Later, smaller, better? Water infrastructure and infant mortality in Finnish cities and towns, 1870-1938

Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=rhof20 The History of the Family ISSN: 1081-602X (Print) 1873-5398 (Online) Journal homepage: https://www.tandfonline.com/loi/rhof20 Later, smaller, better? Water infrastructure and infant mortality in Finnish cities and towns, 1870–1938 Jarmo Peltola & Sakari Saaritsa To cite this article: Jarmo Peltola & Sakari Saaritsa (2019) Later, smaller, better? Water infrastructure and infant mortality in Finnish cities and towns, 1870–1938, The History of the Family, 24:2, 277-306, DOI: 10.1080/1081602X.2019.1598462 To link to this article: https://doi.org/10.1080/1081602X.2019.1598462 © 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. Published online: 22 Apr 2019. Submit your article to this journal Article views: 311 View Crossmark data ARTICLE Later, smaller, better? Water infrastructure and infant mortality in Finnish cities and towns, 1870–1938 Jarmo Peltola a and Sakari Saaritsa b a Faculty of Social Sciences (SOC), History, Tampere University, Tampere, Finland; b Economic and Social History, University of Helsinki, Helsinki, Finland ABSTRACT We analyse the role of modern water infrastructure in reducing infant mortality in Finnish cities and towns in the late nineteenth and early twentieth centuries. Estimates from US data suggest that urban water infrastructures greatly affected the health transition in Western countries, implying policy lessons for developing countries. Finland is a relevant case due to the early onset of mortality decline in a predominantly agrarian context in a country with a low GDP. Our sources enable analysis across population centres of varying size as well as over different phases of development. We construct panel data on infant mortality and the initiation of three major water interventions – piped water, sewers and chlorination –in 37 Finnish cities and towns from approximately 1870 to 1938. We show that in line with previous literature, the interventions had a significant effectoninfantmortality, jointly accounting for roughly 40% of the average decrease in different cities. However, most of the measurable effect was driven by smalland medium-sized cities adopting more advanced technology in the twentieth century rather than by pioneering larger cities in the nineteenth century. Weighting by population size rather than using average effects reduces the estimate to about 32%. Due to low levels of urbanisation, the measurable impact on national mortality decline was only about 4–5 % over the entire period, but roughly twice as high in the twentieth century, when both urbanisation and a decline in urban infant mortality rates gathered pace. Following development economics, our findings emphasise the importance of distinguishing the effects of sanitation by period and developmental context rather than compressing them into a single estimate. KEYWORDS Urban sanitation; infant mortality; mortality decline; Finland 1. Introduction This article analyses the impact of urban sanitation on infant mortality in Finland, historically a predominantly rural ‘developing country’in Europe. Innovations and technology related to clean water in cities have long been seen as crucial for the health transition of Western countries during the ‘Second Industrial Revolution’of the late nineteenth and early twentieth centuries (e.g. Easterlin, 1998, pp. 69–82; Fogel, 2004, pp. 37–39; Riley, 2001, pp. 64–67; Szreter, 1988). The interventions studied in the existing CONTACT Sakari Saaritsa sakari.saaritsa@helsinki.fi This article has been republished with minor changes. These changes do not impact the academic content of the article. THE HISTORY OF THE FAMILY 2019, VOL. 24, NO. 2, 277–306 https://doi.org/10.1080/1081602X.2019.1598462 © 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License (http://creativecommons.org/licenses/by-nc-nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited, and is not altered, transformed, or built upon in any way. literature include the provision of piped water, chlorination, improvements in filtration techniques and the introduction of sewerage. After revising their original estimates, Cutler and Miller (2015)arguedthatfiltration and chlorination accounted for 41% of overall mortality decline and a staggering 59% of infant mortality decline in 13 major US cities between 1900 and 1936 (cf. Cutler & Miller, 2005,pp,1, 13–14). While Cutler and Miller’s estimates have again been significantly degraded as the result of a recent replication effort (Anderson, Charles, & Rees, 2018), Alsan and Goldin (2015,p.3) estimate that approximately 44% of the dramatic overall fall in the infant mortality rate (IMR) in the greater Boston metropolitan area cities and towns between 1880 and 1915 can be attributed to the impact of providing clean water and constructing a sewerage system. 1 The high figures coming from relatively industrialised and developed contexts have emphasised the importance of an adequate water infrastructure in historical health transitions. From the outset, the idea of Western historical developments in water infrastructure providing lessons for contemporary developing countries has been cited as a motivation for the research efforts. Both differences and similarities between the historical West and contemporary global South have been highlighted, showing the complexities of extrapolating from Western experience. A recent paper on Mexico, for instance, deals with the problems caused by badly maintained pipelines by controlling for the age of the network, finding statistically significant effects (Bhalotra, Diaz-Cayeros, Miller, Miranda, & Venkataramani, 2017). On the other hand, Kremer and Zwane (2007) have emphasised the proven effectiveness of classic, large-scale sanitation infrastructure as opposed to the more inexpensive communal schemes favoured in recent development policy. Some of the arguments in development economics are in turn relevant to historical research. In the early stages of economic and social development, studies have highlighted the necessity of complementary inputs. When habitats are crowded and unsanitary, poverty rates high and levels of human capital very low, the mere provision of piped water may not have a significant impact (for an example from India, see Jalan & Ravallion, 2003). In a ‘threshold-saturation model’ outlined by Shuval, Tilden, Perry, and