Public water and sewage investments and the urban mortality decline: - - PDF document

Public water and sewage investments and the urban mortality decline: Sweden 1875-1930

*** Draft, please do not cite or circulate *** Jonas Helgertz Centre for Economic Demography and Department of Economic History, Lund University Martin Önnerfors Centre for Economic Demography and Department of Economic History, Lund University Financial support from the Crafoord foundation for the project “Effects of exposure to health shocks and interventions during the fetal stage and the first year of life on labor market outcomes in Sweden 1900-2012” is gratefully acknowledged. We thank Jonas Wallin and Siddartha Aradhya for helpful comments on

• modelling. For the digitization of data, we acknowledge the assistance of Federica Braccioli, Magda du Toit,

Anita Terzic and Glen Williams.

1 Introduction and aim

The mortality decline that started in the 18th century, and led to an unprecedented rise in life expectancy in Europe and other parts of the world, has been considered to be one of the most significant events in human

• history. In Sweden, the mortality decline began during the early 19th century, being driven by rapidly

declining infant mortality towards the latter half of the century. From the end of the 19th century, infant mortality declined from about 17 percent to less than five percent in 1930, less than half a century later. Coupled with declining fertility, the (first) demographic transition, had substantial consequences at the macro as well as micro level, through reducing the dilution of capital per capita as well as enhancing labor productivity through increased human capital formation (Galor, 2012). In combination, these were fundamental processes for industrialization and rapid economic growth during the 19th and 20th centuries, resulting in substantially changing standards of living. Despite the importance of the mortality decline for the emergence of today’s developed countries, the understanding of which factors drove this decline remains poorly understood. Robert Fogel remained hesitant to call it one of the greatest human achievements, as he was unsure regarding how much of this development was due to human intervention (Fogel, 1986, p.376). The question was not, he argued, whether factors such as economic development, medicine, public health or nutrition were involved in the decline, but the contribution of each respective factor to the decline. Due to the historical importance of the mortality decline, substantial research has been conducted into its

• determinants. A noteworthy characteristic of the mortality decline was that it typically occurred with a

distinct delay in urban areas. This is important, as cities in Sweden and elsewhere across industrializing countries came to host an ever increasing share of the population, why the analysis of urban areas allows for conclusions that are more representative for the country as a whole. This paper examines a newly created database, with annual data covering all Swedish cities between 1875 and 1930. This period was characterized by a steady expansion in public water and sewage provision, as well as being a period when the urban mortality declined to reach rural levels. The data set is unique in the sense that it covers a full range of cities, all within the same national context – this will allow for a more complete picture The aim of the paper is not

• nly to quantify the relevance of the provision of piped water and sewage for the Swedish urban mortality

decline, but also to better understand how this process was affected by the implementation of more advanced water processing methods. It is expected that the provision of (clean) water not only caused an overall reduction in mortality, but more so in terms of deaths due to waterborne disease. The ability to drink and prepare food using clean/safe water, in combination with the ability to dispose of fluid and solid waste, should diminish the spread of waterborne disease, an important cause of death at the beginning of the period examined. The relevance of understanding the role of here investigated public health interventions is not only linked to

• btaining a better understanding of a process of significant importance occurring in the past. Whereas deaths

due to the consumption of unsafe water represents a remarkably rare event in today’s developed countries, it

2

continues to be a problem in many other parts of the world even today. According to the WHO, 1.8 million people die every year from diarrhoeal diseases related to unsafe water supply, sanitation and hygiene (WHO, 2014). The sanitary conditions in several developing countries today are in many ways similar to the disease environment found in western cities in the end of the 19th century (Ferrie & Troesken, 2008, p.2), which makes study of the mechanisms behind historical urban mortality relevant also in a contemporary setting. There is, however, a need for more information on the effectiveness of interventions in improving water and sanitation, to distinguish them from other interventions in reducing mortality (Fewtrell et al., 2005, p.42). The paper is organized as follows. Initially, we will present a section describing the core features of the mortality decline and explanations encountered in the previous literature. Subsequently, we will present existing evidence which links water and sewage provision to mortality, followed by the history of investments into water and sewage in Sweden. Following this, we will describe the data used, as well as the key variables and methods used in the multivariate analysis. Lastly, the results, sensitivity analysis and conclusions sections close.

