Why were infants dying and what were they dying from? Infant mortality patterns in the Greek urban centre of Hermoupolis, Syros (1860–1940)

ABSTRACT

The reduction in infant mortality has been a significant component of mortality decline in all north-western European populations. Infant mortality in Greece has been studied before, though most rates refer to rural populations or short periods of time; while the national ones have been based on multiple assumptions. Only rarely there is available evidence about the evolution of infant mortality in urban Greece in a long-term perspective. This paper, therefore, fills this gap by employing individual-level data, a rare collection of oral histories and qualitative sources from the major urban centre of Hermoupolis, on the Greek island of Syros, for the period 1860–1940. Infant mortality in Hermoupolis was found to be among the highest in the country for most of the study period. Even though it had been argued that infant mortality in Greece declined in the 1930s, Hermoupolis experienced an earlier decline, situated in the late 1890s. Main factors that were found to be related to this decline include wider access to water, changes in the registration system, fertility decline, improvements in living standards and nutrition among lower strata infants and improvements in maternal literacy. Diarrhoeal diseases killed most infants especially during the hot and dry summer months. Despite the widespread practice of breastfeeding in the city, seasonality analysis indicated the early initiation of supplementary food. This paper contributes to the existing literature by extending our understanding of the factors that facilitated the reduction of urban infant mortality beyond Western Europe and North America.

KEYWORDS:

• Nominal record-linkage
• infant mortality
• diarrhoeal diseases
• urban
• Greece
• Hermoupolis

1. Introduction

Infant mortality has been an important component in the study of mortality in the past mainly due to its high levels in pre-transitional demographic regimes (Oris et al., 2004). Falls in early childhood mortality, but mainly infant mortality, were responsible for the mortality decline in the late nineteenth and early twentieth century in Europe, by significantly contributing to the rise of life expectancy and having an indirect impact on fertility decline, although the latter has been challenged in the literature (Reher, 1999, 2004; Schofield & Reher, 1991). The significant differences in the levels and the timing of the decline that existed among most European countries have been attributed to socio-economic, cultural, geographical and environmental factors (Pozzi & Ramiro-Fariñas, 2015).

Previous studies indicate that demographic transition in Greece followed the model, according to which mortality decline preceded that of fertility (Hionidou, 1993, 2007; Valaoras, 1960). Although the timing of the mortality decline has been established to have started at the turn of the century, that of fertility has been debated in the literature. Siampos and Valaoras (1969) situated the onset of the fertility decline in the late nineteenth century, while Hionidou (2020), by resituating the timing of the secular decline of fertility in Greece, recently argued that fertility decline started in the mid-1920s for the urban upper and middle class and in the early 1930s for the rest. Finally, by estimating age-specific fertility rates, Hionidou (1998) and Gavalas (2002) showed that fertility decline began in the late 1920s on the Cycladic islands of Mykonos and Paros.

Although mortality patterns in Greece have been studied both at the national and local level, most rates have been based on indirect techniques and multiple assumptions in order to fill the missing data (Siampos, 1973; Valaoras, 1960; for a review, see Hionidou, 1997b), refer to rural or semi-urban areas or cover short periods of time (Bournova, 2016; Dimitropoulou, 2008; Eliopoulos, 2011; Gavalas, 2001, 2008; Hionidou, 1993, 1997a; Kanellakes, 1955; Moustane, 2014). The discontinuity in the publication of published statistics in Greece were mainly responsible for the lack of studies addressing mortality in the past. The only continuous national infant mortality estimates were constructed by Valaoras (1960). According to his figures, infant mortality rate (IMR) declined continuously from about 200 infant deaths per 1,000 births (‰) in 1860 to 170‰ in 1900 and continued to exhibit the same pattern in the early twentieth century, reaching about 110‰ in the late 1930s. Urban infant mortality was considerably higher than that in rural areas even in the first decades of the twentieth century partly due to extensive under-registration and age-misreporting especially in rural settings (Valaoras, 1988, p. 428); yet, it has been argued that the gap between the two rates had narrowed significantly by the 1930s, as the quality of the published statistics improved substantially (Kanellakes, 1955; Tsaouses, 1971).

The lack of continuous published statistics and the poor quality of civil registration in most Greek populations until the mid-1920s have been great obstacles to studying infant mortality at the national level. This paper, therefore, examines urban infant mortality in Greece by employing individual level civil registration data from the urban centre of Hermoupolis, on the island of Syros. Hermoupolis possesses unique data sources by Greek standards: civil registration for the entire population continuously available from 1859 and a wealth of qualitative sources. Record linkage techniques were used to examine the quality of the data sources and to detect the extent of under-registration. The paper addresses questions regarding the timing of the onset of infant mortality decline, its rate of decline and the factors that facilitated such decline. Seasonality indexes were calculated for the entire period as well as cause-specific death rates for the period 1916–1940 in order to explore the underlying determinants that affected the survival of infants. The findings of this paper will provide continuous urban infant mortality rates for one major Greek city and thus put Greece on the map when studying infant mortality at the European level.

This paper contributes to the existing literature by extending our understanding of the factors that facilitated the reduction of urban infant mortality beyond Western Europe and North America, which have been almost exclusively the focus of previous historical studies of urban mortality in the last two centuries. Finally, this paper analyses the possible impact of urban sanitary reforms (i.e., introduction of water supply) on infant mortality in Greece, an issue that has rarely been explored.

2. Infant mortality decline in an historical context

Infant mortality started its most significant fall in most of the north-western European countries in the second half of the nineteenth century, although there were noteworthy differences in the rates and variations in the timing of the decline (Oris et al., 2004; Schofield & Reher, 1991). Geographical variations, especially between northern and southern European countries, were attributed to climatic differences, as much warmer temperatures in southern Europe amplified diarrhoeal incidence among infants (Caselli, 1994; Oris et al., 2004). In Italy, in particular, the noteworthy regional variations between northern/central and southern/island territories have been attributed mostly to environmental (extremely low temperatures increased neonatal mortality), socio-cultural factors (prolonged breastfeeding in Sardinia mitigated gastrointestinal mortality) and introduction of new health policies (Breschi et al., 2016; Del Panta, 1997; Pozzi, 2002; Scalone & Samoggia, 2018). Infant mortality was also greater in urban rather than in rural populations (Williams & Mooney, 1994; Woods et al., 1989). Urban environments in the nineteenth century faced an ‘urban penalty’, which reduced survival probabilities for their inhabitants and resulted in a great number of deaths, most significantly among young ages, due to high population density, lack of improvements in public health infrastructure and dissemination of epidemic diseases (Kearns, 1988; Reher, 2001).

Most studies attributed the decline in infant mortality to various socio-economic, cultural and environmental determinants and public health movement. McKeown (1976) proposed that improved living standards and nutrition influenced significantly the onset and course of mortality transition. Other studies indicated that maternal education shaped infant mortality decline by improving the status of mothers, access to information and child-care practices (Caldwell, 1979; Woods et al., 1989). The effect of various infant-feeding practices on infant survival has also been reported in the literature (Thorvaldsen, 2008). For instance, exclusively breast-fed infants had much greater survival chances than those who were exclusively or partially artificially fed (Knodel & Kintner, 1977). Many studies also explored the effect of public health policies and medical interventions, which aimed to improvements in milk supply, advanced antenatal and postnatal assistance and greater standards of maternal and infant care (Lucia Pozzi & Barona, 2012; Morel, 1991).