Grosse (1981), the authors hypothesize that returns are low at the lowest levels of development, increase at moderate levels of development and level offagain at high levels of development, where other technology, knowledge and general conditions have already brought mortality down. This model has been fruitfully applied by Gamper-Rabindran, Khan, and Timmins (2010)inanempiricalstudyonpipedwaterprovision and IMR in Brazil for the years 1970–2000 using quantile regressions. In historical analysis with a deeper time dimension, the operation of the model can be based on time-dependent, initially quite severe constraints caused by the levels of available knowledge and technology. For instance, when a basic awareness of germ theory and its implications was still rare, piped water might not have been used to its full potential. The design of the urban environment or housing might have been more detrimental to health than in later periods, making sanitary life virtually impossible. In cities in the late nineteenth and early twentieth centuries, the influx of migrants, growing population densities and increased crowding created new, multidimensional problems that policy makers had to combat with new technology and infrastructure. The largest cities, where the problems were most severe, were often the first to develop better sanitation, but they were still constrained by the magnitude of the challenges (Oris & Fariñas, 2016,p.6;Cain&Hong,2009). The early schemes for providing piped water also occasionally suffered from quality issues, making them ineffective or potentially even harmful (Evans, 1990 pp. 144–161, 190–191; Snow, 1855). 278 J. PELTOLA AND S. SAARITSA The contributions of this article are threefold. First, we add a case from the European periphery to the pool of historical estimates on the magnitude of the effect of urban water interventions on mortality. This makes it possible to compare and analyse the impact of introducing the same technology in smaller towns, in a less industrialised society and with lower incomes and human capital than usual. Second, we make comparisons across periods and city sizes in order to apply the ‘threshold-saturation’ perspective discussed in development economics to the historical data and produce results supporting a more nuanced approach to the estimation of the impact of sanitation on mortality. Finally, we contribute to a reassessment of a previous narrative on Finland’s early health transition and its drivers. 2. Cities and towns, sanitation and mortality in Finland In the 1870s, the Grand Duchy of Finland, an autonomous part of the Russian Empire, had an estimated per capita GDP equal to roughly half the Western European average (the Maddison project, http://www.ggdc.net/maddison/maddison-project/home.htm). The country was agrarian, with less than 10% of a population of approximately 1.8 million residing in cities and towns and approximately 80% supporting themselves in the primary sector (Pitkänen, 2007, Table 1; Vattula, 1983, Table 1.11). Against this backdrop, the beginning of the health transition was quite early. Following the Great Famine of 1867–1868 –the last major subsistence crisis in Europe not caused by war or deliberate policy –a permanent decline in mortality set in. A decline in fertility ensued beginning around 1910, closing a period of rapid population growth. Immediately after the crisis, improved nutrition, supported by an increased supply of cheap grain from Russia transported via a newly built railway, may have played a role (Turpeinen, 1986). However, in the final decades of the nineteenth century, mortality already appeared to be falling faster in cities and towns. This has typically been attributed to urban sanitary reforms in the previous literature, giving them a key role in the entire transition (Pitkänen, 2007; see also, e.g. Nygård, 2004;Lento,1956;Waris,1934). After the famine, migration to cities and towns started to pick up, even if the population shares and absolute numbers remained modest Table 1 and Figure 1. The 10% urbanization mark was passed around 1890, and around half a million people resided in cities and towns by the year 1920. Following independence and civil war in 1917–1918, internal migration accelerated markedly in the 1920s and 1930s. Rural fertility rates remained high, and the relative share of the population of cities and towns grew via internal migration. On the eve Table 1. Indicators of city and town size in Finland, 1870–1940. All Helsinki Others Year Obs. Mean pop. (sd) Min. Max. Pop. Mean pop. (sd) Min. Max. 1870 33 4307 (6021) 258 28519 28519 3551 (4234) 258 19617 1880 35 4813 (8096) 342 43334 43334 3680 (4610) 342 2701 1890 36 6296 (11310) 526 61583 61583 4717 (6261) 526 28946 1900 37 9176 (17172) 907 93576 93576 6832 (9702) 907 38235 1910 38 12023 (25159) 910 147218 147218 8369 (11363) 910 49691 1920 38 14291 (33028) 825 197848 197848 9330 (12647) 825 58367 1930 38 17680 (40761) 726 243560 243560 11575 (15874) 726 66654 1940 38 23191 (53189) 1058 319939 319939 15171 (19886) 1058 80955 Source: SVT VI Väestötilastoa. THE HISTORY OF THE FAMILY 279 of WWII, the population statistically classified as urban was just above 20% out of a total population of approximately 3.7 million. InthecaseofFinland,theconceptof‘city’merits qualification. Partly due to the predominantly agrarian economy, partly due to Nordic institutional history, many of the administrative-statistical units included in this category were small and essentially non-urban compared to the industrial cores of the Western world. Like in Sweden, of which Finland was a part until 1809, being denoted a city was based on rights granted by the Crown, which introduced a degree of historical arbitrariness into the process (see Enflo, Henning, & Nousiainen, 2016, pp. 14–15). Many small towns on the western coast of Finland had been granted this status to legalise trade by sea before the nineteenth century. Having experienced reversals of fortune and a loss of significance, these towns would by the twentieth century at times consist of fewer than a thousand inhabitants. During the years 1905–1938, after the promotion of Lahti to the status of a city, the number of legally denoted cities remained stable at 38. Their average population was only about 13,000 in 1910, with Helsinki having 147,000 inhabitants, Turku 49,000 and Tampere 45,000; but 29 cities had populations of less than 7,000 (Table 1). As Figure 2 demonstrates, Helsinki