2 Background

2.1 The mortality decline: hypotheses and empirical evidence

This paper focuses on a period which begins when Sweden has already experienced approximately 70 years

• f steady mortality decline. The period here examined takes place during the third phase of the demographic

transition, characterized by declining fertility as well as mortality. Compared to when the mortality decline was initiated, at the beginning of the 19th century, the crude death rate had declined from between 25-30‰ to around 20‰ in 1875. Indeed, by the end of the period analyzed – in 1930 – CDR had reached a level that is comparable with the level today. The pattern of decline for the country as a whole, however, disguises considerable heterogeneities between urban and rural areas. As elsewhere in the industrializing world, mortality tended to be higher in urban areas due to a range of different factors, including a higher degree of crowding as well as being hubs for migration flows, facilitating the spread of disease (Reher, 2001). As a consequence, the mortality decline occurring during the time period analyzed in this paper is to a significant driven by improving survival in the cities. This is illustrated by Figure 1, showing annual infant mortality rates calculated for urban and rural areas, separately. In the beginning of the analysis period infant mortality in urban areas were around 30 percent higher, having converged to the rural levels by 1930. Figure 1 here. To some extent, the mortality decline was a process that occurred simultaneously across Europe and North

• America. The reasons behind the mortality decline have been widely debated, and a number of potential

explanations have been put forward. One important hypothesis links the declining mortality to increasing standards of living. An important proponent of this view was McKeown, Brown & Record (1972, p.382), who argued that improved nutrition and economic growth were the primary cause of mortality decline. More

3

specifically, they conclude that the reduction in mortality in 19th and 20th century Europe was “probably due to a significant increase in food supplies”, and rule out other factors by comparing their timing to the timing

• f the mortality decline. The main cause for the improvement in nutrition was major agricultural reform,

allowing for substantially more efficient farming and increased yield. This conclusion has, however, been criticized: Aaby (1992, pp.174–175) argues that the importance of nutrition in the mortality decline was greatly exaggerated, and also thinks that his “negative exclusion” of other factors is an unsatisfactory method

• f reaching a reliable conclusion. Easterlin (1999, p.265) also opposes the conclusions of McKeown and

colleagues, by arguing that the time-scale used in reaching the conclusions is wrong: the take-off dates for economic growth that paved way for better nutrition does not coincide with the time-scale of the mortality decline in the countries that McKeown references to. A more recent opinion is given by Soares (2007, p.254), who notes that although McKeown is right to some extent, there are still substantial parts of the mortality decline that cannot be explained by nutrition or economic growth. Another central theory that aims to explain the mortality decline focuses on the influence of improvements in (health) care and medicine. The link between breastfeeding and infant mortality is well established, serving to increase the child’s resistance to infections during a period of great susceptibility to disease (see for example Woods, Watterson & Woodward, 1988). Evidence supporting the hypothesis for the mortality decline is, however, scarce, plausibly linked to the unavailability of data. The effect of medical advances on mortality is not as disputed as the nutritional argument, at least not when it comes to timing. Easterlin (1999, p.269) argues, along with McKeown & Record (1962), that medical advances had little to do with mortality decline until earliest in the late 19th century. From this point on, however, Easterlin (1999, p.273) argues that medical advances such as methods, vaccines and drugs led Europe into the first phase of the epidemiologic transition (as described by Omran, 1971). However, Razzell (1974, p.11) argues that the inoculation for smallpox (which is also mentioned by McKeown) at least should be considered as a factor in the decline of mortality in late 18th/early 19th century. From the late 19th century and onward, Lee (2003, p.170) sees the progress in medicine as the most important factor in reducing mortality: he mentions that the 20th century is the century when the fatal infectious diseases are brought under control through medical methods, vaccines and drugs. A factor that has also been lifted in the debate as potentially important for the mortality decline is the proposed diminished virulence of some critical viruses (the smallpox virus, for example): when viruses for some reason become less aggressive during the 18th and 19th century, mortality is also diminished (PHAS, 2005, p.49). This (supposedly spontaneously) decreased virulence would speak against the effect of medicine, and has been argued by Fridlizius (1985) to be of special importance in the case of the early Swedish mortality decline.