The largest part of the most recent literature, however, has attributed much of the infant mortality decline to successive sanitary reforms. Various studies attributed significant declines in infant and childhood mortality in Europe but also the US to the introduction of sewerage, and water supply or water filtration and chlorination systems during the late nineteenth and early twentieth century (Alsan & Goldin, 2019; Cutler & Miller, 2005; Floris & Staub, 2019; Macassa et al., 2006; Peltola & Saaritsa, 2019). A few studies also showed that the effect of water supply and sewerage systems in infant and childhood mortality decline were complementary to each other rather than substitutes (Alsan & Goldin, 2019; Gallardo–Albarrán, 2020; Helgertz & Önnerfors, 2019). Decline in fertility is the final factor that has been found to have played an important role in infant mortality decline (Woods et al., 1988, 1989), although has been challenged in the literature (Reher, 1999, 2004; Van De Walle, 1986).

Within the Greek context, by studying the population of Athens in the first half of the twentieth century, Bournova (2016) suggested that infant mortality decline in the 1930s resulted from improvements in public health and medical care. Hionidou (1993, 1997a), on the other hand, attributed infant mortality decline to the decline in fertility rather than to the introduction of public health measures, as they were rather virtually non-existent in rural Greece before the 1950s. Other studies connected infant mortality decline with improvements or changes in the registration system (Bournova, 2016; Eliopoulos, 2011; Gavalas, 2001, 2008; Kanellakes, 1955).

3. Setting: Hermoupolis, Syros

Hermoupolis, the capital city of the island of Syros and of the Cycladic group of islands, was a significant Greek urban centre (with a population of 21,540 in 1879) and a major industrial centre. It was by far the most important port of modern Greece and the largest centre of transit trade in the Eastern Mediterranean Sea in the mid-nineteenth century (Delis, 2015). The city’s importance started to decline in the late 1870s, although it maintained its prominence until the opening of the Corinth Canal in 1893 (Travlos & Kokkou, 1980). Hermoupolis then entered a long phase of decline, which led to a significant loss of inhabitants (the population had fallen to 18,760 in 1896). Hermoupolis, therefore, was relegated to a provincial city by the beginning of the twentieth century, albeit with extensive administrative responsibilities as the capital of the Cyclades Islands. In the 1920s, the population of the city increased significantly (reaching 21,416 in 1928) due to the arrival of the Greek Orthodox refugees from Asia Minor, but by 1940 it had declined again to the pre-1920s level (18,925).

The majority of the inhabitants of the island of Syros were Catholic, except for Hermoupolis, where almost all the inhabitants were Greek Orthodox. Catholic inhabitants were registered in the Ano Syros Registry Office, and therefore their deaths were reported there even if they lived in Hermoupolis, as they were buried in the Catholic cemetery in Ano Syros. The town of Ano Syros is situated very close to Hermoupolis, but despite such proximity, the two settings had entirely separate administrations.

Hermoupolis experienced very high mortality levels in the second half of the nineteenth century with a crude death rate (CDR) of 32‰ in 1896, much higher than the national average (22–25‰) at the time (Valaoras, 1960). Kolodny (1969) argued that high mortality in Hermoupolis resulted from the unhealthy working conditions and very low living standards among the working-class residents. Lack of public hygiene, high in-migration, and high population density in the city have been also associated with high mortality levels (Raftakis, 2019). By the late 1920s, CDR had reduced to 21‰, reaching levels much closer to the national average, as estimated by Valaoras (1960) and Siampos (1973).

4. Public health

Greek governments rarely undertook initiatives to improve public health in the country (Copanares, 1933). Instead, local authorities had to take care of sanitary reforms and healthcare. Public hygiene was found to be poor in Hermoupolis despite the initiatives of the local authorities (GSAS/DA/PDS 23/12/1857; 9/11/1859). After its creation in the 1820s, the city had limited space to expand, since the distance between the sea front and the hills was very limited and therefore was cramped around the port. Most people in Hermoupolis, according to a contemporary physician, lived in damp, dark dwellings, whereas upper and middle classes lived in healthier houses (Tsakalotos, 1914). Public health reports stated the unhealthy conditions of the Hermoupolis housing up to the early-1930s (GSAS/YPA/YP917/1931).

Hermoupolis had a hospital as early as 1827¹, the largest quarantine area in Greece by 1842 and a strong association of physicians from the mid-nineteenth century. Even though the local government attempted to improve public hygiene, the lack of a water supply system, the inefficient sewage disposal, high population density, bad housing and the very poor living standards among the working classes resulted in frequent outbreaks of infectious diseases and the prevalence of diarrhoea-related diseases (Fragkides, 1894). Regular smallpox and diphtheria vaccinations were implemented by the local authorities from the 1880s, although smallpox vaccination was made compulsory as early as 1835. Such practice, which carried on throughout the study period, has been argued, had great impact on the evolution of mortality in the city (Raftakis, 2019).

Residents of Hermoupolis got their water mainly from artesian wells and private water cisterns that collected rainwater. Many inhabitants also relied on public springs for potable water (interviewees #10,13). Nonetheless, according to a contemporary report, all types of water were contaminated in 1933 (Alivizatos, 1933; Pagkalos, 1933). An underground water supply system was constructed only in the mid-1920s. Oral accounts, however, refer to restricted access to the central water supply even until the mid-1930s (interviewees #2,9). Hermoupolis had acquired a sewage disposal system by the mid-nineteenth century (its construction started in 1848 and it was one of the oldest in the country), although very limited improvements took place until the late 1930s. The sewerage system actually consisted of public open halls, situated on the main streets, where the inhabitants emptied their pots, and flowed via central sewers into the sea (Raftakis, 2019). According to contemporary physicians in 1895, the existing sewers were totally ineffective and their complete absence would actually have been beneficial for the hygienic conditions of the city (Iatrike Etaireia Syrou, 1895).

5. Data and methods

Two different types of archival material were employed for the purposes of this study: civil registration certificates preserved in the General State Archives of Syros (GSAS hereafter) and Hermoupolis Registry Office (Leksiarchika Vivlia Apovioseon, Leksiarchika Vivlia Genneseon), and a wealth of qualitative sources (including, local press, public health reports (Ygeia & Pronoia Archeio, YPA) and a rare collection of oral interviews²). Civil registration certificates have been available in the city continuously since 1859 and cover the entire population of the city. Hermoupolis data is of unique quality by Greek standards. Registration in Hermoupolis, unlike other areas in Greece, did not start in 1859, the year of the creation of the civil registration system but commenced in the 1820s (for example, the city kept municipal registers since its creation in 1822). Hermoupolis, therefore, had a long-established tradition of keeping a wide range of sources.

