Religious communities

6Since the Reformation, the Netherlands has been a religiously divided country (Knippenberg, 1992). As a general rule of thumb, it can be said that the north of the country was Protestant, whereas the southeast was Roman-Catholic. Roman-Catholicism was the dominant religion in Brabant, Limburg, and the southern parts of Gelderland, whereas in the other provinces most individuals were Protestant. The spatial division between a Protestant north and a Roman-Catholic south makes it seem like there were two homogeneous religious regions in the Netherlands. However, the ‘north’ was not a religiously homogeneous zone. First, religious communities often collided with local borders, creating denominational and theological differences between places. Roman-Catholic religious communities can be found along the Holland coastline, in the east of the country, the south of Zeeland and – to a lesser degree – in rural South Holland and western Utrecht (Knippenberg, 1992). Second, Protestantism refers to a widely diverging group of Churches among which the Dutch Reformed Church was by far the largest. The dominant Dutch Reformed Church was an amalgam of different religious communities, “ranging from ultra-liberal to ultra-orthodox” (Kok, 2017, 60). As such, most provinces were a patchwork of liberal-Protestant, orthodox-Protestant and Roman-Catholic religious communities.

7In total, approximately half of the Zeeland population was liberal Protestant, whereas one quarter was Roman-Catholic and another quarter orthodox Protestant (see tab. 4). Religion had a significant impact on daily life, as the church instilled values and norms regarding family, sexuality, and marriage (Kok, 2017). A wide range of literature has shown that religion is also associated with mortality in later life. In general, religion is thought to decrease mortality rates by promoting temperance of alcohol, tobacco, and diets, helping followers cope with adversity, and stimulating social support (Koenig, King, Carson, 2012). Increased levels of survival have been found among followers of religious communities with high levels of social control and strong regulations on the abstinence from alcohol and tobacco, e. g. Mormons, Seventh-Day-Adventists, and Old Order Amish (Berkel, 1979; Fraser, 1999; Hamman, Barancik, Lilienfeld, 1981; Lindahl-Jacobsen et al., 2013; Temby & Smith, 2014). These groups seem to show an extreme example of a general principle: increased religious participation is associated with decreased alcohol intake, less tobacco consumption, and a lower chance on a poor diet (Koenig, King, Karson, 2012). These mechanisms are not only present in contemporary society, but also at the turn of the 20th century. In the Netherlands, members of the highly secularised Liberal Protestant churches had higher levels of old-age mortality than members of the Orthodox Protestants church, Roman-Catholic church, or Jewish families, which were known for their strong social control (Kok, 2017). Elsewhere, inactive Mormons in Utah had higher mortality levels and a lower chance to become long-lived than Mormons who remained active in the Church of Latter-Day Saints (Lindahl-Jacobsen et al., 2013; Temby & Smith, 2014). Hence, increased involvement in 20th century religious communities has been associated with increased survival rates.

Environmental conditions

8In Dutch historiography, regional differences in survival have traditionally been explained by environmental conditions. Coastal regions had much higher mortality rates than inland provinces (Devos & Van Rossem, 2017; Hofstee, 1981; NMBG, 1879; van den Boomen & Rotering, 2018; Wolleswinkel – Van den Bosch et al., 2001). The difference between coastal and inland areas is often summarised as a difference between clay and sand. The coastal regions with clay soils had a worse disease climate, less access to clean drinking water, and different agricultural traditions than inland regions with sandy soils (Devos & Van Rossem, 2017; Hofstee, 1981). The first two explanations hypothesise that differences between porous sandy soils and waterlogged clay soils are in fact a juxtaposition between drylands and wetlands (Hedefalk, Quaranta, Bengtsson, 2017; Munro et al., 1997). The wetlands were an excellent breeding ground for pathogens. Waterlogging created pools and puddles which were fertile breeding ground for mosquitoes that transmitted malaria. For clean drinking water, people were dependent on collecting rainwater, as groundwater was often salinised or contaminated by feces and waste from nearby cesspools and canals. Hence, drinking surface water or using it for cleaning purposes was not without risk (Hofstee, 1983; van den Boomen & Ekamper, 2015). A third explanation focuses on the relation between soil type and intensive agriculture (Devos & Van Rossem, 2017). Sandy soils were relatively unfertile and required intensive labour. As a result, inland farming was not restricted by the availability of land, but the amount of land a farmer could work, whereas in coastal regions farms were larger and relied on hired labour. Hence, the soil indirectly determined whether individuals produced their own food or were hired labourers.