2.2 Public health investments and mortality

Historically, publicly coordinated attempts of promoting health can be linked to the prevailing views on how

4

disease spreads and causes mortality. Indeed, the idea of contamination as a physical concept has existed in various forms throughout history, namely that there exists an exogenous factor which causes illness and death, as well as how this is spread. Before the break-through of bacteriology in the 1870s, the dominant theory of disease was the miasma theory (Cutler & Miller, 2005). According to this, disease was primarily spread through vapours (miasmas) that also carried an offensive smell. Cutler (ibid) notes that this view seems to be based on a kind of Pavlovian learning, where it was noted that people being exposed to foul

• dours were more often sick.

With the breakthrough of the bacteriological view on communicable diseases, authorities started to assume a different position concerning public health interventions. Across the cities of the industrializing nations of Europe and North America, a common intervention was represented by the construction of underground sewage and water lines, intending to provide the city’s inhabitants with water deemed safe for drinking and

• ther domestic use, as well as the removal of solid as well as liquid waste. Studies on the role of public

health improvements into clean water technologies and sewage provision on mortality have been conducted

• n several geographical settings. Alsan & Goldin (2015) study the impact of clean water and sewage

technologies on infant mortality in the Boston area during the late 19th/early 20th century. Using a difference-in-differences framework, they find that the effects of water and sewage systems act as each

• ther’s complements rather than substitutes. Indeed, the effect of the provision of one becomes reinforced by

the provision of the other, and with quite a substantial effect on infant mortality. The results suggest that 37 percent of the total decline in infant mortality can be attributed to the introduction of clean water and sewage. Apart from a general effect, Alsan & Goldin (2015, p.16) also find different effects on mortality based on socioeconomic status, where the more marginalized Irish population has experienced a greater decline in infant mortality as a result of being exposed to both sewage and water services compared to the population of British descent. Cutler & Miller (2005) also analyze the impact of water and sewage provision as well as the use of filtration and/or chlorination processing methods in thirteen US cities on mortality between 1900 and 1936. The authors consider its influence on both overall and child mortality and mortality in typhoid fever, with the latter being used as proxy for water-borne diseases. The combined use of filtration and chlorination methods is found to exercise a substantial influence on all studied outcomes, from reducing child mortality by 50 percent to total mortality, amounting to 13 percent. The results remain robust to a range of sensitivity tests, also suggesting quite a considerable social rate of return. Despite the caveat associated with how the parameters used to calculate the social rate of return were obtained, the authors claim that the results indicate a 23:1 rate of return on investment (95% CI; 7:1 – 40:1). Using typhoid (mortality) as an instrument of water quality exposure during early life in 75 US cities, Beach et al. (2016) investigate its influence on the human capital formation process. Examining outcomes among males observed in 1940, their findings suggest that the eradication of typhoid fever through the adoption of clean water technologies increased educational attainment by one to nine months, and earnings by up to nine

5

percent. Beemer, Anderton & Leonard (2005) also examine the United States, studying the mortality differences between different city districts in Northampton, New England in the late 19th century. At the start of the study period, the city was characterized by harboring a large industrial area, the waste from which was not disposed of by means of a sewage system. At the same time, the mortality rate for people residing in the industrial area was almost twice as high compared to the residents of inner-city commercial districts. After the completion of a sewage system which included the industrial areas, the mortality rates became almost equal across the city. While the methods and the data used do not allow for making causal inference, having controlled for SES, gender and literacy, the authors underline the sewer coverage as plausibly being a main determinant of the decline in mortality in the city. Using annual statistics between 1899 and 1929, Cain & Rotella (2001) investigate ho, w investments into various sanitary measures – water, sewage and refuse collection and disposal – influence waterborne disease mortality in the US. Their findings suggest that a one percent increase in each of the examined sanitation types would have saved 18 lives in an average-sized city. Troesken (2001) also focuses on waterborne disease mortality in 13 southern US cities, 1889-1921, finding suggestions of discriminatory provision of water services through the effects of public ownership. Turning to a European context, Kesztenbaum & Rosenthal (2014) examine the relationship between sewerage provision and life expectancy in 80 Parish neighborhoods between 1880 and 1913. Their results suggest nontrivial effects, with a one standard deviation increase in the fraction of buildings connected to the sewerage network being associated with an extended life expectancy at age one amounting to about two