All death certificates for the period 1859–1940 were transcribed into the Hermoupolis Mortality Database (HMD hereafter) amounting to 45,247 records (Table 1).³ Each death certificate contains information about the informant as well as about the deceased person (full name, age in days, weeks or months for infants, marital status, place of origin, occupation, relationship to the informant, mother’s full name, cause and place of death from 1916 onwards, date of death and reporting). The death certificates of those infants who had died at the age of 0–15 months around each census year were linked to their respective birth records using the technique of nominal record-linkage (Williams, 1989).⁴ The record-linkage involved two stages: first, linking the death records with the birth records; and second, linking the death records to the Foundling Manuscript Books (Vivlia Ektheton) in order to obtain the necessary information for those foundlings whose birth record was either missing or was not linkable. For some years, the registration of the death of foundlings was reported at the end of each month or even year retrospectively by the foreman of the local hospital in an appendix to the death manuscript books, otherwise it was stated on death certificates.

Table 1. Total number of infant deaths and live births and their mean per annum, Hermoupolis,1860–1940

	N (records)
Total number of infant deaths	10,108
Mean per annum	124,8
Total number of live births	50,794
Mean per annum	627

Source: HMD

Of 1,153 infant deaths recorded in the civil registration in the census years from 1861 to 1940, 869 were successfully linked to a corresponding birth record giving an overall linkage rate of 75% (Table 2). Comparatively speaking, the linkage rate in the nineteenth century was not significantly different from the one undertaken (about 80%) by Williams (1989), and somewhat lower than those in Revuelta-Eugercios and Ramiro-Fariñas (2016) and Jaadla and Puur (2016) studies, 88% and 89%, respectively. The trend of increasing successful linkage rate (from 64% in 1861 to 92.6% in 1940) is important evidence to suggest that the civil registration system gradually improved.

Table 2. Successful linkage ratio (%) by age for every census year, Hermoupolis, 1861–1940

N (days)	1861	1870	1879	1889	1896	1907	1920	1928	1940
0 days	0.0^4	–	0.0^1	0.0^1	–	0.0^3	50.0^2	33.3^3	50.0^2
1–5 days	0.0	84.2	85.7	54.5	61.5	55.6	62.5	76.9	100.0
6–10 days	0.0	66.7	57.1	22.2	80.0	100.0	100.0	83.3	100.0
11–28 days	50.0	61.5	93.3	54.5	86.7	100.0	83.3	81.8	100.0
29–364 days	74.0	78.2	74.2	65.1	81.1	91.2	83.3	81.9	92.7
Successful linkage rate	64.3	76.4	75.6	60.3	79.7	87.9	81.6	80.2	92.6

¹one death recorded; ² two deaths recorded, ³ three/four deaths recorded; ⁴ nine deaths recorded

Source: HMD

To a great extent, the failure to link a higher percentage of infant deaths (a quarter of all unlinked cases) is attributed to the high number of foundlings. The linking of their deaths was often virtually impossible, as very little information was available from either source. Hermoupolis experienced high foundling mortality during the second half of the nineteenth century due to the existence of a public hospital in the city, which accepted foundlings and placed them with hired wet-nurses (Arfanes, 1900), as found also in other studies (Ramiro-Fariñas, 2007; Revuelta-Eugercios & Ramiro-Fariñas, 2016). Loukos (1994) estimated that 1,426 foundlings were abandoned in the local hospital during the period 1873–1910, with half of them dying within the first year of life. Under-registration or mis-registration of the foundlings in urban areas in the past has been a very common problem which researchers need to overcome in order to explore the actual levels of infant mortality. By simply dividing the annual number of foundling infant deaths to annual number of births, it was calculated that foundling mortality accounted for a relatively substantial fraction of infant deaths, while it started to decline from the 1910s onwards (Figure 2).

Given that Hermoupolis was the major urban area of Syros, it is possible that mothers from other parts of the island might have abandoned their infants in the Hermoupolis local hospital. The pregnant woman would have to travel on foot to Hermoupolis, give birth on her own there and then deposit the newborn baby. Alternatively, she would give birth secretly in her place of origin and then travel on foot with the baby to Hermoupolis. As the sole purpose of disposing the foundling was to retain her anonymity and conceal her association with the illegitimate child either of the above scenarios are highly unlikely. In both cases, the woman would come across acquaintances who would ask questions. Oral evidence also indicates that the Catholics of Ano Syros left unwanted babies in or near Catholic churches in Ano Syros. The Catholic priest then judged what would happen to them, allocating them to Catholic families (interviewee #9). It is more probable that women who could afford to do so, did travel from Syros to bigger urban centres such as Athens or Piraeus in order to retain anonymity and to avoid the social stigma. The hypothesis that mothers had travelled from other Cycladic islands to Hermoupolis in order to abandon their children is highly improbable, as in most cases women would not travel to Syros since anonymity could never been ensured there. Those who could afford such a trip, they would have travelled to Athens for similar reasons as their counterparts from Syros.⁵

One of the most important source problems when studying demographics of historical societies is the under-registration of infant deaths but also that of births. Under-enumeration of the number of births could result in an inflated estimate of infant mortality, while that of deaths in a deflated estimate. In Hermoupolis, a very likely ‘loss’ was the under-registration of births or deaths of those infants (irrespective of their sex) who died within a few hours or even days of birth, given that parents may have seen no reason to register them (Hionidou, 1993). The percentage of unlinked deaths gives a broad indication of the quality of the data. Certainly, in the first half of the period, under-registration was most likely. The fact that deaths within 24 hours after birth (Table 2) were registered already since 1861 is an important indication regarding the limited levels of under-registration of infant deaths. On the other hand, the failure to link any deaths that occurred within 24 hours after birth until 1907, suggests that under-registration of births was more likely for those deaths occurring within the first day or days of life (up to 5 days, see also linkage-ratio by age in Table 2). Evidence from neighbouring Mykonos indicated that when infants survived more than a few hours, a hastened baptism would take place and a proper burial would follow.⁶ Thereby, the registration of the death was ensured. In addition, oral evidence showed that stillbirths were not reported either as deaths or as births, since parents buried them informally rather than in the municipal cemetery (Hionidou, 1997a, p. 158). Since stillbirths in Hermoupolis were not registered prior to 1912, it is hypothesized that a similar pattern could also take place there. In any case, stillbirths were ignored in the linkage exercise and therefore were excluded from the calculation of all rates. In Hermoupolis, systematic registration of stillbirths took place only from 1912 onwards (Raftakis, 2019). An obvious problem of course is that the mis-classification or mis-reporting of stillbirths and early neonatal deaths can cause false estimations of infant mortality. Prior to 1912, therefore, it was not possible to identify the stillbirth levels, a standard issue in historical demography, as only rarely stillbirths were reported in civil registration prior to the twentieth century in most European populations (Woods, 2009).