9In Zeeland, differences in access to clean drinking water were practically non-existent, as everywhere groundwater was salinised due the vicinity to the sea (Priester, 1998). To keep livestock hydrated, farmers dug basins to collect rainwater and inhabitants collected their own drinking water from roofs in rainwater tanks (Priester, 1998; Hoogerhuis, 2003). Some households had their own tanks, but it was not uncommon that households in the city shared these water tanks (Hoogerhuis, 2003). Although Zeeland had no sandy soils (WUR-Alterra, 2006), significant differences existed in the use of clay and sandy clay soils (Priester, 1998). The difference between clay and sandy clay corresponds with different polders in Zeeland known as oudland (old land) and nieuwland (new land). These terms – old and new land – refer to the origin of different regions. Oudland was the first embanked land in the Zeeland delta, which was reclaimed between 1000 and 1200. It is the lowest lying land in Zeeland and consists of clay resting on layers of peat. Not surprisingly, oudland was more likely to waterlog and these pools and puddles of brackish water were an excellent breeding ground for malaria mosquitoes. So much so, that Zeeland was notorious for its intermittent fevers, which was especially dangerous to people new to the region and young children (Van der Kaaden, 2003; van Poppel, Ekamper, Mandemakers, 2018).

10Oudland was also less suited for agricultural purposes. Due to high soil acidity and salinity of oudland soil, possibilities for agriculture were limited as even crops resistant to saline and acid soils – beets, potatoes, grain, and madder – were hard to grow on these fields. Nieuwland, where the soil consists of a mixture of sand and clay, was much more fertile and often used for agriculture (Priester, 1998). Paradoxically, the lesser quality of the soil might have been beneficial for female and male agricultural labourers who lived in the oudland. Although livestock farming in Zeeland was rare, grassland was more common in the oudland regions (Priester, 1998). Farmer’s wives were often involved in cheesemaking to contribute to the household income. In general, this meant that households were somewhat wealthier (Van Cruyningen, 2005) and that women could work in an economic niche, which is known to increase female survival (Humphries, 1991; McNay, Humphries, Klasen, 2005). It also might have helped that farms in the oudland were smaller (Priester, 1998) and required fewer farm labourers. The lower dependency on farm labourers meant that more households produced their own food and were not dependent on their employer or the market to acquire food. In general, farming seems to have been beneficial for men, as farmers and – to a lesser degree – farm labourers lived longer than other men, while such an effect has never been found for women (Alter, Dribe, Van Poppel, 2007; Edvinsson & Broström, 2012; Ferrie, 2003; Mourits, Smith, Janssens, 2019; Schenk & van Poppel, 2011; Smith et al., 2009; Zimmer, Hanson, Smith, 2016). Hence, in their own way, both female and male agricultural labourers could probably benefit from living in the older Zeeland polders.

Rural-urban differences

11Besides religion and factors associated with soil use, we control for several characteristics of urban contexts. For Zeeland, the 19th century was a period of economic decline. Before the 19th century, Zeeland had been a relatively wealthy province. Between 1600 and 1800, Zeeland was home to the West-Indische Compagnie and had a positive net migration rate to attract labourers for its expanding economy (Priester, 1998). However, during the 19th and early 20th century, the province was characterised by outmigration and many people left for Rotterdam or the new world (Wintle, 1992; Zwemer, 2014). Steady outmigration probably was a way to cope with the economic stagnation that occurred in Zeeland. The tendency to leave the province can be attributed to an excess of labour force in the province (Priester, 1998). This phenomenon also occurred in other saturated habitats that had little room for spatial or demographic expansion (see e. g. Voland & Dunbar, 1995). Besides relieving demographic pressure from the less attractive municipalities in Zeeland, the outmigration also prevented urbanisation. Whereas elsewhere in the Netherlands cities grew like never before, resulting in overcrowding and slums (Van der Woud, 2010), the population of Zeeland and its urban centres – Middelburg and Vlissingen – grew only modestly [2] (see fig. 6). Therefore, urban centres were not as plagued by overcrowding as the rising metropoles (Hoogerhuis, 2003). This might have prevented survival disadvantages that plagued cities in the Netherlands during the 19th century (NMBG, 1879; Van Poppel, 1989).

