above 75 % of all emigrants from Halland in 1891 were in the age 15-35 years. This corresponds well with the figures for the total amount of emigrants from Sweden and this is why only individuals in that age span are included in the study. To investigate any difference in the propensity to emigrate within this age group, the population is divided into four 5-year groups. Next variable is marital status and it is clear that the vast majority in the study was unmarried. High marriage age was something that was characteristic not only for Sweden, but also for the rest of Western Europe. Moreover, the average marriage age in Halland tended to be among the highest in Sweden. Next variable is the individual s occupation. The data is extracted from the 1890 census and the occupational titles are coded by the HISCO system. Since the sample is rather small it was necessary to merge similar occupations into only four categories. The first group represents landowners and includes peasants, freeholders and tenants. The second group represents the landless and includes crofters, cottagers and lodgers. The third group is very diversified and includes all other occupations but primarily maids, farm hands and farm workers. The forth group consists of sons and daughters still living in the parental home having not yet really entered into employment and therefore lacking occupational titles in the records. Table 2 shows the group of variables reflecting household structure and socioeconomic status. The first one is the father s occupation at the individual s time of birth, capturing the impact of the socioeconomic status of the parental household on the individual s likelihood of emigration later in life. The occupational categories are constructed in the same way as the variable showing the individual s occupation, with the one difference that the fourth group represents the poorest in society, the paupers. The next group of variables reflects the impact of the presence of parents in 1890 on the probability of migration in 1891. Four categories are compared: both parents being alive, either father or mother being dead, and both parents being dead. The next variable also concerns the family structure. The number of older siblings alive in 1890 reflects the impact of parity on future possibilities to stay at home and perhaps take over the farm one day. The variable is categorized into four groups with increasing numbers of siblings. The last variable captures the importance of previously migration of family members. Here family refers only to siblings and parents, and not more distant relatives. The variable is divided into two categories; those with a family history of 6
previous migration, and those without any previously migrated parents or siblings. All variables are constructed as dummies. Table 3 and 4 report the distribution of the variables within the two samples. The tables show that there are differences between movers and stayers, for instance in the distribution of gender, marital status, parental presence and previous family migration history. Next, we will test whether such bivariate associations between potential determinants and emigration remain in a multivariate analysis. 7
3. Results The purpose of a case-control study is to identify and evaluate different factors that may have an impact on the probability that an event will occur. To measure the extent to which there is an association between the event and the different independent variables the odds ratio. The odds for a certain event to happen is the probability for that event to occur divided with the probability for the event not to occur. The relative odds for the event to occur is expressed as the odds ratio, i.e. the odds of a specific category divided by the odds of the reference category. Since the odds of the reference category is set to 1, an odds ratio > 1 indicates that the probability is larger for the category in question than for the reference category, while an odds ratio < 1 indicate a smaller probability. The odds ratio cannot, however, be automatically interpreted in terms of probability and percentages. One way to make it easier to interpret the results is to determine the anti logarithm of the odds ratios. Then it is possible to think of the relations in terms of percentage changes. Two