Although birth and death under-registration is considered to be less important from 1896 onwards, death registration was more complete than that of births throughout the study period, as the presence of a priest at a burial was mandatory, while the burial could not be conducted without a death certificate (Stefanos, 1884). The Greek legislation dictated that every death must be reported within 24 hours of the event, unlike a birth which had to be reported within 5 days (or even 10 days after 1924) of the delivery. Therefore, there are no serious indications of under-registration of infant deaths, while it seems that the registration of births improved over time. Unlike Hermoupolis, under-registration of both births and deaths in many rural and urban settings in Greece was found to be especially high until in 1925, when civil registration properly commenced (Eliopoulos, 2011; Gavalas, 2001; Kanellakes, 1955; Siampos & Valaoras, 1969; Valaoras, 1960, 1988).

Migration has been an important factor in increasing urban mortality. Although Hermoupolis population experienced high in- and out-migration, especially in the nineteenth century – a standard issue in the study of urban demography – it is argued here that migration had a minimal impact on the levels of infant mortality in the city (apart from the 1923 refugee influx). Infants very rarely migrated. Young single males were more likely to migrate in and out of the city (Hionidou, 1999), while oral evidence suggested that wet-nurses employed by affluent families were coming to the city from the neighbouring islands (interviewees #1,13; Hionidou, 2005).

To examine the quality of the Hermoupolis sources, to detect the extent of under-registration and to correct the figures from the effect of such under-registration and age-misreporting, four rates of infant mortality were calculated for every census year: Reported (a) and Linked IMRs (b-d) (Table 3). For the calculation of the Reported Rate (a), the annual number of reported infant deaths and births were used, while for the calculation of the Linked Rate (b), dummy births were created for all cases of registered infant deaths for which there was no recorded birth certificate and added to the annual number of births (see also Table 3). When the two rates are compared, they seem to follow very similar patterns, although there are some disparities in the second half of the nineteenth century. The Reported IMR (a) is somewhat higher than Linked Rate (b) because of age-misreporting at death. In the case of infants, age-rounding up to age 1 was common. To reduce the net error of age-misreporting, the exact number of surviving days was calculated by linking the recorded date of birth to that of death. From 1920 onwards the discrepancies between the two rates had become minimal.

Table 3. Reported and Linked IMRs for every census year, Hermoupolis, 1861–1940

	IMR (a)	IMR (b)	IMR (c)	IMR (d)
1861	201.3	195.0	119.3	–
1870	190.4	188.0	141.0	–
1879	166.2	181.0	126.4	156.3
1889	234.6	223.7	170.0	221.0
1896	251.8	248.3	199.2	239.4
1907	264.1	273.5	211.6	255.3
1920	148.3	150.2	123.7	–
1928	185.1	183.2	158.1	–
1940	130.8	127.0	117.8	–

(a)all infant deaths with reported age>1 year/365 days divided by annual number of reported births.

(b) infant deaths linked to a birth record/Foundling Manuscript Book with linked age>365 days and remaining/unlinked deaths with reported age>1 year/365 days divided by annual number of reported births and dummy births for all unlinked cases.

(c) non-foundling infant deaths linked to a birth record with linked age>365 days and remaining/unlinked non-foundling deaths with reported age>1 year/365 days. The number of births has been adjusted to the linkage rates for non-foundling infants (see also main text).

(d) all infant deaths with reported age>1 year/365 days divided by annual number of reported births and dummy births for all foundling infants that were registered to the Foundling Manuscript Book (Loukos, 1994, p. 249, Figure1).

Source: HMD

However, if we assume that many deaths were unlinked to a birth record due to significant under-registration of births, the linkage rate for non-foundling infants should reflect the completeness of birth registration, and therefore the number of births could be inflated for these missing births. For this reason, Linked Rate (c) was estimated by adjusting the number of dummy births according to the linkage rate for non-foundling infants (for instance, for a linkage rate of 80% then births could be inflated by 100/80). Lastly, in order to capture the large contribution of foundling infants in the estimation of IMR, an attempt to assign dummy births to all foundling infants which were registered to the Foundling Hospital Books was made (Linked IMR (d)). Both rates are much lower than the previously estimated IMRs. The real explanation is likely to include some mix factors. Nonetheless, given that certification of birth was relatively complete except in those cases where infants died before registration (most likely soon after delivery), while it improved over time and that a significant number of foundling deaths was indeed linked to a birth record and/or the Foundling Hospital Books, Linked IMR (b) is more likely to reflect or at least be closer to the genuine levels of infant mortality in Hermoupolis.

Calculating IMR around the census years, however, did not allow us to explore the short-term fluctuations of infant mortality for specific spans of years due to the long gaps between the publications of censuses (for example, from 1907 to 1920). Therefore, the annual IMR, neonatal and post-neonatal rates along with their 5-year moving average to eliminate erratic fluctuations were calculated as well (Figure 1).

Figure 1. Annual and 5-year moving average IMR, NMR and PNMR, Hermoupolis, 1860–1940

The annual number of reported births has been used as a denominator for the calculation of all rates.

Source: HMD.

Figure 2. Annual and 5-year moving average IMR for all infants, non-foundlings and foundlings, Hermoupolis, 1860–1940

The annual number of reported births has been used as a denominator for the calculation of all rates.

Source: HMD.

Finally, mortality by cause of death was analysed using the successive schemes of the International Classification of Diseases (ICD-3 and ICD-4) for the years 1916–1940, when cause-of-death information is available. The accuracy of historical causes of death has been challenged in the literature, as often confounded by changing conceptions of diseases, diagnostic nomenclature, recording practices and nosologies. Causes of death among infants, in particular, were extremely unreliable as its largest share was attributed to the ‘undefined’ category or were commonly vague descriptions of symptoms rather than actual diseases (Alter & Carmichael, 1996, 1999; Kintner, 1986; Reid & Garrett, 2012; Williams, 1996). Despite such limitations, cause of death data, which is rarely available in Greece at the individual level, is expected to contribute significantly to the study of the underlying determinants that affected the survival of infants in the city.

6. Results and discussion

6.1. Infant mortality

Figure 1 shows the course of infant mortality along with neonatal and post-neonatal mortality rates from 1860 to 1940. Fluctuations in post-neonatal mortality levels were mainly responsible for the fluctuations of IMR, in accordance with other European experience (Oris et al., 2004; Woods et al., 1988) or that of other Mediterranean island communities (Breschi et al., 2012; Hionidou, 1997a; Pujadas-Mora, 2009). Considering that post-neonatal comprises the largest component of infant mortality, the trend of the two, is almost identical. The high levels of post-neonatal mortality therefore reflected the importance of exogenous causes throughout the study period. Although neonatal mortality comprised a smaller fraction of the overall infant mortality, its gradual increase in the last decade of the nineteenth century and its fluctuations soon after played a key role in the course of infant mortality.

Infant mortality in Hermoupolis was high throughout the study period, reflecting the trend of the high CDR. After an increase in the 1860s, probably as a result of the gradual improvement of the registration coverage, the relatively lower IMR in the 1870s was followed by a greater early-childhood mortality (ECMR) (Figure 3). Following the Mediterranean pattern, this trend of higher ECMR than IMR was observed in Spain, Italy and neighbouring Mykonos (Hionidou, 1993; Pozzi, 2002; Ramiro-Fariñas & Sanz-Gimeno, 2000b, 2000a). Yet, this pattern was also typical of large urban populations in the nineteenth century (Jaadla & Reid, 2017; Wolleswinkel-van den Bosch et al., 2000; Woods, 1993).