12Although Zeeland was not densely populated, there were multiple towns with a regional function in Zeeland. Most individuals never had to leave the island, unless they left to find work elsewhere (Bras, 2002). Each island had one or multiple larger towns that catered to the needs of the surrounding villages. These regional centres functioned as gateways to the outside world and had small docks where regular passenger and cargo shipments arrived and left. Oftentimes, these ships sailed to Middelburg and Rotterdam at least once per week [3] (De Kanter & Utrecht-Dresselhuys, 1824), so that the islands were never really secluded from the outside world. Unsurprisingly, the regional centres of trade had a different social composition than the surrounding farming villages, and most of the agricultural labourers rarely frequented these towns (Bras, 2002).

Neighbourhood matrix

14For our analyses, we use shapefiles provided by Boonstra (Boonstra, 2007). Municipal borders in Zeeland changed considerably during our period of observation. Between 1812 and 1862 the number of municipalities decreased from 144 to 113 municipalities. During the period of observation, the number of municipalities decreased even further until 86 municipalities in 1962 (Van der Meer & Boonstra, 2006). To be able to compare our birth cohorts, we selected the shapefiles containing the municipal borders of 1860 to map differences between Zeeland municipalities and compute neighbourhood matrices. Due to restrictions in our data, we merge multiple municipalities, so that 101 remained [4]. The resulting map for Zeeland is shown in fig. 1.

Fig. 1

Islands and municipalities in Zeeland (Hermsen, 2018)

Islands and municipalities in Zeeland (Hermsen, 2018)

15We tested which neighbourhood matrix was most accurate to calculate the Moran’s I statistic of spatial autocorrelation. Neighbourhood matrices define which municipalities border one another and are required to compute clustering of longevity. Because Zeeland is an island region, water could form a natural border between nearby municipalities. Therefore, neighbourhood matrices can only be constructed for municipalities that neighbour one another. This can be done in two ways, depending on whether one defines neighbours by geographic divides or human travelling routes. The first approach sees water as a natural border and sees municipalities that directly border one another as neighbours. The second perspective not only identifies municipalities with a land connection as neighbours, but also those that are connected by ferry. To be able to determine whether longevity clustering associated more strongly with environmental features or human travelling routes, we constructed two neighbourhood matrices.

16The neighbourhood matrix that controlled for land connections contained all adjacent municipalities which were retrieved using the queen’s method. The graphical representation of the corresponding neighbourhood matrix is shown in fig. 2, left. The second neighbourhood matrix contained both land and ferry connections that sailed regularly between the islands. We added the ferry lines as operating in 1824 (De Kanter & Utrecht-Dresselhuys, 1824) to the neighbourhood matrix retrieved using the queen’s method. The graphical representation of the matrix is shown in fig. 2, right. The ferry lines are shown as dotted lines. It turned out that spatial autocorrelation was higher when we used the neighbourhood matrix without ferry connections. Neighbouring municipalities were more similar if we only took geographic adjacency into account and not human travel routes. Hence, we used the queen’s method to further study the spatial clustering of longevity in Zeeland.

Fig. 2

Neighbouring municipalities (solid) + ferry connections (dotted) in Zeeland

Neighbouring municipalities (solid) + ferry connections (dotted) in Zeeland

Method

17To measure spatial clustering of longevity, we ran empty multi-level models with belonging to the top 10% as the dependent variable. From the residuals we calculated the Moran’s I, which measures spatial autocorrelation. This procedure checked whether the chance to become long-lived was similar for neighbouring municipalities or randomly distributed over the province. To determine whether the spatial clustering of longevity was stable over the life course, we repeated the regression analyses four times. Each time, we used a different starting age – 5, 20, 30, and 50 instead of 0 – to see whether spatial clustering of longevity in the null models’ residuals increased, decreased, or remained stable if we moved the observation window. An increasing Moran’s I means that spatial clustering is stronger at older ages, whereas a decreasing Moran’s I hints at the importance of the environment in early life. A stable Moran’s I, on the other hand, indicates that environment features have a stable effect over the life course on an individual’s chances to become long-lived.

18If longevity clustered spatially, we violated an assumption of the multi-level regression model and cannot interpret the models until we can explain the spatial clustering. Therefore, we added environmental variables (dominant religion, soil type, population density, population size, net taxable income per capita, economic focus, and the migration rate) to see if these variables explain spatial clustering. The main goal of these models is to see whether the Moran’s I decreases and becomes insignificant. In such cases, we conclude that certain environmental indicators explain spatial clustering of longevity. This does not necessarily mean that the environmental features themselves explain clustering, but rather that they are a proxy for an underlying mechanism (e. g. soil type is a proxy for waterlogging, agricultural traditions, etc.).