was in a league of its own and continued to grow much faster than any other city throughout the period. The smaller towns were characterised by wooden construction, unpaved streets and water infrastructure based of wells and ditches. The few cities of more considerable size and industry, such as Helsinki, Turku and Tampere, were for the most part similar aside from a small but growing core of buildings and homes made of brick or stone and grid plans for the central streets. Yet, most of these places adopted innovations from the Second Industrial Revolution to deal with sanitation during the period under study. The Finnish localities studied here cannot be compared with contemporary metropoles in the USA or the UK, and the term ‘cities and towns’is applied throughout the article with 0250000 500000 750000 1000000 5 % 25 % 20 % 15 % 10 % 1800 1810 1820 1830 1840 1850 1860 1870 1880 1890 1900 1910 1920 1930 1940 1950 Year Urban population, share Urban population, total (right axis) Figure 1. Urbanisation in Finland, 1800–1950. Notes: Parish of Tornionjokilaakso annexed to Grand Duchy in 1809 (population 11 800). Province of Viipuri annexed to Grand Duchy in 1811 (population 185 000). Orthodox Christian population included since 1830 (population 25 200). Territorial losses with complete evacuation and resettlement in 1940 and 1944 (population affected 430 000). Source: (Suomen Tilastollinen Vuosikirja, 1950, p. 6), Table 7. 280 J. PELTOLA AND S. SAARITSA this caveat in mind. This is not entirely unusual. While large and growing cities in industrial countries have been the typical context for discussing sanitary interventions, water infrastructure investment has also been analysed in outlying regions and small towns both in developing countries and in the peripheriesofwealthier countries (Gamper-Rabindran etal., 2010;Watson,2006). Some of the work on major early industrial countries has at times actually been based on data from smaller localities (Alsan & Goldin, 2015; Gallardo Albarrán, 2018). Furthermore, literature on infant and child mortality in the Nordic countries has suggested that certain phenomena usually associated with larger cities, such as the urban penalty, might also have applied to small, non-industrial communities with a high population density or to rapidly industrialising rural localities (Edvinsson, Garðarsdóttir, & Thorvaldsen, 2008,p.459;Lazuka,Quaranta,&Bengtsson,2016, pp. 1, 41). 2.1. Urban IMR and mortality trends A number of historical and contemporary papers have analysed the impact of sanitation specificallyonIMR(Alsan&Goldin,2015; Gamper-Rabindran et al., 2010;Watson,2006). While child mortality (1–5 years) is usually seen as a better indicator of environmental conditions due to a lack of distortion by birth defects and variations in breastfeeding (e.g. Oris, Derosas, & Breschi, 2009, p. 367), infant mortality has its merits as an outcome variable of sanitary investment. Infant mortality tends to form a high share of total mortality in ‘pre-transition’populations. It is highly likely that the first ingestion of water used for food preparation and drinking as well as washing occurs during the first year of life, as strictly exclusive breastfeeding for 12 months tends to be rare in most past and present populations. 2 A lack of water for hygiene increases early risk of infection (Gamper-Rabindran et al., 2010). 3 Data on IMR is also readily available due to its early establishment as a public health indicator, first pioneered in mid-eighteenth-century Sweden (Laurent, Figure 2. The distribution of city sizes in Finland, 1870–1930. Circles weighted by absolute population, y axis in logs. Source: SVT VI Väestötilastoa. THE HISTORY OF THE FAMILY 281 2017, pp. 44–47). In a long time series, all-cause IMR does not suffer from historical inaccuracies in establishing causes of death (Koskinen & Martelin, 2007, pp. 180–181). For these reasons, infant mortality has often turned out to be the dependent variable of choice in sanitation studies. In the case of Finland, the IMR is available for the entire relevant period and produces consistent estimates. 4 As a conservative first pass, we therefore focus on the measurable effect of sanitation on infant mortality. Over the period, both urban and rural mortality measured by the crude death rate (CDR) fell considerably. In terms of CDR, the urban penalty disappeared around the 1890s. For infant mortality, this took until the 1920s. Nascent urbanisation created new health problems in the largest cities. Still, both rural and urban IMR fell steadily together, resulting in a remarkably stable urban penalty until a step-like drop in the 1920s, when cities and towns finally took the lead in the decline (Figure 3). 5 How large a share of the overall mortality decline has urban IMR represented in the case of Finland? Population statistics make it possible to decompose the changes in the CDR into urban and rural by period, and to further extract the share of urban IMR. The findings are predictable (Table 2). Over the period from 1870–75 to 1930–35, overall CDR 50 001 051 002 1880 1890 1900 1910 1920 1930 1940 Year Urban IMR Rural IMR Figure 3. Urban and rural infant mortality rates in Finland, 1880–1940. Source: SVT VI Väestötilastoa, Väestönmuutokset. Table 2. Reductions in urban and rural mortality and contributions of urban and rural mortality decline to overall mortality decline over 30-year periods, from 1870–75 to 1930–35. Urban (CDR 1900– 1905 = 23.8) Rural (CDR 1900– 1905 = 21.5) Total (CDR 1900– 1905 = 21.7) Contribution to total CDR change, % ΔCDR % ΔCDR % ΔCDR % Urban Rural 1870–75-1900–05 −6.3 −26.6 −2.1 −10.0 −2.6 −11.8 10.0 90.0 1900–05-1930–35 −6.1 −35.0 −5.6 −28.7 −5.8 −30.4 46.4 53.6 Source: Computations from data based on Suomen Tilastollinen Vuosikirja, 1950 (1951), pp. 44–45. 