Figure 3. IMR, ECMR, and CBR for every census year, Hermoupolis, 1861–1940

ECMR in 1889, 1896 and 1940 has been estimated (Ramiro-Fariñas & Sanz-Gimeno, 2000a), while both ECMR and CBR have been calculated only around census years due to the availability of mid-year population

Source: HMD.

The sharp rise of IMR in the 1880s but especially in the 1890s most probably reflected the decline of the economy in the city, which in combination with the unstable financial and political situation of the country at the time (i.e., 1893 public insolvency) deteriorated the working conditions and living standards among the lower strata (see for example, Helios, 16/01/1894, 464). According to Fragkides (1894, pp. 54–62), the almost annual occurrence of epidemics and the high prevalence of diarrhoeal diseases among infants were the products of high population density and lack of public health infrastructures. The highest ever recorded IMR in 1890–91 should be associated with smallpox and whooping cough epidemics that occurred in the city and a flu epidemic (presumably he refers to the Russian flu; for recent evidence, see Ramiro et al., 2018), which afflicted two-thirds of the inhabitants in early 1890 (Fragkides, 1894, pp. 58–60). Furthermore, this rise of IMR can be partly attributed to increases in neonatal mortality and the gradual improvement in the registration coverage. Another explanation could be the impact of food price inflation (e.g., bread, see Patris, 26/05/1892, 1354, p. 3), as a result of the economic decline in the city at the time, which may have affected the nutrition of mothers and ultimately that of infants (see also Scott & Duncan, 1999; Wu et al., 2004). While IMR reached its peak towards the late-nineteenth century, ECMR had already started to decline by the late 1880s. The timing of the decline coincided with the introduction of vaccination practices, implemented by the local authorities (Raftakis, 2019); however, further research should be undertaken to investigate the impact of vaccination policies into the decline of ECMR in Hermoupolis but also elsewhere.

Subsequently, IMR appears to have fallen fairly continuously from the late 1890s onwards; yet, almost annual fluctuations characterised the mortality regime of the city. Although most studies have argued that decline in infant mortality in Greece began or accelerated in the late 1920s and 1930s (Bournova, 2016; Hionidou, 1997a; Kanellakes, 1955), the timing of decline in Hermoupolis is in agreement with the national trend as estimated by Valaoras (1960, p. 132). After an increase of IMR mostly as a result of a rise in neonatal mortality, it followed again a somewhat steep declining trend. According to Loukos (2015, p. 66), the reduction in childhood mortality in Hermoupolis was partly the result of the establishment of philanthropic organizations and/or institutions in the 1910s, which aimed to provide childcare, medical care and protection of lower strata children and allowed mothers to work outside their home, an issue that often overlooked in the literature (Mooney & Tanner, 2006). Such practices, thus, increased survival possibilities for many infants in Hermoupolis, including foundlings, as shown also in Figure 2, where foundling mortality declined significantly in the early twentieth century (for Athens, see Maccas, 1911, p. 25).

On the other hand, neonatal mortality had already declined in the early 1910s. Falls in neonatal mortality have been connected with advances in medical care, since most neonatal deaths are caused by genetic makeup or conditions occurring before or during the birth and the quality of care during the delivery itself (Kirby, 1993; Reid, 2001). Despite the increased interest of the Greek state in improvements of obstetrical care and midwifery education in the 1910s (4063/68/1912; 4063/169/1916), it seems that no special improvements took place in Hermoupolis even by the end of the study period. According to the local press and census sources, the number of midwives had increased gradually in the city (Tharros, 6/10/1933, 566, p. 2; NSSG, 1909, 1923, 1935a); yet, oral evidence indicates that even in the 1920s and later, all women gave birth in their houses with the help of unqualified ‘practitioner’ midwives (interviewee #11). A possible reason for the drop in neonatal mortality was the onset of stillbirth registration from 1912 onwards. Even though stillbirths have been excluded from our calculations, those stillbirths that occurred prior to 1912 and may have been reported as neonatal deaths were included in the calculations, as it was impossible to identify them. Therefore, the decline – at least partly – is very likely to be an artefact of a change in the registration process rather than constitute an actual decline.

Despite the relatively low levels of neonatal mortality throughout the study period, the extraordinary rise in 1919 deserves special attention. Other studies have interpreted the occurrence of excess-neonatal mortality during the months after the 1918 pandemic peak as an effect of the increase in prematurity rate, most likely because of infections to mothers in the first or the second trimester of pregnancy (Reid, 2005). It seems that those infants born during the first months of 1919 were at the greatest risk of neonatal mortality. The exceptionally high levels of mortality due to perinatal causes, as shown in Figure 6 and will be discussed in the next section, is an important indication for this hypothesis. Hermoupolis experienced a prolonged and intense wave in late fall of 1918, when influenza-related mortality increased substantially also among young children (Raftakis, 2019). Post-neonatal mortality, however, had already increased since 1917 possibly as a result of the naval blockade of part of the country, which led to a food crisis.⁷

Unlike the experience of other Greek populations, for instance, Athens, Patras, Mykonos or even Greece as whole (Eliopoulos, 2011; Hionidou, 1997a; Kanellakes, 1955; Valaoras, 1960),⁸ where mortality had increased as a result of the effects of the period of almost continuous wars from 1912 to 1923, the following arrival of the Asia Minor refugee and its aftermath (Valaoras, 1939, pp. 21–22), the decline in Hermoupolis was only temporarily interrupted. Certainly, the temporary increase in post-neonatal mortality in 1923–25 should be associated with the Asia Minor refugee influx on the island.⁹ Subsequently, IMR further declined, reaching a level much closer to the national average. As no straightforward explanation for the infant mortality decline in the first decades of the twentieth century could be put forward (apart from changes in stillbirth registration and improvements in living standards, which more likely had short but immediate effect), the framework introduced by Woods et al.(1989)was adopted. According to this framework, infant feeding practices, women’s education, decline in fertility and improvements in public hygiene were identified as having contributed to the great decline in infant mortality in England and Wales.

Although no national statistics on breastfeeding are available, it is believed that breastfeeding was widespread in Greece (Maccas, 1911). In the case of Mykonos, as in Paros, the minimum average duration of breastfeeding was typically around 12 months (Gavalas, 2001; Hionidou, 1997a). In the case of Hermoupolis, oral evidence suggested that the minimum average duration of breastfeeding was between 8 and 12 months (interviewees #1,9,13,19). It does not appear that major changes in breastfeeding patterns had taken place in the early twentieth century, thus breastfeeding could not have made an important contribution to a decline in infant mortality, although it may have helped to stabilize the rates (Raftakis, 2019). Nonetheless, wet-nurses were employed by the local hospital to take care the high volume of foundlings in the city. In many cases, wet-nurses did not have enough breastmilk to feed both their own children and the foundling infants, so they had to buy cheap impure milk due to the limited compensation that they received by the local government, or even used flour pulps and/or leftover food. Due to this inadequate nutrition and contaminated food, many foundlings often got sick or even died of intestinal disorders or gastroenteritis (Arfanes, 1900; Loukos, 1994). Given that from the 1910s onwards foundlings received better care and nourishment, it is expected that such measures had a cumulative effect on the reduction of infant mortality in the city.