Did longevity cluster in Zeeland

20Fig. 3 shows the average chance to belong to the top 10% survivors for the Zeeland municipalities. Scores are centred around the mean. Light tints indicate that inhabitants of a municipality had a below-average chance to belong to the top 10% survivors, whereas dark tints show that inhabitants had an above-average chance to belong to the top 10% survivors. Differences in the chance to belong to the top 10% survivors are shown in terms of standard deviations. The darker the colour, the stronger the deviation from the mean.

Fig. 3

Clustering of individual chances to become long-lived in Zeeland

Clustering of individual chances to become long-lived in Zeeland

21In most municipalities, individual chances to become long-lived diverged little from the mean. Nevertheless, there are multiple municipalities where the chance to belong to the top 10% survivors differs more than 1 standard deviation from the average chance to become long-lived in Zeeland. These municipalities are spread over the entire province. A small group of municipalities with low chances to belong to the top 10% survivors is located in southern Zuid-Beveland, whereas groups of municipalities with above-average chances to belong to the top 10% survivors are located in western Walcheren and on Schouwen. However, the Moran’s I of 0.11 is not significant, which indicates that the resemblance between these neighbours is likely based in chance.

Is the spatial clustering of longevity in Zeeland sex-specific?

22The average chance to belong to the top 10% survivors in each of the size Zeeland municipalities is shown by sex in fig. 4. Fig. 4, left, shows that the chance to belong to the top 10% survivors is quite similar for women from neighbouring municipalities. Although many observations are still centred around the mean, municipalities with a standard of more than 1 from the mean now neighbour each other. Municipalities with above-average chances to be-long to the top 10% survivors are located on Schouwen and west Zeeuws-Vlaanderen, whereas municipalities with below-average chances to belong to the top 10% survivors are located in southern Zuid-Beveland. A Moran’s I of 0.29 indicates that longevity clustered spatially for women in Zeeland.

Fig. 4

Clustering of individual chances to become long-lived in Zeeland, women and men

Clustering of individual chances to become long-lived in Zeeland, women and men

23There is little indication that longevity also clusters spatially for men. Municipalities where an unusually high number of men belongs to the top 10% survivors are more spread over the island. Areas with below-average chances to belong to the top 10% survivors seem to be missing altogether, whereas there is a small group of municipalities with above-average chances to belong to the top 10% survivors in western Walcheren. The insignificant Moran’s I of 0.09 underlines that there is little evidence of spatial clustering of longevity for men.

Is the spatial clustering of longevity in Zeeland stable over time?

24Now that we have established that there is significant clustering for women, but not for men, we test if the clustering of longevity is stable over time. For both sexes, we estimated whether the chances to belong to the top 10% survivors in neighbouring municipalities resembled each other more when we excluded mortality between ages 0-5, 0-20, 0-30, and 0-50. The outcomes of this procedure are shown in tab. 1.

25When we estimate chances to belong to the top 10% survivors from birth, we retrieved a Moran’s I of 0.29 for women. If we estimate the chance to belong to the top 10% survivors from ages 5 or 20, the Moran’s I increased slightly to 0.33. Estimating the chance to belong to the top 10% survivors from age 30 or 50 marginally increased the clustering of longevity further to 0.35. Except for the first 5 years of life, clustering of longevity, thus, seems to be stable for women.

26When we estimate the male chance to belong to the top 10% survivors from birth, the retrieved Moran’s I is 0.09. This spatial clustering hardly changed if we used a different interval. Estimating male chances to belong to the top 10% survivors from age 5, 20, 30, or 50 caused the Moran’s I to vary between 0.08 and 0.12. Spatial correlations were all insignificant. Thus, also for men, the spatial clustering of the chance to belong to the top 10% survivors seems to be stable over time.

Which environmental factors affected longevity in Zeeland?

27In tab. 2 and 3 we explore to what extent the majority religion, dominant soil type, population density, number of inhabitants, net taxable income rate, percentage of the work force in agriculture or trade, and the net migration rate can explain spatial clustering of female and male chances to belong to the top 10% survivors. For each variable, we report the logit, standard deviation, and p-value to indicate whether regression estimates had a significant positive or negative association with individual chances to belong to the top 10% survivors. The Moran’s I after exclusion of the variable is used to indicate whether the variable explains spatial clustering of longevity. For the full model we report the Moran’s I before and after inclusion of variables, and show in fig. 5 the average chance to belong to the top 10% survivors after correction for the mentioned environmental effects.