• years. For England, while generally stressing the role of an improved diet as the primary force, McKeown &

Record (1962, p.120) also conclude that improved public sanitary conditions were an important driver behind the decline in water-borne disease mortality during the late 19th century. Van Poppel & Van der Heijden (1997) examine the Dutch city of Tilburg, failing to find a negative relationship between the provision of piped water on infant and child mortality during the early 1900s. For Sweden, there exist a number of case studies on selected cities, examining the mortality consequences of public investment into health and sanitation. Nilsson (1994) examines the city of Linköping, first obtaining water and sewage services in 1876, as well as describing its decision and implementation process. The findings suggest that infant mortality declined faster in the neighborhoods that were first connected to the water and sewage mains, also characterized by a lower death rate in gastroenteric diseases. In contrast, Castensson, Löwgren & Sundin (1988) examine four Swedish cities, however, failing to find any noticeable connection between the construction of piped water and mortality. Molitoris (2015) uses individual level data for Stockholm, covering the time period 1878-1926, examining socioeconomic differences in cause specific

• mortality. The findings resemble those published by Burström et al. (2005), with a disappearance of the

socioeconomic gradient in food and waterborne disease mortality (diarrheal in Burström et al). While unable

6

to pinpoint the mechanism which drives this development, Molitoris (2015) suggests the expansion of access to piped water occurring during the time period as a possible explanation. Burström et al. (2005) suggest that technical the technical water improvements together with public education and sanitary laws create a synergy that reduce waterborne mortality, but that the technical improvements alone would be insufficient. Nilsson (2013, p.108) writes, on the specificity of the Swedish mortality decline and water/sewage interventions, that these interventions did most likely not start the decline in waterborne mortality, but rather helped to keep bringing down the urban mortality during the whole period. He stresses other factors, such as rising living standards and advances in medicine, as the main drivers of the urban mortality decline. Sundin et al. (2005, p.47) makes a similar argument: Sweden went into a different mortality regime starting in the 1810:s, and experienced a steady decrease in mortality from then on. The raging cholera epidemics of the 1830:s (ending with the last major epidemic in 1866) served as a strong motivator for cities to invest in water and sewage systems, but the systems did not by themselves start the decline. Apart from a general increase in living standards, Nilsson mentions the increase in literacy as an important factor behind the mortality decline.

2.3 History of water and sewage provision

This study examines the time period 1875-1930, a period during which Sweden experienced rapid urbanization alongside industrialization. Over the time period examined, the population in locations designated by the Swedish authorities to be cities increased from about 620,000 to over two million, corresponding to an increase from about 14 to 34 percent of the total population. Despite Sweden following a trend that could also be observed elsewhere in the industrializing world, the typical size of the Swedish city still remained relatively modest. As late as 1930, the median population of the 114 Swedish cities only amounted to roughly 7,000 individuals, with only 20 percent of the cities hosting a population exceeding 15,000. The growth of the Swedish cities naturally to a great extent consisted of the in-migration of individuals from urban areas, attracted by the opportunities in the growing industrial and service sectors. Schön (2000) illustrates how about 70 percent of the Swedish work force were employed in agriculture around 1875, declining by about 50 percent until 1930. Access to water in 19th century urban areas, before the introduction of piped water, often came from public wells (Edvinsson 1992, p.115, Sundin & Willner 2007, p.137). Apart from this groundwater supply, surface water (such as nearby rivers and lakes) was also used in households - and both of these types of local water sources created a vulnerability to human-induced water pollution (Castensson, Löwgren & Sundin, 1988, p.280). Due to this, the groundwater from which the wells were supplied could be contaminated by inadequate waste handling, which led to reoccurring outbreaks of epidemics from water-borne diseases. The first Swedish city to implement a rudimentary piped water delivery system was Uppsala, where a wooden pipe system was built in 1649. Already before the take-off of the urbanization process, Swedish authorities recognized that substantial