In Greece, female illiteracy was pervasive in the late nineteenth century (93% of all women could not even write their own name; Tsoukalas, 1977, p. 413), while by the late-1920s had declined significantly (around 40%). In the case of Hermoupolis, census material exhibited a substantial rise in the literacy rate of all women aged 15–49 in the first decades of the twentieth century (from 40% in 1907 to 64% in 1928: NSSG, 1909, 1923, 1935b), while evidence from school registers showed a significant increase in female school attendance in the late 1920s (GSAS/DA/Archeio Gymnasiou, 211). It is therefore expected that increasing female literacy at the time of infant mortality decline would have contributed to the trend, however the potential effect of the ‘educated’/literate mothers was concealed, since it coexisted with that of the older ‘less-educated’/illiterate mothers (Hionidou, 1997a, p. 167).

Although age-specific fertility rates have not been calculated since it was beyond the scope of this study, crude birth rates (CBR) have been calculated around census years. In Figure 3, it is evident that CBR began to decline in the 1920s, a few decades later than ECMR and IMR. Certainly, the lack of more sophisticated fertility rates prevents firm conclusions from being drawn, though the decline of infant mortality in Hermoupolis (at least for the post-1920 era) could be partly the product of the mortality-fertility interaction during the demographic transition (Reher, 1999, 2004).

The limitation of this study to investigate the mortality-fertility interaction leads to the exploration of the possible impact of certain exogenous factors such as sanitary reforms on the course of infant mortality. The introduction of the underground water supply system in the mid-1920s could be also somehow related to the decline in infant mortality or at least coincided with the timing of further decline. Although spatial analysis would generate useful results, the positive impact of the introduction of water supply (Alsan & Goldin, 2019; Cutler & Miller, 2005; Floris & Staub, 2019; Peltola & Saaritsa, 2019), sewerage system (Kesztenbaum & Rosenthal, 2017) or the combination of the two (Alsan & Goldin, 2019; Gallardo–Albarrán, 2020; Helgertz & Önnerfors, 2019) on reductions in infant and childhood mortality has been established in the literature. Cause-of-death analysis along with seasonality indexes will be presented in the next section in order to discuss the potential influence of water supply and sewerage system on infant mortality in Hermoupolis.

When the IMR in Hermoupolis is compared with those of neighbouring semi-urban Mykonos and rural Paros, the much higher incidence of infant deaths in Hermoupolis is revealed (Gavalas, 2001; Hionidou, 1993, 1997a). High levels of infant mortality also existed in Athens (Kanellakes, 1955) and Patras (Eliopoulos, 2011), two significant contemporary urban centres. As shown in Figure 4, wide urban–rural discrepancies have been estimated by Kanellakes (1955) and Tsaouses (1971) using published statistics. This study therefore confirms the existence of an urban mortality disadvantage in Greece, in agreement with most European experience (Kearns, 1988; Ramiro-Fariñas & Oris, 2016; Reher, 2001; Williams & Mooney, 1994). When infant mortality declined, urban rates reached levels much closer to those of rural areas. The main reason for the low rural infant mortality in Greece has been assumed to be the high level of under-registration in those areas (Valaoras, 1988). On the other hand, urban infant mortality was substantially inflated by the mortality of foundlings. Foundling hospitals or hospitals that received foundlings existed in all major Greek urban settings, for which high levels of foundling mortality have been estimated (Eliopoulos, 2011; Hatzivasileiou, 1935; Kanellakes, 1955; Loukos, 1994; Maccas, 1911; Moustane, 2014; Papavasileiou, 1903). Excess urban infant mortality may have also been the result of poor environmental and hygiene conditions in most urban centres in Greece. Given that public health in Greece was almost non-existent until the late 1920s, while in the countryside improvements in public health occurred only after World War II (Copanares, 1933), it is reasonable to suggest that such unhealthy conditions were primarily responsible for the high urban infant mortality in nineteenth- and early twentieth-century Greece.

Figure 4. IMR in various places in Greece, 1861–1940 (around census years)

Hermoupolis: HMD; Greece: (Valaoras, 1960); (Kanellakes, 1955); (Tsaouses, 1971); Athens: (Kanellakes, 1955); Patras: (Eliopoulos, 2011); Mykonos: (Hionidou, 1993); Paros: (Gavalas, 2001).

The high levels of infant mortality recorded in Hermoupolis can be therefore compared with those of some of the unhealthiest English and German cities during the second half of the nineteenth and early twentieth century (Millward & Bell, 2001; Vögele, 1994). While the IMR in Hermoupolis was particularly higher than that of other Mediterranean island communities, such as Alghero or Palma de Mallorca, which experienced diverse timings of infant mortality decline (Breschi et al., 2016; Pujadas-Mora, 2009). Hermoupolis infant mortality levels, however, are comparable with those of the Mediterranean island community of Malta, where IMR was extremely high due to the high prevalence of diarrhoeal-related diseases (Sawchuk et al., 2013).

Given the importance of exogenous causes of mortality in Hermoupolis, in order to further comprehend the environmental conditions associated with post-neonatal mortality, seasonality indexes and cause-specific mortality rates during infancy were calculated and presented in the next section.

6.2. Seasonality of infant mortality and causes of death

Seasonality of infant deaths by age has been calculated and presented in Table 4.¹⁰ A division into four periods has been adopted in order to detect any differences in infant mortality over time. It was necessary to group deaths by quarter in order to capture the full length of the seasonal pattern. In addition, Greece adopted the Gregorian calendar in 1924, so all seasonal indices prior to that year have been adjusted to take account of the Julian calendar and to minimize the effect of the 13-day discrepancy between the two calendars. In all four periods, seasonality of infant deaths followed very similar trends. Neonatal mortality displayed a prominent winter peak throughout the study period. The effect of the seasonality pattern of births could be partially associated with the observed peak in the winter months, since more than the average births occurred in those months. By examining the births of those infants who died around the census years (as employed in the linkage exercise), seasonality of births peaked in the first trimester of the year (Raftakis, 2019, p. 219). The limited use of births for the exploration of birth seasonality is a significant limitation of this study and therefore should be treated with caution. Nonetheless, the high occurrence of births during the winter months has been observed in late-nineteenth century Hermoupolis (Fragkides, 1894, p. 33), but also on neighbouring islands (Hionidou, 1993, p. 45). For post-neonates, a strong summer peak persisted throughout the study period. More specifically, the summer peak for infants, aged 1–5 months, only slightly declined during the last period, while that of post-neonates, aged 6–11 months, experienced the highest rate of increase. According to Davenport (2019, p. 187), seasonality analysis of infant deaths by age can be used as an indirect method to investigate breastfeeding practices. It is widely accepted that breastfeeding provides significant protection against diarrhoea, respiratory tract infections etc. (Goldman, 1993; Hanson, 1999). On the other hand, weaned infants are exposed to higher risks related to inadequate nutrition and/or contaminated food, especially during the summer months. These summer peaks in infancy are often associated with large occurrence of diarrhoeal diseases in this age-group and less often with non-universal or early cessation of breastfeeding (Davenport, 2019). Therefore, despite the widespread practice of breastfeeding in the city, the persistence of summer peaks after the first month of life may indicate the early initiation of supplementary food. Given the inadequate nourishment of foundlings at least up to the pre-1910 period, the lower incidence of foundling deaths during the summer months in the nineteenth century, when compared to non-foundling infants, is an unexpected finding, though the pattern was reversed in the twentieth century (Figure 5(a-b)). A possible reason for this trend may be that foundlings had developed resistance to hazards derived from inadequate nourishment already since the neonatal period.