Fig. 5

Clustering of individual chances to become long-lived in Zeeland after model correction, men and women

Clustering of individual chances to become long-lived in Zeeland after model correction, men and women

28Tab. 2 shows that women living on clay had a better chance to belong to the top 10% survivors than women living on sandy clay. Moreover, controlling for soil type reduced the spatial clustering. Besides living on clay soil, living in a larger municipality increased chances to belong to the top 10% survivors for women. This effect was weaker in the first years of life and did not explain clustering. The net taxable income per capita, share of the work force working in trade, and net migration rate were all negatively associated with female chances to belong to the top 10% survivors. Moreover, all variables explained clustering from age 5.

Tab. 2

Women

0 – Top 10% 5 – Top 10% Variables N log + sd p-value Moran’s I log + sd p-value Moran’s I Religion liberal protestant 52 ref. ref. ref. ref. orthodox protestant 19 .01 (.07) .930 .05 −.04 (.06) .458 .10 roman catholic 17 .07 (.08) .374 .05 −.00 (.06) .973 .10 no dominant religion 13 .01 (.08) .915 .05 −.02 (.07) .759 .10 Dominant soil type sandy clay 75 ref. ref. ref. ref. clay 26 .16 (.07) .018 .11 .11 (.06) .050 .14 Population density < 1000 citizens p. ha. 66 ref. ref. ref. ref. 1000-2500 citizens p. ha. 28 .10 (.07) .184 .01 .07 (.06) .245 .07 > 4000 citizens p. ha. 7 −.20 (.14) .146 .01 −.14 (.12) .226 .07 Number of inhabitants 101 .06 (.04) .204 .05 .07 (.04) .060 .11 Net taxable income p.c. 101 −.03 (.04) .457 .05 −.05 (.03) .083 .16 % workforce in agriculture 101 −.06 (.05) .281 .03 −.01 (.04) .832 .08 % workforce in trade 101 −.10 (.03) .002 .15 −.07 (.03) .012 .15 Net migration rate 101 −.06 (.03) .015 .06 −.07 (.02) .001 .18 Moran’s I: 0 model .29 .33 Moran’s I: full model .04 .10

Women

30 – Top 10% 50 – Top 10% Variables N log + sd p-value Moran’s I log + sd p-value Moran’s I Religion liberal protestant 52 ref. ref. ref. ref. orthodox protestant 19 −.07 (.06) .218 .13 −.09 (.06) .110 .14 roman catholic 17 .02 (.06) .728 .13 .02 (.06) .701 .14 no dominant religion 13 −.03 (.07) .672 .13 −.02 (.07) .743 .14 Dominant soil type sandy clay 75 ref. ref. ref. ref. clay 26 .13 (.06) .029 .16 .11 (.06) .055 .16 Population density < 1000 citizens p. ha. 66 ref. ref. ref. ref. 1000-2500 citizens p. ha. 28 .07 (.06) .268 .09 .08 (.06) .173 .10 > 4000 citizens p. ha. 7 −.13 (.12) .277 .09 −.09 (.12) .438 .10 Number of inhabitants 101 .08 (.04) .029 .14 .08 (.04) .034 .14 Net taxable income p.c. 101 −.07 (.03) .036 .19 −.06 (.03) .044 .19 % workforce in agriculture 101 −.00 (.04) .954 .12 .01 (.04) .898 .13 % workforce in trade 101 −.07 (.03) .015 .16 −.07 (.03) .017 .17 Net migration rate 101 −.08 (.02) .000 .22 −.08 (.02) .000 .23 Moran’s I: 0 model .35 .35 Moran’s I: full model .12 .12

29Tab. 3 shows that for men living on clay soils also associated with better chances to belong to the top 10% survivors. Especially in the first 5 years of life, this variable explained a large share of the – insignificant – clustering of longevity. Besides soil type, the share of the work force in agriculture was positively associated with male chances to belong to the top 10% survivors. However, the variable did not explain spatial clustering. Net taxable income per capita and the net migration rate were negatively associated with chances to belong to the top 10% survivors. Of these two variables, the net migration rate explained the – insignificant – clustering of longevity for men.