regressions have been estimated. The first one is based on the variables reflecting individual characteristics and the results are displayed in Table 3 and 4. Table 3 shows the odds ratio and Table 4 the marginal effects of each variable, expressed as percentage changes. As can be seen there are some variables that shows significant results. If we look at the variables reflecting the gender and marital status the probability to emigrate is 17 % lower for women than for men and 28 % lower for married than for unmarried persons. Another factor that seems to have had a rather strong effect on the emigration decision is the individual s own profession. Compared to the group of landowners the probability for emigration among the landless group was almost 42 % higher. As to the influence of age, it is clear from the Tables that being 30-35 years was associated with a 15 % lower likelihood of emigration compared to the reference category (15-20 years). Controlling for factors reflecting individual characteristics, variables capturing household structure and socioeconomic status has been added in the second regression and the results are presented in Table 5 and 6. As before Table 5 shows the odds ratio and Table 6 the marginal effects. The variables from the previous regression are showing similar results. The probability to emigrate among females are now almost 20 % lower than for men and married people show 27 % lower probability to move than unmarried persons. The current occupation among the individuals is still important and has increased its impact. Now we 8
can see that except for the landless group, which shows a probability to move that is almost 50 % higher than the landowners, the probability of emigration of people in other occupations is almost 40 % higher than the reference category. The influence of the age variable is still rather week, but the overall picture is a negative association of age and emigration, which is clearest for the age group 30-35. Among the variables added in the second regression there are especially two that stand out because they represent large effects that are highly statistically significant. It is rather obvious that the variable reflecting the numbers of previous emigrants in the family has a strong impact on the decision to emigrate. The probability of emigration was 32 % higher for individuals with a family emigration history compared to those without. The other variable is the one showing the effect of both parents being dead or alive in 1890. It turns out that the propensity to emigrate was 22 % lower if both parents were dead compared to if they were both alive. Having one or the other of the parents widowed does not seem to have had that much effect on migration decisions though. The coefficients of the variables of father s occupation and number of older siblings alive in 1890 were not statistically significant for this sample. 9
4. Discussion When trying to understand people s choices in different situations, one quickly realizes that most decisions are influenced by many considerations and circumstances. People are unique and affected by a wide variety of factors. Such a crucial decision as to emigrate to North America in the late 1800 s was of course based on careful considerations that were exclusive to each potential emigrant. Most research on determinants of migration has been conducted on an aggregated level data. As being performed on individual level data this study is an attempt to, in a more intrusive way, approaching the individual reasons behind a person s decision to move or to stay. The results show that the factors most decisive for the decision to migrate were those primarily related to the individual s life situation. Young people with some kind of employment and income but with no family or farm of their own were those who were most prone to emigrate. This outcome was perhaps not so surprising. A little bit more unexpected was, on the other hand, the fact that the social background didn t seem to have had any impact on the migrant decision. To emigrate meant for most people an opportunity to improve or maintain their social status. This was probably an issue within most social classes in times with a growing population in combination with scarce working opportunities. The presence of parents