282 J. PELTOLA AND S. SAARITSA fell from approximately 24.3 to 15.9 per 1000. In cities and towns, the drop was from a much higher 30.1 all the way down to 17.7 per 1000. However, urbanisation was still quite modest in the nineteenth century, and the effect of a decline in mortality in cities and towns on overall mortality remained small. Over time, cities and towns gradually started to grow faster and simultaneously took the lead in the health transition. Their contribution to the total fall in mortality grew markedly as a consequence, accounting for about half of the total decline between 1900–1905 and 1930–1935, despite the fact that urbanisation still barely reached 20%. In the early 1870s, the deaths of infants still constituted roughly 25% of the national CDR. With respect to the total fall in urban CDR, the falling IMR accounted for approximately 43% in the period from 1870–1875 to 1930–1935. The contribution increased markedly in the twentieth century, when the urban IMR fell quickly and the urban IMR penalty vanished (Table 3 and Figure 4). Estimating how much of the decline in urban IMR can be attributed to water interventions will make it possible to experimentally estimate the contribution of this particular, measurable part of the impact of urban sanitation on general mortality trends. Table 3. The contribution of infant mortality to the crude death rate and its decline over 30-year periods in Finnish cities and towns, 1870–75 to 1930–35. Period Total CDR Infant CDR* Infant share, % Total CDR change Infant CDR* change Infant share of change, % 1870–75 23.1 5.9 25 1900–05 17.1 4.3 25 −6.0 −1.6 26.3 1930–35 11.1 0.8 7 −5.9 −3.5 58.6 Source: Computations from SVT VI Väestötilastoa: SVT VI, Väestönmuutokset 1865–1940; SVT VI 29:1–3, Väestön tila, 1750–1890; Suomen Tilastollinen Vuosikirja, 1879–1940. *Refers to deaths of infants per 1000 people. .05 .1 .15 .2 .25 .3 0001repshtaedfoerahS 010 20 30 40 50 0001repshtaeD 1860 1880 1900 1920 1940 Year CDR CDR infants Infants of CDR, share (right axis) Figure 4. The share of infants in the crude death rate, 1865–1938 (sources: SVT VI Väestötilastoa: SVT VI, Väestönmuutokset, 1865–1940; SVT VI 29:1–3, Väestön tila, 1750–1890; Suomen Tilastollinen Vuosikirja, 1879–1940). ”CDR infants”refers to deaths of infants per 1000 people. THE HISTORY OF THE FAMILY 283 2.2. The spread of sanitation As this section shows in detail, the introduction of a modern water infrastructure in Finland occurred in two distinct phases, with different characteristics. In the nineteenth century, piped water was initiated in some of the largest, growing cities. Since the technology was novel, there were more quality issues, sometimes including confirmed or suspected outbreaks of disease spread through the supply system itself (see Evans, 1990, pp. 144–161, 190–191 for a similar experience from Hamburg, Germany; Snow, 1855 is a classic on London). Dissemination and take-up were gradual. Sewers were introduced in eight Finnish cities and towns before 1900, but importantly they often preceded piped water. The aim was to contain local waste water issues, and the concept and technology differed from what came afterwards. The pioneering largest cities were also dealing with a host of problems related to early urbanisation, which applied negative pressuresatthesametimethatoverallmortality rates were improving. In the twentieth century, sanitation spread to smaller cities and towns, and it was introduced as a modern ‘package’with simultaneous water and sewer service (see Table A1) and the application of chlorination also started. This later phase, particularly from the 1920s onwards, was underpinned by quickly declining mortality rates, growing incomes and generally improving social capability (on this concept, see Abramovitz, 1986). Despite the general poverty of most in the country, the technical elite of Finland was informed and cosmopolitan. Developments in more advanced European countries were keenly followed and local conditions carefully studied with modern techniques (e.g. Hietala, 1987,1992; Laakkonen, 2001; Niemi, 2007). 6 Waterworks providing piped water were first introduced by Helsinki (1876), Tampere (1882) and Viipuri (1892), with three more cities and towns (Oulu, Turku and Hanko) following suit in the 1900s. The initiation of such waterworks clustered around the years 1909–1917, when 11 new cities and towns began piping water. Figure 5 plots the establishment of new facilities and the 010 20 30 40 Number of waterworks 50 100 150 200 RMInabrU 1860 1880 1900 1920 1940 1960 Year Urban IMR Number of waterworks Figure 5. Urban infant mortality and the number of Finnish cities with waterworks, 1870s–1950s. 284 J. PELTOLA AND S. SAARITSA Table 8. Sanitary interventions and infant mortality in Finnish cities, 1870–1938 (comparison of fixed effects estimates across data and specification variants). Variant No large cities Population weighted Population weighted, no large cities 20th-century interaction terms I–III IV V I–III IV V I–III IV V I–III IV V Models (Separate) (Multiple) (Interact) (Separate) (Multiple) (Interact) (Separate) (Multiple) (Interact) (Separate) (Multiple) (Interact) Water −0.143 −0.113 −0.450 0.065 0.052 0.036 −0.077 −0.029 −0.436 −0.238 (0.064)*** −0.136 (0.059)** −0.471 (0.166)***(0.064)** (0.064)* (0.171)** (0.073) (0.045) (0.062) (0.066) (0.058) (0.126)*** Sewers −0.122 −0.042 −0.083 −0.033 −0.079 −0.086 −0.090 −0.072 −0.099 −0.144 (0.065)** −0.054 (0.063) −0.087 (0.053)(0.057)** (0.048) (0.044)* (0.057) (0.048) (0.053) (0.063) (0.059) (0.061) Chlorine 0.018 0.092 0.094 −0.226 −0.217 −0.218 0.028 0.056 0.061 (0.079) (0.081) (0.081) (0.075)*** (0.069)*** (0.068)*** (0.084) (0.095) (0.095) Joint effect −0.062 −0.438 −0.244 −0.268 −0.046 −0.474 (.086) (0.185)** (0.070)*** (0.089)*** (0.107) (0.188)*** Notes: Dependent variable: ln IMR. All models include year and city fixed effects. Coefficients extracted from models I–V, as in Table 7. Separate: coefficients extracted from separate regressions for each intervention; multiple: coefficients extracted from multiple regression: interact: coefficients extracted from specifications with the PipedXSewers interaction. Standard errors clustered by city in parentheses. * p< .10, ** p< .05, *** p< .01. THE HISTORY OF THE FAMILY 291 greater impact in smalland medium-sized localities. Did the effect of introducing piped water and sewers also appear to change over time? Table 8 presents estimates for an interaction term with a dummy for the twentieth century for these two interventions. The estimates are negative and significant, whereas the (unreported) constituent effects are not. 19 In this specification, piped water and sewers were significantly more clearly related to reductions in IMR in the period after 1900. The findings would seem to point to the same direction. The early phase, when the very first piped water systems were launched in the largest cities, did not seem to affect IMR as much at the city population level. Only later, when smalland medium-sized towns followed suit using tried and improved technology and coordinated