Tab. 3

Men

0 – Top 10% 5 – Top 10% Variables N log + sd p-value Moran’s I log + sd p-value Moran’s I Religion liberal protestant 52 ref. ref. ref. ref. orthodox protestant 19 .16 (.06) .005 .04 .07 (.04) .094 .03 roman catholic 17 .05 (.06) .403 .04 −.06 (.05) .163 .03 no dominant religion 13 .03 (.07) .684 .04 .00 (.05) .972 .03 Dominant soil type sandy clay 75 ref. ref. ref. ref. clay 26 .17 (.06) .002 .01 .13 (.04) .002 .00 Population density < 1000 citizens p. ha. 66 ref. ref. ref. ref. 1000-2500 citizens p. ha. 28 .06 (.06) .302 −.05 .04 (.04) .311 −.02 > 4000 citizens p. ha. 7 −.18 (.12) .118 −.05 −.16 (.08) .057 −.02 Number of inhabitants 101 −.05 (.04) .162 −.08 −.03 (.03) .262 −.03 Net taxable income p.c. 101 −.04 (.03) .157 −.08 −.05 (.02) .031 −.02 % workforce in agriculture 101 .04 (.05) .376 −.07 .08 (.03) .013 −.02 % workforce in trade 101 −.06 (.03) .041 −.00 −.04 (.02) .078 .01 Net migration rate 101 .02 (.02) .310 −.07 −.04 (.02) .032 −.00 Moran’s I: 0 model .09 .10 Moran’s I: full model −.06 −.02

Men

30 – Top 10% 50 – Top 10% Variables N log + sd p-value Moran’s I log + sd p-value Moran’s I Religion liberal protestant 52 ref. ref. ref. ref. orthodox protestant 19 .05 (.04) .251 .02 .04 (.04) .320 .02 roman catholic 17 −.02 (.05) .697 .02 .00 (.05) .936 .02 no dominant religion 13 .01 (.05) .789 .02 .02 (.05) .708 .02 Dominant soil type sandy clay 75 ref. ref. ref. ref. clay 26 .14 (.04) .002 .03 .11 (.05) .013 .02 Population density < 1000 citizens p. ha. 66 ref. ref. ref. ref. 1000-2500 citizens p. ha. 28 .03 (.05) .516 .00 .03 (.05) .594 .01 > 4000 citizens p. ha. 7 −.15 (.09) .084 .00 −.12 (.09) .202 .01 Number of inhabitants 101 −.03 (.03) .301 −.00 −.03 (.03) .303 −.00 Net taxable income p.c. 101 −.07 (.02) .006 .01 −.07 (.02) .009 .01 % workforce in agriculture 101 .07 (.03) .037 .01 .06 (.03) .054 .01 % workforce in trade 101 −.03 (.02) .224 .02 −.02 (.02) .320 .02 Net migration rate 101 −.05 (.02) .005 .04 −.05 (.02) .009 .05 Moran’s I: 0 model .09 .08 Moran’s I: full model .01 .01

Religious communities

31Membership of close-knit religious groups is thought to stimulate survival, as social control enforces more temperate lifestyles and stimulates networks of care and social support (Fraser, 1999; Hamman, Barancik, Lilienfeld, 1981; Kok, 2017). However, we found no relation between the majority religion in a municipality and an individual’s chance to become long-lived. Inhabitants of municipalities in Zeeland with a Liberal Protestant, Roman Catholic or Orthodox Protestant majority all had the same chance to become long-lived. These findings contradict an earlier study on the relationship between survival and religion in the Netherlands, where Liberal Protestants had a shorter later-life expectancy than members of any other religious denomination (Kok, 2017).

32There are four reasons why we might have missed this association between religion and individual chances to become long-lived. First, being Liberal Protestant meant something different in Zeeland than in the rest of the Netherlands. In the Netherlands, Liberal Protestant churches were highly secularised and enforced little control on their religious communities. In Zeeland, however, most of the Liberal Protestant churches were Ethisch Hervormd [Ethical Reformed] and, as such, not only embraced modernity and new cultural practices, but also valued tradition and the authority of the Bible (Knippenberg, 1992). As such, Liberal Protestant churches in Zeeland might have diverged less from the other denominations in terms of social control than elsewhere in the Netherlands. Second, because effects of belonging to a religious denomination were not measured on the individual level, we cannot rule out that religion had no effect on an individual’s chances to become long-lived. It might be that the effect of belonging to a Liberal Protestant Church was too small to be picked up by our analyses. Third, differences between religious denominations in survival might exist between regions, but not within regions. In such cases, differences in survival are caused by regional cultures rather than religious communities. Finally, we only observed the individuals who stayed in Zeeland. It might be that those with weak ties to their religious communities chose to leave Zeeland altogether, so that amongst the stayers few individuals remained with weak ties to their religion. Therefore, additional individual-level studies are necessary to understand how religion affected human survival.