7

measures were required to combat the sanitary conditions believed to be associated with the elevated mortality which characterized urban areas in Sweden. As early as 1857, according to a royal decree, the lack

• f “good” water for drinking and other domestic purposes was identified as a key reason for the presence of

disease and epidemics. In underlining the extent of the sanitary issues facing – in particular – the urban areas, the decree also highlighted the health consequences of overcrowded cemeteries - especially when in proximity to residential houses - and the accumulation of refuse in public areas. Thus, despite the lack of a full understanding regarding the mechanism linking exposure to “dirty” water or various forms of waste, such contaminants were considered to be synonymous with poor public health. However, it was not until 1874 – plausibly linked to the breakthrough of bacteriology and the improved understanding of the mechanism of disease that followed - when centrally coordinated attempts at publicly provided and planned health care in the Swedish cities started to become a reality. Following the passing of Hälsovårdsstadgan, health boards were to be established in all cities, commissioned to monitor the health and sanitary conditions as well as to making sure that laws were being adhered to. The health boards always had as permanent seat for the city’s head medical doctor, in order to ascertain the necessary competence in regards to the spread of disease and treatment methods. Whereas the health board had limited formal responsibilities for city council decisions on public health investments, their de facto influence on health practices through, for example, the ability to fine violators of Hälsovårdsstadgan, was considerable. The following decades came to be characterized by considerable public health interventions (Edvinsson, 1992, p.75). As the health boards became established across Swedish cities, the mandatory inspection of meat, pork and milk alongside the creation of isolation/epidemic hospitals became common sight. Arguably, the most noticeable change in the cities were the work going on in the streets in laying down water and sewage mains as well as service lines in order to provide their inhabitants with a comparatively abundant supply of water and the ability to get rid of liquid (and solid) waste. Clearly, establishing a functioning system for the delivery of water directly into existing buildings, residential and otherwise, as well as the removal of waste, came at a substantial cost. Indeed, usually, a sewage management system was constructed in conjunction with, or not long after, the installation of a water-carriage system in the city (Castensson, Löwgren & Sundin, 1988, p.284). Most frequently, the decision to construct a water and/or sewage system was made by the city council, also implying that it was to be publicly financed (Lindman, 1911). In certain cases, such as Borås, private entrepreneurs were responsible for the construction of the city’s water provision network (Berglund & Palm, 2005). Lindman (1911) provides an illustration of the pace with which water and sewage systems came to characterize Swedish cities. More specifically, whereas only seven out of Sweden’s approximately 100 cities were able to provide their inhabitants with piped water by 1875, only 28 cities lacked these services by the end of 1909. The laying down of sewage mines was a slower process initially, but gathered pace rather quickly. As a result, only 12 cities were completely without these services by 1910. Although the knowledge of disease transmission through water was present, the motivation for introducing piped water in a city often also concerned other aspects than the improvement of public health (Edvinsson,

8

1992). For the city of Sundsvall, the question of whether to establish a piped water system was first raised in 1858, pleading fire safety, industry water supply, and public health as motivation. As in several other cities, the suggestion was, however, initially voted down by the council (ibid., p.116). Questions about costs, taxes, and coverage led the question to remain disputed in the council until its acceptance in 1877, some 18 years

• later. Bjur (1988, p.89) tells a similar story of political inertia in Gothenburg, and how officials eagerly

followed the miasma/bacteriology debate while trying to decide on the implementation of piped water in the