Table 4. Seasonality of neonatal and post-neonatal deaths, Hermoupolis, 1860–1940

	Dec-Feb	Mar-May	Jul-Aug	Sept-Nov	N (deaths)
0–28 days
1860–79	1.35	1.00	0.83	0.83	599
1880–99	1.54	0.89	0.74	0.84	623
1900–19	1.36	1.18	0.64	0.83	641
1920–40	1.45	0.88	0.77	0.91	413
1–5 months
1860–79	0.73	0.96	1.55	0.75	1210
1880–99	0.61	1.02	1.64	0.72	1380
1900–19	0.64	1.00	1.68	0.67	1011
1920–40	0.71	0.74	1.54	1.01	746
6–11 months
1860–79	0.68	0.93	1.55	0.83	1109
1880–99	0.63	0.87	1.69	0.79	1045
1900–19	0.70	0.85	1.56	0.87	733
1920–40	0.53	0.65	1.85	0.96	558

All rates prior to 1924 have been adjusted to Gregorian calendar.

Source: HMD.

Figure 5. A-h Seasonality indexes of infant mortality, Hermoupolis, 1860–1940

All rates prior to 1924 have been adjusted to Gregorian calendar. Figure d includes data for the period after the introduction of piped water, while Figures e-h include data for the period 1916–1940.

Source: HMD.

Figure 6. Cause specific mortality rates during infancy per 1,000, Hermoupolis, 1916–1940

Cause-specific mortality was calculated as deaths of infants from specific group of diseases by total births multiplied by 1,000.

Source: HMD.

Using individual level causes of death for the period 1916–1940, cause-specific mortality rates have been calculated during infancy. Cause-of-death reporting of infant deaths in past populations was rather inaccurate, mainly because the description of symptoms was vague, and therefore highly unreliable (Alter & Carmichael, 1996; Kintner, 1986; Reid, 2002; Williams, 1996). Keeping this in mind, the broad cause-of-death groups were employed to generate numerically robust results and to smooth the reporting inconsistencies and ambiguous terms. Diarrhoeal diseases (diarrhoea, enteritis, gastroenteritis etc.) form one group and infectious diseases the second. The other groups are perinatal conditions, respiratory diseases, nervous system diseases and other diseases (Figure 6). Certainly, atrophy and athrepsia constitute two vague but commonly reported causes, which were often linked to gastrointestinal infections by contemporary physicians. They form a separate group due to their important contribution to infant mortality and the somewhat different seasonal pattern to that of diarrhoeal-related diseases (Figure 5(c–e)).

Diarrhoeal-related diseases were the largest category of causes of death in this period, and if deaths due to atrophy and athrepsia are also included, they accounted for about half of all infant deaths for most of the study period. Although diarrhoeal-related mortality had started to decline in the 1930s, the exceptionally rise in 1939 requires special attention. Yet, no explanation has been found for this peculiar event. Perinatal mortality comprised a significant part of infant mortality, especially during the first years of the study period, while it decreased substantially in the 1930s. The noteworthy increase in neonatal mortality in combination with the exceptionally high incidence of diseases related to the perinatal period in 1919 may be associated with the 1918/19 influenza pandemic. Despite the relatively low levels of respiratory diseases, greater rates are noticed in the late-1910s and mid-1920s. Lastly, infectious and other diseases comprised an insignificant fraction of infant deaths.

High levels of infant deaths due to diarrhoeal-related diseases were found during the first semester of life, accounting for 52% of all infant diarrhoeal deaths, whereas 29% were between the sixth and the eighth month. The vast majority of infant deaths due to atrophy and athrepsy also occurred in the first semester (72% and 83%, respectively), though a quarter of them were of neonates. It seems, therefore, that most infant diarrhoeal-related deaths were concentrated amongst infants of less than eight months, while the minimum average duration of breastfeeding has been found to be 8 to 12 months (interviewees #1,9,13,19; Raftakis, 2019). Since infants were dying while being breast-fed, it is suggestive that other factors may have affected infant diarrhoeal mortality. Early initiation of supplementary food could perhaps explain why infants were dying from diarrhoeal diseases at such young ages. Contemporary physicians argued that mothers should suckle their babies for about a year, with supplementary feeding starting at the third or the fourth month of an infant’s life (Melissinos, 1904; Ministry of Hygiene, 1930; Zinnes, 1877). Mothers were also advised to dilute milk with water (Kaires, 1927, p. 394), while they often used flour pap (Melissinos, 1904, p. 31). Evidence from the local press showed that powdered milk was available on the island, which suggests that water was also used for its preparation (Tharros, 8/12/1933, 575, p. 2).¹¹ Yet, milk examinations in the 1930s showed that milk supply in Hermoupolis was relatively clean (GSAS/YPA/YP917/1932). Water in Hermoupolis was found to be mostly contaminated; hence, it can be assumed that it was an important medium for the high prevalence of diarrhoeal diseases especially among the very young. As the city relied on public springs, it is known that during the summer months less water was available on the islands due to droughts, so it can be reasonably assumed that the available contaminated underground water or water stored in cisterns, which were never cleaned (Alivizatos, 1933; Pagkalos, 1933), was one of the main reasons for the extremely high rates of diarrhoeal mortality especially during the summer months, as shown also in Figure 5(c-d).¹² More specifically, the pronounced summer excess mortality associated with diarrhoeal-related diseases and its persistence among all post-neonates in the 1930s, despite the introduction of water supply in the city, is further suggestive of the early use of supplementary food.

The existence of a sewerage system in the city, in addition, was identified by contemporary physicians as being responsible for the transmission of infectious diseases (especially typhoid fever) and high levels of infant diarrhoeal mortality (Fragkides, 1894, p. 55; Iatrike Etaireia Syrou, 1895), despite the positive link between the introduction of sewerage system and declining mortality found in other studies (Kesztenbaum & Rosenthal, 2017). The lack of a piped water supply in the city prior to the mid-1920s is more likely to be responsible for the ineffectiveness of the existing sewerage system, and perhaps it could be supported that they were complementary to each other, as suggested also recently in the literature (Alsan & Goldin, 2019; Gallardo–Albarrán, 2020; Helgertz & Önnerfors, 2019).