Environmental conditions

33Soil type has historically been associated with different disease environments and access to clean drinking water (Hofstee, 1981). However, inhabitants of clay soils had a better chance to become long-lived than inhabitants of sandy clay, even though the clay soils were excellent breeding grounds for malaria mosquitoes. Most likely, the inhabitants of the clay soils lived longer, due to how the land was used (Devos & Van Rossem, 2017). Agriculture was less common on the clay soils, because the high salinity and acidity of the soil made it hard to grow crops. Hence, farmers held more stock. Just like in Holland, farmer’s wives took this opportunity to produce cheese. This may have provided these households with some extra income (Van Cruyningen, 2005), but more importantly, cheesemaking gave women a chance to perform economically valued labour in an economic niche. An extensive study on 19th century excess female mortality in Britain showed that the dairy economy in Devon and Cornwell also strongly correlated with increased female survival, indicating that female chances to work in an economic niche were probably larger in rural regions that focused on dairy farming rather than agriculture (McNay, Humphries, Klasen, 2005). Furthermore, farmers on clay soil worked less land than the farmers on sandy clay. This means that the ratio between farmers and farm workers is lower (Priester, 1998), so that more people had their own land which they could work on. Farmers were more secure of income than farm workers, which might have benefited their survival.

34The agricultural orientation was also very important. Living in a wealthier municipality seemed to be detrimental for men as well as women, due to the higher inequalities in the municipality. The wealthier municipalities were not the cities, but transport hubs or farmland near Goes and Middelburg. Most likely a focus on cash crops and large-scale farming, meant that a smaller number of farmers employed a larger number of farm labourers. These farm labourers did not have their own land to build and were more dependent on the market for access to food. The negative effects of cash-crop and large-scale farming were similar for men and women, which indicates that it did not affect female occupational opportunities worse than other agricultural regions. Rather, less market integration generally meant that more food was produced and remained in the region, so that more food was also available to everyone in society (Komlos, 1998). Hence, the association between soil and human survival might be dependent on an individual’s chances to have one’s own farm and access to food in general, which assured better access to food for both men and women.

Rural-urban differences

35City size had a positive influence on a woman’s chances to become long-lived, whereas no association was found for men. Reversely, a strong focus on agriculture seems to be good for men, while it has no effect on female chances to become long-lived. This is in line with the idea that the countryside was healthy for men as they did not work in poorly ventilated workshops and spent less time in the pub, whereas cities were healthy for women due to more positive gender roles and increased chances to perform economically valued labour (Humphreys, 1991; Janssens, Messelink, Need, 2010; Mc-Nay, Humphries, Klasen, 2005; Reher, 2001). However, our findings contradict urban mortality penalties that have been found for cities in the Netherlands (Drukker & Tassenaar, 1997; NMBG, 1879) and other countries in Western Europe (see e. g. Eggerickx & Debuisson, 1990; Kesztenbaum & Rosenthal, 2011; Vögele, 1998). There are two explanations why we do not find evidence of an urban penalty in Zeeland. First of all, Zeeland never urbanised as strongly as other Western European regions. Rising metropoles like Rotterdam or Antwerp were just outside the provincial borders and the larger town in Zeeland did not suffer from the negative effects of overcrowding which came with large levels of immigration. Hence, municipality size in Zeeland was not related to the strong negative effects that larger cities in Western Europe experienced. Second, the positive association between living in a larger town and individual chances on becoming long-lived were only found after controlling for soil type, net taxable income per capita, and the percentage of traders in the work force. This is in line with more recent studies on mortality after age 50 in Utah and Québec (Gagnon et al., 2009; Temby & Smith, 2014). These studies showed that after controlling for other factors, living in a larger town was beneficial for survival. This is in line with the debate on human stature, which found evidence of both urban penalties and urban premiums (see e. g. Baten, 1999; Martínez-Carrión & Moreno-Lázaro, 2006; Mironov & A’Hearn, 2008; Reis, 2009; Steckel, 1995; Tassenaar, 2019). In the Netherlands, these urban premiums were available for small-and medium-sized cities (Tassenaar, 2019). Therefore, we conclude that living in larger towns also had its benefits, which has received too little attention in historical research due to our focus on the urban mortality penalty.