seems, however, to have been of great importance for the propensity to emigrate. Both parents being alive meant probably more support when making the crucial decision. Partly through economic contributions and also perhaps by reducing the pressure and expectations of taking care of a widowed parent or younger siblings. The family history of previous migration was also a factor that had a very clear impact on the decision to migrate. Any siblings or parent, regardless of the family s social status, that already had made the trip had a great influence on new presumptive emigrants back home. In addition to financial support, which was very frequent, they could contribute with useful information about the New World or also perhaps by helping out with a job or a place to stay. This study is a first draft with preliminary data analyses. It will eventually be extended with a larger number of observations. The idea is to make a sample of 250 movers and 500 stayers. It would then be interesting to adjust the variables and perhaps to change categorizations. It could also be of interest to test interactive associations between different variables. 10
Tab.1 Variables reflecting human capital Tab.2 Variables reflecting household structure and socioeconomic status Variables Frequence Percent Variables Frequence Percent Sex Fathers occupation male 157 50.16 landowner 179 56.83 female 158 49.84 landless 76 24.13 Total 315 100.00 other occ. 57 18.10 Age pauper 3 0.95 15-20 127 40.32 Total 315 100.00 21-25 67 21.27 Parents alive 1890 26-30 71 22.54 Both parents 31-35 50 15.87 Yes 115 36.83 Total 315 100.00 No 199 63.17 Marital status Total 315 100.00 married 55 17.46 Father unmarried 260 82.54 father alive 279 88.57 Total 315 100.00 father dead 36 11.43 Own occupation Total 315 100.00 landowner 24 7.62 Mother landless 13 4.13 mother alive 255 80.95 other occ. 78 24.76 mother dead 60 19.05 missing 200 63.90 Total 315 100.00 Total 315 100.00 Both parents dead 1890 Yes 20 6.35 No 295 93.65 Total 315 100.00 Numbers of older siblings alive 1890 0-1 158 50.16 2-3 94 29.84 4-8 63 20.00 Total 315 100.00 Previous migrants mighist 0 226 71.75 1-7 89 28.25 Total 315 100.00 11
Tab.3. Distribution of movers and stayers. Variables reflecting human capital Tab.4.Distribution of movers and stayers. Variables reflecting household structure and socioeconomic status Movers Stayers Movers Stayers Variables Freq. Percent Freq. Percent Variables Freq. Percent Freq. Percent Sex Fathers occupation male 73 63.48 84 42.00 landowner 61 53.04 118 59.00 female 42 36.52 116 58.00 landless 34 29.57 42 21.00 Total 115 100.00 200 100.00 other occ. 19 16.52 38 19.00 Age pauper 1 0.87 2 1.00 15-20 55 47.83 72 36.00 Total 115 100.00 200 100.00 21-25 26 22.61 41 20.50 Parents alive 1890 26-30 26 22.61 45 22.50 Both parents 31-35 8 6.96 42 21.00 Yes 82 71.30 117 58.50 Total 115 100.00 200 100.00 No 33 28.70 83 41.50 Marital status Total 115 100.00 200 100.00 married 5 4.35 50 25.00 Father unmarried 110 95.65 150 75.00 father alive 12 10.43 24 12.00 Total 115 100.00 200 100.00 father dead 103 89.57 176 88.00 Own occupation Total 115 100.00 200 100.00 landowner 1 0.87 23 11.50 Mother landless 5 4.35 8 4.00 mother alive 17 14.78 43 21.50 other occ. 32 27.83 46 23.00 mother dead 98 85.22 157 78.50 missing 77 66.96 123 61.50 Total 115 100.00 200 100.00 Total 115 100.00 200 100.00 Both parents dead 1890 Yes 4 3.48 16 8.00 No 111 96.52 184 92.00 Total 115 100.00 200 100.00 Numbers of older siblings alive 1890 0-1 50 43.48 108 54.00 2-3 39 33.91 55 27.50 4-8 26 22.61 37 18.50 Total 115 100.00 200 100.00 Previous migrants mighist 0 64 55.65 162 81.00 1-7 51 44.35 38 19.00 Total 115 100.00 200 100.00 12