roll-outs among populations of manageable size, did the effects become statistically visible. There are plausible explanations for these results. The early introduction of sanitation took place in challenging circumstances. Piped water often had quality issues before filtration developed; in Tampere, it actually caused a typhoid epidemic in 1916, and in Helsinki there were suspicions of the same in the 1890s (Herranen, 2001, p. 61). Dissemination and take-up were slow, and sometimes inhabitants persistently preferred the old unsanitary wells to new water outlets if the latter involved payments or were inconveniently located (Herranen, 2001, p. 62; Pietikäinen, 2018, pp. 27–30). In larger cities with more inequality between residential areas, access to piped water spread more slowly. Factors like the growth of slums might have pulled mortality rates up at the same time that better sanitation was pulling them down, creating local confounders. In the twentieth century, the newly treated populations were smaller and more homogenous. Modern filtration was the norm, and sewers were immediately linked to water provision. Even if sanitation was bound to improve conditions in the larger cities compared to a counterfactual completely without interventions, at the level of the entire population this was not necessarily measurable in the short run. Within cities, a more fine-grained analysis could yield more positive results (cf. Kesztenbaum & Rosenthal, 2017;Troesken,2004). Between cities, the latecomers were the ones registeringclearer successes. The strong effect of chlorination, which results in a one-offtreatment of a population dependent on a preexisting water system and largely eliminates quality issues, is the only separate estimate remaining significant in the population-weighted regressions. In Finnish cities, this was a twentieth-century technology. 4.2. Robustness checks and event study analysis The sensitivity of the results is tested with added controls and event study analysis. Log population, which was left out of the main specification due to concerns about the mechanical correlation between population and a dependent variable measuring mortality, has been reinserted. We also control for a variable measuring the share of industrial workers out of the total city population. Similar controls are typically applied based on changing health risks due to industrialisation. The measure could only be constructed from 1886 onwards, however, leading to a loss of 16 years’worth of data, including several early interventions; it also presented some issues regarding smoothness. 20 All the results discussed above are fully robust to the inclusion of these two variables, and the estimates are not reported separately. Furthermore, city-specific trend variables have been added (Tables 9 and 10). These variables control for potential factors affecting 292 J. PELTOLA AND S. SAARITSA mortality linearly over time in city-specific ways, but they also further saturate the model. In the main set of regressions, piped water still retains its significance, while models II–IV now lack significant estimates,notably even for chlorination. In model V, the sewer variable again becomes borderline significant, but the behaviour of the piped water estimate and the interaction term defies meaningful interpretation. Joint effects are unstable. Turning to Table 10, with the four largest cities removed the findings are partly robust to the inclusion of city trends, with a significant estimate on piped water but the p-value for sewers dropping to .12 in the separate regressions (models I-III). In the case of weighted regressions, chlorination also loses all significance. Instead, the coefficients for piped water become significant. With the largest cities removed, the coefficients are substantively similar to those in Table 8, but they now show significant estimates for piped water and sewers separately. The interaction term for the twentieth century does not yield any results. In sum, it would seem that the findings on differences between cities of different sizes and periods are sensitive to assuming local trends. However, the assumption of linear trends in mortality over the entire research period at the city level can be considered strong. Losing some results due to unit trends is not unusual, and it is not necessarily enough to reject the findings (cf. Beach, Ferrie, Saavedra, & Troesken, 2016, pp. 49–50; Kesztenbaum & Rosenthal, 2017, p. 181). In particular, the disappearance of any effect from chlorination in the specifications suggests treating the findings with caution. They do provide occasional support for a stronger impact in the late adoption of sanitary interventions by smalland medium-sized cities and towns through slightly higher coefficients without the large cities in the weighted regressions. Event study graphs derived from Equation (2) have also been drawn to complement the regressions. Ideally, these graphs should indicate a discontinuity at the point of impact and afterwards; they can also capture preand post-trends. The graphs in Figures 6–8present Table 9. Sanitary interventions and infant mortality in Finnish cities, 1870–1938 (fixed effects estimates with additional controls and city and town specific trends). lnIMR I II III IV V Piped water −0.094 −0.082 −0.372 (0.047)* (0.057) (0.094)*** Sewers −0.065 −0.014 −0.102 (0.046) (0.056) (0.052)* Chlorination −0.091 −0.081 −0.038 (0.084) (0.086) (0.086) PipedXSewers 0.358 (0.091)*** lnPopdens −0.075 −0.079 −0.088 −0.078 −0.091 (0.054) (0.054) (0.054) (0.053) (0.051)* Popgrowth (t/t-5 yrs) 0.000 0.000 0.000 0.000 0.000 (0.001) (0.001) (0.001) (0.001) (0.001) lnPopulation 0.192 0.219 0.214 0.189 0.182 (0.107)* (0.109)* (0.106)* (0.111)* (0.111) Share ind workers 0.340 0.324 0.324 0.333 0.279 (0.617) (0.617) (0.632) (0.620) (0.623) City trends Yes Yes Yes Yes Yes R 2 0.42 0.42 0.42 0.42 0.42 Sanitation joint effect −0.176 (0.113) −0.511 (0.143)*** Notes: Dependent variable: ln IMR. All models include year and city fixed effects and cityand town-specific trends. Number of observations = 2424; number of groups = 37. Standard errors clustered by city in parentheses. * p< .10, ** p< .05, *** p< .01. THE HISTORY OF THE FAMILY 293 Table 10. Sanitary interventions and infant mortality in Finnish cities, 1870–1938 (comparison of fixed effects estimates across data and specification variants with additional controls and cityand town-specific trends). Variant