36Individuals from municipalities with net outmigration had a higher chance to become long-lived. It is unlikely that the negative relationship between the migration rate and chance to become long-lived is attributable to compositional effects, as migrants on average live longer than stayers (Alter & Oris, 2005; Puschmann, Donrovich, Matthijs, 2017; Van den Berg et al., 2020). Moreover, in Zeeland immigration and emigration rates themselves are not significantly correlated to individual chances to become long-lived (correlations are −.11 and −.10, respectively). Most likely the association between the net migration and individual chances to become long-lived are attributable to demographic pressure. The net migration rate distinguishes between regions with little migration, some outmigration, and more outmigration. There was no room for spatial or demographic expansion in Zeeland, as there was no land acquisition during the 19th century and the economy did not grow. Outmigration from Zeeland relieved the demographic pressure on Zeeland and left better economic opportunities for those who stayed behind. Moreover, it is to be expected that the individuals who left were least likely to attain a good position in society. Therefore, small levels of net outmigration already had a notable effect on local communities.

37Our analyses also showed that trade has a negative effect on female survival. The more traders there are in a municipality, the lower the chance for someone to become long-lived. It is not possible to determine whether this effect is also present for men, as effects of farming and trade are multicollinear. The positive effect of living in a town with many farmers in the work force is so strong and robust for men that they negate the possible effects of the share of traders in the work force. Nevertheless, the share of traders in the work force had a significant correlation with individual chances to become long-lived, indicating that there might also be a connection between trade and male chances to become long-lived.

38There are multiple explanations why the share of traders in a municipality correlates with (female) chances to become long-lived. First, the share of the workforce in trade might be indicative for effects of urbanisation in general, as the variable identifies the municipalities that function as regional centres on the Zeeland islands. However, it is unlikely that urbanisation in Zeeland decreased individual chances to become long-lived. Zeeland never really industrialised in the 19th century and most of the pockets of industrialisation were focused on mechanisation of the countryside, rather than the cities (Zijdeman, 2010). Furthermore, little overcrowding occurred (Hoogerhuis, 2003) and our analyses showed that population density had no effect on offspring survival. Moreover, because the share of traders in a province explains a significant part of longevity clustering, it is unlikely that it is a proxy for another environmental effect that was also included in our regression models. Second, it might be that the share of the workforce in trade is a proxy for work in another sector. Especially work in the cottage industry has been shown to be detrimental for survival (Van Rossem, Deboosere, Devos, 2017). However, a closer inspection of our data – as shown in tab. 6 – shows that the share of workers in labour is not correlated to work in any other major sector in Zeeland. As such, a municipal focus on trade cannot be a proxy for work in another economic sector. A third explanation, is that the most unfortunate worked in trade as a last resort to secure oneself of some income. These individuals might have been hucksters, peddlers, and petty merchants that catered to the rural populations on the island. In Zeeland, traders mainly lived along the canals on Walcheren or in the island centres (see fig. 13), which were ideal locations to buy goods and later sell them in the surrounding countryside. These small traders are known to have been insecure of food and income. Poverty might explain why living in a place with traders is bad for survival to exceptional ages. Earlier studies have shown that small entrepreneurs show higher mortality in later life (Edvinsson & Broström, 2012; Mourits, Smith, Janssens, 2019). Therefore, the effects of trading on the chance to become long-lived are most likely a composition effect.

39To substantiate our hypothesis, we took a further look at the 1930 census from which we derived our information on the share of traders in the work force. Aggregate information on the number of owners, managers, foremen, and labourers working in trade were available on the provincial level. Tab. 7 shows that traders usually worked alone, although some also had personnel. This is in favour of our argument that most of the traders were petty merchants. The 1930 census further included aggregated information on what the Zeeland traders sold. As shown in tab. 8, 40% of trade occurred in wholesale, distributive trade, and retail. This is a wide range of activities from which it is hard to conclude who these traders were. However, another 40% of the traders worked as a peddler, grocer, or cloth salesman. None of these individuals produced their own goods and instead sold goods of relatively low value. The low profits that could be gained on these products probably made these individuals among the poorest in Zeeland. Hence, living in a town with a focus on trade might be synonymous with living in a town with increased levels of poverty. When a sizable share of the work force was a huckster, peddler, or petty merchant – and, thus, lived in poverty – it is to be expected that this had its effects on the overall chances to become long-lived.