Previous research has established that infant diarrhoeal mortality is closely related to personal hygiene, sanitation and improvements in water supply (Cairncross et al., 2010; Cutler & Miller, 2005). However, the relatively low incidence of cholera during infancy in Victorian England showed that infants were relatively protected from waterborne diseases and therefore improvements in water supply had only little impact on infant or diarrhoeal mortality unlike other age-groups (Davenport et al., 2019; Woods, 2000). In the case of Hermoupolis, no firm conclusions can yet be drawn regarding the impact of the introduction of water supply on infant diarrhoeal mortality. Various studies on contemporary societies showed that improved sanitation had greater effect on the reduction of diarrhoeal incidence during childhood rather than improvements in water supply (Fink et al., 2011; Fuller et al., 2015). Evidence from Hermoupolis has shown that wider access to water, irrespective of its quality, may have enabled improvements in personal hygiene among the residents of Hermoupolis (Raftakis, 2019) in line with Floris and Staub (2019), who also found that sanitary interventions increased attention on hygienic practices in several Swiss towns, among others. In particular, oral histories in Hermoupolis indicated that working-class residential areas were found to be exceptionally clean by the 1930s thanks to the initiative of individual housewives (interviewees #1,11,13). Furthermore, the local press had been reporting on the filthiness of Hermoupolis since the mid-nineteenth century (see for example, Panope, 24/12/1877, 400; To Vema, 20/07/1913, 132; Tharros, 13/05/1927, 222); yet, these reports had disappeared or at least reduced significantly by the 1930s. Therefore, it could be suggested here that such improvements in personal hygiene and sanitation may have reduced infant exposure to faeces and this had a global effect on infant mortality, by reducing the incidence of infectious diseases or chronic conditions that increased mortality from other causes. Finally, further associations between parish information and the expansion of the water supply in the city might establish stronger links and shed more light on the effectiveness of wider water availability in an urban setting.

7. Conclusions

This paper discussed the course of infant mortality in Hermoupolis, Syros – an important industrial Greek centre – for the period 1860–1940, based on individual-level data and employing record-linkage techniques in order to test the quality of the available data. Fluctuations in post-neonatal mortality were responsible for the variations in the levels of IMR throughout the study period, reflecting the centrality of exogenous causes. The course of infant mortality seems to have been consistent with the fluctuations in living standards: when the economy of the city declined, infant mortality increased dramatically. IMR started to decline in the late 1890s, returning to the pre-1880s levels, while it further decreased in the mid-1910s most likely as a result of changes in stillbirth registration and improvements in living standards and nutrition among lower strata infants by the local philanthropists. After a short interruption possibly because of the arrival of the Asia Minor refugees and its aftermath, infant mortality reached relatively low levels in the 1930s. This study argued that the decline in infant mortality may have been also to a certain extent the result of the mortality-fertility interaction during the demographic transition (Reher, 1999, 2004) and improvements in maternal literacy. The introduction of the underground water supply in Hermoupolis in the mid-1920s, although not expanded across the whole city, is expected to have had an indirect impact on the decline in infant mortality, in accordance with other studies (Alsan & Goldin, 2019; Cutler & Miller, 2005; Floris & Staub, 2019; Peltola & Saaritsa, 2019), as it enabled improvements in personal hygiene amongst the residents of the city.

In addition, this paper identified that diarrhoeal-related diseases were responsible for the majority of infant deaths for the years 1916–1940, when cause-of-death information is available. Seasonality analysis results indicated that the hot and dry summers in Hermoupolis caused a large number of infant deaths. It was that time of the year when infants had a great risk of dying due to diarrhoeal-related diseases. Most of those infant deaths occurred before the eighth month of life and while the infants were still breastfeeding. Although breastfeeding was rather universal, according to oral evidence, seasonality analysis indicated the early initiation of supplementary food. The restricted access to clean water in the city is considered to have been the most important factor as when supplementary food or formula milk was given to infants it was mixed with impure water. Furthermore, the milk supply has been found to have been relatively clean, as regular tests were carried out by the local authorities at least in the 1930s. Even so, if the clean milk was mixed with contaminated water, it would turn into a fatal medium for the transmission of the diarrhoeal-related diseases amongst infants.

The findings of this paper suggested that the unhealthy urban environment in Hermoupolis was responsible for reducing the survival of many infants and confirmed the wide urban-rural differences that existed in Greece throughout the study period (Kanellakes, 1955; Tsaouses, 1971), in line with other European experience (Kearns, 1988; Ramiro-Fariñas & Oris, 2016; Reher, 2001; Williams & Mooney, 1994). Finally, this paper sets the ground for the study of infant mortality in Greece, however additional research is clearly needed to establish its levels and trends in the country as a whole, but also both in rural and urban areas.

Acknowledgments

I wish to thank the two anonymous referees for their insightful comments and suggestions. I would like also to thank Violetta Hionidou who provided helpful comments on earlier drafts of this paper. Special thanks to Mrs Angeliki Psilopoulou and Taina Hietala, the staff of the General State Archives of Syros, the Hermoupolis Registry Office and the Hermoupolis Library for their assistance during my field trip in Hermoupolis, Syros.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. Hermoupolis hospital was used mainly by the Orthodox inhabitants of the island, while the Catholics would use primarily the French Catholic hospital in Ano Syros (interviewee #22; Hionidou, 2009).

2. The Syros interviews were conducted as part of a project examining the 1941–42 famine in 2000 by Hionidou (2006), but included also questions on public health, living standards and breastfeeding practices. For reasons of anonymity, all interviews denoted by a number.

3. Infant deaths in 1859 have been omitted from our calculations to avoid possible effects of early under-registration.

4. Linkage was restricted to census years, as only then information on mid-year population and age structure of the population was available. The current study is part of the project ‘Mortality transition in Hermoupolis’, for which abridged life tables have been constructed around census years.

5. Loukos (1994, p. 257) suggested that foundlings could also have derived from neighbouring islands, though without providing any further evidence.

6. For the Greek Orthodox Church, the death of unbaptised children was considered to be a mortal sin. When a new-born was in imminent danger of death, either the midwife or a priest would perform a hastened baptism (aerovaphtisma or kantelovaphtisma) (Paradelles, 1995, p. 141).

7. Although Valaoras (1960, p. 127) has identified both events as responsible for huge increases in mortality, very little attention has been paid to their impact due to the scarcity of data.

8. Increases in infant mortality especially in Athens have been also linked to improvements in data coverage (Kanellakes, 1955, p. 138).

9. Although many refugees in infancy were found to have died in the city, it was not possible to estimate their real contribution as such information is only partly available.

10. In the lack of temperature data in Hermoupolis, data from Athens were employed due to geographical proximity: June–September were the hottest months, while December–February were the coldest ones (Exner et al., 1929, 1934).

11. Evidence from the contemporary medical journals Iatrike Proodos and Asklepios suggests that powdered milk existed in the country already since the early twentieth century.

12. The high prevalence of infant diarrhoeal diseases over the summer months in many Greek urban areas was also evident to several contemporaries (Kardamates, 1905; Prinaris, 1848; Stefanos, 1884).