No large cities Population weighted Population weighted, no large cities 20th-century interaction terms Models I–III (Separate) IV (Multiple) V (Interact) I–III (Separate) IV (Multiple) V (Interact) I–III (Separate) IV (Multiple) V (Interact) I–III (Separate) IV (Multiple) V (Interact) Water −0.109 −0.095 −0.485 −0.084 −0.075 −0.273 −0.093 −0.046 −0.449 0.026 (0.083) 0.019 (0.084) −0.208 (0.156)(0.055)* (0.072) (0.153)*** (0.037)** (0.042)* (0.057)*** (0.050)* (0.064) (0.117)*** Sewers −0.082 −0.018 −0.087 −0.058 −0.026 −0.113 −0.103 −0.073 −0.111 −0.029 (0.048) −0.003 (0.060) −0.001 (0.050)(0.051) (0.067) (0.058) (0.042) (0.045) (0.049)** (0.051)* (0.066) (0.067) Chlorine −0.089 −0.012 −0.008 −0.016 −0.028 −0.015 −0.061 −0.012 −0.006 (0.196) (0.196) (0.197) (0.027) (0.028) (0.031) (0.142) (0.165) (0.166) Joint effect −0.125 −0.580 −0.125 −0.580 −0.132 −0.565 (0.201) (0.255)** (0.202) (0.255)** (0.182) (0.229)** Notes: Dependent variable: ln IMR. All models include year and city fixed effects and cityand town-specific trends. Coefficients extracted from models I–V, as in Table 7, with additional controls for the share of industrial workers and log total population. Separate: coefficients extracted from separate regressions for each intervention; multiple: coefficients extracted from multiple regression: interact: coefficients extracted from specifications with the PipedXSewers interaction. Standard errors clustered by city in parentheses. * p< .10, ** p< .05, *** p< .01. 294 J. PELTOLA AND S. SAARITSA Figure 7. Event study: piped water and infant mortality. Coefficients extracted from regressions, as in Equation (2), with the ends of the event window binned and year t-1 used as a reference; 95% confidence intervals based on standard errors clustered by city; controls are those shown in Table 7. Figure 8. Event study: sewers and infant mortality. Coefficients extracted from regressions, as in Equation (2), with the ends of the event window binned and year t-1 used as a reference; 95% confidence intervals based on standard errors clustered by city; controls are those shown in Table 7. THE HISTORY OF THE FAMILY 295 estimates with a 10-year event window in each direction around the point at which each intervention was initiated, modelled separately (models I-III). All of the models include year and fixed effects and controls, as in Table 7, and they cluster standard errors at the city level. The ends of the event windows have been binned into categories of ‘10 years or more’.The year before initiation (t-1) is the reference category. This is a common approach, although conventions on the size of the window, binning and combining years currently vary in the literature (cf.Alsan&Goldin,2015, p. 40; Hjørt, Sölvsten, & Wüst, 2017,pp.92–94; Helgertz & Önnerfors in this issue for variants). The figures superimpose coefficients and confidence intervals from three regressions each: one on the full data, one on twentieth-century observations only and one with the four largest cities left out. 21 While the resulting graphs are not ideal in terms of clarity or consistency, significant parts do support a treatment effect. 22 The pattern is clearest for piped water. There is no visible trend preceding the intervention, after which the coefficients drop to an asymptotically lower level below zero. The pattern is slightly stronger with the data excluding the large cities, in line with previous findings. The graph for sewers also has a pattern, although this time with signs of a preand post-trend of about three years around the time of initiation. This could be connected to anticipation, awareness and gradual dissemination in the context of early cases of ‘reactive’sewer construction, where there was no close association with piped water. Of the three estimations on sewers, the one done on the full data including the nineteenth century would in this analysis actually seem marginally more suggestive than the others. This would conform to the hypothesis regarding the effectiveness of some of the early, stand-alone sewer schemes. For chlorination, the dispersion of the estimates is higher, and all are on the low side. Figure 9. Event study: chlorination and infant mortality. Coefficients extracted from regressions, as in Equation (2), with the ends of the event window binned and year t-1 used as a reference; 95% confidence intervals based on standard errors clustered by city; the controls are those shown in Table 7. 296 J. PELTOLA AND S. SAARITSA All in all, one key result remains robust: piped water had a clear effect on IMR, and this appeared stronger in smaller cities and towns. 4.3. Contribution to total mortality decline Previous Finnish literature has suggested a key role for urban sanitation in initiating the overall national mortality decline since the 1870s (e.g. Pitkänen, 2007). However, no attempts at measurement have been made thus far. It is therefore of interest to conduct an experiment with the figures at hand. What could these estimates of the effect of sanitation have meant for total mortality decline? As mentioned, only all-cause IMR provided clear and consistent estimates in our data; nothing comparable could be found using CDR or the available truncated data on cause-specific mortality (see endnote 4). Until the 1920s, urban IMR based on total aggregate births and deaths in Finnish cities and towns was above the unweighted mean in the cities and towns discussed so far (Appendix Figure A1), evidently due to the influence of the largest cities. It declined from 205 to 62 during the period from 1870–75 to 1935–38, or by approximately 70%, as opposed to a decline in the unweighted average of approximately 65% (from 176 to 62). If applying the coefficient of joint significance from the unweighted regressions, approximately 37% of this decline could be attributed to sanitation over the whole period. However, this coefficient estimates the unweighted average effect across cities and towns, and can be considered an overestimation for our purposes. Disregarding population size inflates the larger effect found in smaller cities and towns. In principle, population-weighted regressions should be more appropriate. If we would take the coefficient of joint significance from these regressions (Table 8), the implied percentage change over the whole period would be approximately −22.6%. 