Tab.3. Logistic regression Number of obs = 315 Wald chi2 (17) = 44.43 Prob > chi2 = 0.000 Log pseudolikelihood= - 184.23411 Pseudo R2 = 0.1088 Robust emig1891 Odds Ratio Std.Err. z P>IzI (95% Conf. Intervall) female.4543143.1166348-3.07 0.002.2746817.7514204 agegroup2.9086888.2954152-0.29 0.768.4804953 1.7184670 agegroup3 1.0155670.3361883 0.05 0.963.5308005 1.9430600 agegroup4.4618341.2218729-1.61 0.108.1801169 1.1841800 married.1918288.1299455-2.44 0.015.0508519.7236367 ownprof2 5.9886110 6.5530290 1.64 0.102.7012940 51.1389800 ownprof3 4.7449950 4.9670100 1.49 0.137.6098278 36.9202300 ownprof4 2.4578430 2.6897470 0.82 0.411.2877659 20.9927300 Tab.4. Marginal effects after logit y = Pr(emig1891) (predict) =0.32538841 variable dy/dx Std.Err. z P>IzI (95% C.I.) female* -.1718420.05553-3.09 0.002 -.280669 -.063014 agegro2* -.0208119.06998-0.30 0.766 -.157974.116350 agegro3*.0033959.07288 0.05 0.963 -.139449.146240 agegro4* -.1516970.08245-1.84 0.066 -.313298.009904 married* -.2816659.08231-3.42 0.001 -.442997 -.120335 ownprof2*.4190828.21752 1.93 0.054 -.007252.845418 ownprof3*.3618337.23501 1.54 0.124 -.098781.822449 ownprof4*.1869687.21202 0.88 0.378 -.228576.602513 (*) dy/dx is for discrete change of dummy variable from 0 to 1 Referencecategories Sex male 0 Age agegroup1 age 15-20 Marital status unmarried 0 Own profession ownprof1 landowner 13
Tab.5. Logistic regression Number of obs = 315 Wald chi2 (17) = 67.01 Prob > chi2 = 0.0000 Log pseudolikelihood= - 168.66074 Pseudo R2 = 0.18414 emig1891 Odds Ratio Robust Std.Err. z P>IzI (95% Conf. Intervall) female.3912056.1100272-3.34 0.001.2254256.6789015 agegroup2.9397356.3533390-0.17 0.869.4497337 1.9636130 agegroup3.9880561.3453016-0.03 0.973.4980895 1.9599990 agegroup4.4968527.2406173-1.44 0.149.1923115 1.2836600 married.2013318.1435586-2.25 0.025.0497700.8144366 ownprof2 9.0232470 9.3913990 2.11 0.035 1.1733620 69.3895000 ownprof3 5.6706180 4.9658660 1.98 0.048 1.0191010 31.5532000 ownprof4 2.2106180 2.0339820 0.86 0.389.3641885 13.4184200 fathprof2 1.5093190.4574814 1.36 0.174.8332550 2.7339090 fathprof3.9857108.4114410-0.03 0.972.4349641 2.2338070 fathprof4 2.1913310 2.1651340 0.79 0.427.3159964 15.1961500 fathalive.6656765.2855080-0.95 0.343.2871995 1.5429180 mothalive.7608942.2740482-0.76 0.448.3756215 1.5413390 parsdead.2575026.1884071-1.85 0.064.0613737 1.0803910 oldsibl2 1.4092720.4533082 1.07 0.286.7502333 2.6472400 oldsibl3.97996180.3398076-0.06 0.953.4966494 1.9336080 mighist26 3.9835630 1.1953760 4.61 0.000 2.2123080 7.1729500 Tab.6. Marginal effects after logit y = Pr(emig1891) (predict) =.31230751 variable dy/dx Std.Err. z P>IzI (95% C.I.) female* -.1995471.05876-3.40 0.001 -.314714 -.08438 agegro2* -.0132588.07961-0.17 0.868 -.169292.142775 agegro3* -.0025774.07487-0.03 0.973 -.149322.144167 agegro4* -.1352704.08421-1.61 0.108 -.300317.029776 married* -.2674039.08300-3.22 0.001 -.430083 -.104724 ownprof2*.4959770.17530 2.83 0.005.152393.839561 ownprof3*.3981666.19246 2.07 0.039.020949.775384 ownprof4*.1622580.17793 0.91 0.362 -.186483.510999 fathpr2*.0915736.06881 1.33 0.183 -.043286.226433 fathpr3* -.0030857.08932-0.03 0.972 -.178155.171983 fathpr*.1862159.24596 0.76 0.449 -.295861.668293 fathale* -.0818561.08033-1.02 0.308 -.239292.07558 mothale* -.0567247.07222-0.79 0.432 -.198276.084827 parsdead* -.2180483.07904-2.76 0.006 -.372972 -.063125 oldsibl2*.0754232.07227 1.04 0.297 -.066225.217071 oldsibl3* -.0043374.07414-0.06 0.953 -.149652.140977 mighist*.3153291.06771 4.66 0.000.182623.448035 (*) dy/dx is for discrete change of dummy variable from 0 to 1 Referencecategories Sex Father s profession male 0 fathprof1 landowner Age Parental lifstatus1890 agegroup1 age 15-20 parentsalive both parents alive 1890 Marital status Older siblings 1890 unmarried 0 olsibl1 0-1 older siblings alive 1890 Own profession Previous migrants ownprof1 landowner No previous migrants 0 14
Appendix Sources of data 1890 census Churh registers from the county of Halland: Cathechetical examination registers for 1856-1891 Migration registers for 1891 References Bohlin J., Eurenius, A.-M., 2010. Why they moved Emigration from the Swedish countryside to the United States 1881-1910. Explorations in Economic History 47, 533-551. Carlsson, S., 1976. Chronology and composition of Swedish emigration to America. In: Runblom, H., Norman, H. (Eds), From Sweden to America. Acta Universitatis Upsaliensis, Minneapolis Uppsala. University of Minnesota Press. 114-148. Hatton, T.J., Williamson,J.G., 1993. After the famine: emigration from Ireland, 1850 1913. Journal of Economic History 53, 575-600. Hatton, T.J., Williamson,J.G., 1998. The Age of Mass Migration. Causes and Economic Impact. Oxford University Press, Oxford. Sjaastad, L.A., 1962. The costs and returns of human migration. Journal of Political Economy 70, 80-93. Todaro, M.P., 1969. A model of labor migration and urban unemployment in less developed countries. American Eonomic Review 59, 138-148. 15