23 This would yield a contribution by sanitary interventions of approximately 32.4% rather than 40% to the overall decline in infant mortality in cities and towns from 1870–75 to 1935–38. In terms of the contribution to total national mortality decline, the low levels of urbanisation for most of the period automatically lead to very low estimates. In the period from 1870–75 to 1930–35, the contribution of the decline in urban infant deaths to total CDR decline could be estimated to have been only about 12% on the basis of the figures presented in section 2.1.If32–40% of this decline were due to sanitation, this would have implied a measurable contribution of only about 4–5% to the overall mortality decline throughout the period. While the finding is in many ways obvious, it is relevant for assessing the previous Finnish literature, which emphasizes the role of urban sanitation already in the nineteenth century. In the early twentieth century, the contribution grew due to both growing urbanisation and the rapid decline in urban mortality, which was largely driven by a decline in urban IMR at that time. The decline in urban infant deaths accounted for approximately 27% of the total national decline in CDR from 1900–05 to 1930–35. When applying the above estimates of the effect over the whole period, approximately 9–11% of the decline in CDR could at this point be attributed to improved sanitation. It is also possible to estimate coefficients separately for the twentiethcentury observations in the data. 24 In this case, unweighted regressions like those in THE HISTORY OF THE FAMILY 297 Table 7 would yield an estimate of roughly one half of the decline in urban IMR in the 1900s being attributable to sanitation. As much as about 14% of the total decline in CDR would thus have stemmed from this source. The joint significance for the tenuous population weighted regressions would actually be lower with truncated data. The coefficient would imply merely a 17.5% change due to sanitation, yielding approximatelyan8%changeintotalCDR. All in all, the figures would suggest a sizable and growing impact on infant mortality within the cities and towns, possibly reaching as much as one half in the twentieth century. Due to modest urbanisation, the effect on the overall decline in mortality throughout the country would have remained small. In the twentieth century, however, it can be estimated to have been roughly double the average throughout the whole period. This growth in importance was not simply a mechanical outcome of growing urbanisation, but a confluence of several factors: the rapid decline in urban IMR, its growing share of the overall decline in national mortality and the greater effectiveness of the sanitary interventions themselves. 5. Conclusions In this article, we have provided quantitative estimates of the impact of the introduction of major sanitary reforms –piped water, sewers and chlorination –on infant mortality in the cities and towns of Finland in the years 1870–1938. We discovered significant and sizable effects resulting from the three key sanitary interventions, which is comparable with findings in previous literature on e.g. the United States. On average, chlorination in particular seemed to be associated with steep declines in IMR. When breaking this down by city size and period, however, variations emerge. At the level of city and town populations, it is difficult to discern the effects of sanitation for the larger pioneer cities of the nineteenth century. On the other hand, the effects in smaller cities and towns that followed appear more robust, and in particular piped water seems to have had a greater impact when introduced in such contexts in the twentieth century. Our findings suggest that while sanitation became a crucial technology in lowering urban mortality, its impact appears different when analysed according to variations in surrounding conditions, time and space. Observations in development economics regarding the importance of complementary inputs, such as basic disease avoidance skills and quality of living environment, seem relevant for historical analysis as well. More manageable and more developed, early twentieth century small towns were more responsive to the more mature water technology installed than their troubled nineteenth century counterparts. Combining national mortality statistics with our estimations enables an experimental accounting of the strictly measurable part of the contribution of the urban sanitary interventions to overall mortality trends. In the twentieth century, this measurable contribution reached 8–14%. In this period, the weight of urban IMR decline in total CDR decline was augmented over and above the growth in urbanisation by an acceleration in the decline itself. Our results suggest that urban sanitation did have a significant role in this acceleration. However, in the case of Finland, it seems placing much emphasis on the role of urban sanitary investment in the initial nineteenth-century onset of mortality decline, 298 J. PELTOLA AND S. SAARITSA as has been done in some previous literature, is not well founded. The large and expanding cities that first started to develop water infrastructure in the nineteenth century were challenging environments compared to the smalland medium-sized towns that followed suit in the increasingly prosperous twentieth century. The early technology was less functional and more risky. Focusing on average estimates over the long run without accounting for such variations would preclude the observation that populations with the most acute problems seemed to gain less from sanitary interventions than those already better offwhen such efforts first began. Investments were more effective and their measurable returns higher in the later period, when necessary complementary aspects of development had already improved over those prevailing in the nineteenth century Grand Duchy. Notes 1. The 37% effect in log points reported by Alsan and Goldin (2015,p.18)hasbeen converted to percentage change for comparability using figures from (Alsan & Goldin, 2015, p. 3) and footnote 9. See Cutler and Miller (2015) for technical discussion. The original publication by Cutler and Miller in 2005 presented an estimate of a 74% share of the decline in infant mortality attributable to sanitation (p. 13). Alsan and Goldin pointed out in 2015 that their way of interpreting regression coefficients on log IMR directly as percentage change was not appropriate for large, discrete changes, and that the transformation (e β -1) should have been applied (p. 1, fn 3). This led Cutler and Miller to revise their estimate to 59% in an erratum (2015). Most recently, Anderson et al. (2018)have acquired the original data and code from Cutler and Miller, and seem to be suggesting that after correcting a large number of errors in e.g. timing of interventions and data transcription, the appropriate unweighted estimate could in fact be closer to 2% (although joint effects or their standard errors are not directly reported by the authors). With the population weighted specification preferred by Anderson et al., this could potentially go up to 8% assuming the joint effect is statistically significant (Our computations from Anderson et al., 2018, Table 15, columns 5 and 6,; Cutler & Miller, 2005,p. 13, Table 5). 2. See section 2.3 on historical Finland; on Sweden, see Bengtsson, 2009,p. 150, fn 16; Oris et al., 2009, p. 36, fn 17; for current global estimates indicating rates well below 100% for exclusivity during the first six months, see the WHO database at http://apps.who.int/gho/ data/node.main.52?lang=en . Even partial breast-feeding could still have provided additional protection against infections also in the case of exposure. 3. While Davenport, Satchell & Shaw-Taylor (in this issue) argue against a link between water quality and the IMR in mid-nineteenth-century England, the contextual nature of the possible protective factors they identify, such as using boiled water or food handling practices, and the number of significant estimates of IMR across many other cases, suggest this connection is still widely plausible. Channels are explicitly discussed in Alsan & Goldin, 2015, p. 14, for instance. 4. While some direct association of sanitation with the crude death rate (CDR) is detectable in the nineteenth century, it is not sufficient enough to build on. Cause-specific mortality data are only available at the city level from 1896 onwards; the truncation is unhelpful for the task of evaluating the role of urban sanitation in the mortality decline starting from the 1870s. 5. Kari Pitkänen has observed that this was mainly due to high mortality among infants born out of wedlock, whereas for children born to married couples the penalty ended already in the years 1916–20 (Pitkänen, 1983). 6. Archival sources document detailed technical discussion of and reporting on planning and the preparations for introducing piped water in Helsinki and Tampere in reports to THE HISTORY OF THE FAMILY 299 municipal authorities, including comparisons of filtration methods, chlorination and the health hazards of lead pipes. (Kansalliskirjasto, Pienpainatekokoelma 1 B, Kunnallishallinto, Teknilliset laitokset, 1810–1944.) 7. In Tampere, wastewater from such a scheme was dumped into the Tammerkoski river, a potential source of water for other residents. Tampereen kaupungin terveydenhoitolautakunnan kertomukset, 1883–1904, Tampereen kaupungin arkisto. 8. It is not immediately clear from the manuscript what the metrics were for shares of cities covered by sewers. The source was a query sent to municipalities Backman, 1923, p. 97. 9. In some cases, the description is not clear and the author had still provided the highest score on the index that he applied to eight of the cases, even when the lack of service for suburbs was specifically mentioned. 10. The previous analysis is not always unequivocally convincing in light of closer scrutiny of the statistics. In Helsinki, Turpeinen (1995) attributes the fall in overall mortality in the early twentieth century to improved filtration from 1909–1910 onwards. However, there was a marked collapse in the number deaths from waterborne disease, particularly enteritis, already in 1902. The closure of certain unsanitary wells in working-class quarters was a likely culprit (Waris, 1934). 11. The city of Lahti, which was granted city rights only in 1905, was dropped from the regressions due to a lack of figures for constructing a population density variable. 12. Fifty observations are missing, while the panel regressions were run with approximately 2500 observations for the entire period. 13. Parish-level tables for population changes, National Archives. We collected the data from microfilms in a mirror archive at Statistics Finland. 14. See Alsan & Goldin, 2015,footnote 3, and Cutler & Miller’s 2015 erratum. 15. Credit for these important points belongs to Jonas Helgertz and Andrew Hinde (conference discussions). 16. Thanks to Susan Hautaniemi Leonard for bringing this up. 17. However, on the ultimate size of effect in Cutler and Miller’s work, see Anderson et al., 2018. The separate estimate for tap water is also close to that of Ogasawara and Matshushita, 2018, p. 206 for Japanese cities the years 1922–1940, although the period, measurements and specifications differ. Ogasawara and Matsushita estimated approximately 13.5% of the decline in IMR could be attributed to an increase in tap water consumption. While the specifications again differ significantly, Helgertz and Önnerfors also present separate estimates of the effect of water and / or sewage on IMR in this issue, with the effect being at approximately 6% in their basic model for Swedish cities adopting such systems in the years 1875–1930, and with consistently higher estimates for waterborne disease mortality. The experiments on the truncated Finnish data available on waterborne disease mortality have yielded no significant results. 18. In fact, piped water has a significant coefficient with the wrong (positive) sign. 19. The series available in SVT (Official Statistics of Finland) includes craft workers until 1908, after which only industrial workers are included. The series was constructed by applying the yearly percentage changes in the early data to a backwards-extrapolated industrial workers’share. 20. With the exception of chlorination, which has collinearity issues in the last specification. 21. The lack of statistical significance for individual coefficients is standard in the literature. 22. The mean of joint effects coefficients for models IV and V for the population weighted regressions. 23. Unreported, available on demand. Acknowledgments Funding from the Academy of Finland is acknowledged by Saaritsa. The authors would like to thank the participants of the session “Effects of Sanitary Interventions on Mortality Outcomes”at the 41st annual meeting of the Social Science History Association in Baltimore, MD, in November 300 J. PELTOLA AND S. SAARITSA