• city. Lindman (1911) describes a discussion regarding the need for water and sewage mains going on more or

less in parallel across all Swedish cities during the late 19th century. Furthermore, he describes what was frequently a considerable delay between decision to implementation. In the city of Skellefteå, the decision to provide the inhabitants with piped water took place already in 1876, but was not constructed and ready for use until 1895. A similar story of extensive debate and delays of implementation of water technologies was also common in American cities during this period: Cutler & Miller (2005, p.6), citing McCarthy (1987), mention the city of Philadelphia, where it took 20 years from decision to the implementation of water filtration methods. A common solution for supplying water to a city was by means of pumping from a nearby lake or river, or relying on the ground water. Empirical studies have suggested that the introduction of piped water and sewage not always lead to a direct improvement of the individual citizen’s ability to enjoy a supply of safe

• water. Duffy (1992) describes the non-optimal sewage handling system that was first constructed in

Philadelphia: The primary sewage outfall was emptied in a river upstream, and close to, the city’s primary water intake, which created a circular water system. Auto-contamination of drinking water in Swedish cities has also been documented. A satirical picture of the early water supply in Stockholm can be seen in Figure 9,

• Appendix. The picture shows how one woman disposes of waste water into a sewage system, connected to

the same body of water from which another gentleman obtains his drinking water (Söndags-Nisse, 1866). Along with the construction of piped water and sewage, the need for purification of the water consequently arose rather quickly. Cutler & Miller (2005, p.5) mentions three common methods: sand filtering (also known as slow filtering), mechanical filtering (rapid filtering) and chlorination. In Swedish cities, all of these methods have been used, but in different combinations and forms. In Göteborg, for example, it was discovered in after water pipes had been built in 1890, that the water quality in the water source was unsatisfactory: this meant that the project was delayed so that a water treatment plant could be built (Bjur, 1988, p.132). The plant, which used slow filtering technology, started delivering water to the city four years later (ibid., p.134).

2.4 Spread of waterborne diseases

A general, and logical, view on cities and diseases is to assume that higher population size and density will affect the mortality from waterborne disease, by creating a disease environment that facilitates disease spread (WHO 2008, Alirol et al. 2011). The details of how this actually works might, on the other hand, be complex,

9

and the transmission and persistence of waterborne diseases within a community can be modelled in many different ways. The pattern of transmission will vary with the characteristics of the disease, but can consist of a single environmental reservoir (“water/food → person” transmission), between infected and susceptible individuals (“person → person transmission”) and also a more complicated mode wherein an infected individual spreads infection into a water source which is consumed by a susceptible individual (“person → water/food → person” transmission), and also hand-to-mouth transmission through, for example, a restroom (Tien & Earn 2010, Singh & Sharma 2014, p.95). Within the field of spatial epidemiology, several types of models exist, from ones that assume a static population with homogeneous mixing, to a fully interactive model with random exchanges and full mobility of individuals – in order to try and predict transmission and diffusion over space and time (Brockmann, David & Gallardo, 2009). Within the scope of this study and using the current (aggregated) data, it is not possible to get closer to any specific spatial dynamics of disease

• spread. However, based on the fact that population size and density of a community is an important part of

the theoretical models of disease spread, it seems reasonable to assume that the Swedish cities will have a lower risk of spread of waterborne disease compared to previous research on larger cities (for example, Cutler & Miller 2005). Figures for the average urban population density of Sweden exist from 1880 and

• nwards, and was 563 persons/km2 in 1880, peaked at 908/km2 in 1900, and came down to 155/km2 in 1950

(SCB, 1969). This average was probably largely driven by Stockholm, which had a density of 9638/km2 in 1900, but specific data for other Swedish cities does not exist. The average Swedish population density can be compared to the most densely populated cities in the world today, which are well above 15.000 persons/km2. Considering the focus of most previous research on larger cities, the Swedish experience represented in the current dataset will contribute with a new view on waterborne disease mortality reduction (that is, the experience of comparatively smaller cities) - which can be assumed to be different based on the dynamics between population size and disease spread. On the same note of disease spread, water and a sewage systems can theoretically have a separate and/or combined effect on mortality. As discussed by Alsan & Goldin (2015) and Duflo et al. (2015), there are both epidemiologic complementarities and externalities connected to the different combinations of water and

• sewage. Sewage systems create a barrier by transporting away possible contaminants from the urban

environment, lowering both the risk of contaminating a reservoir, and person→ person contamination. Water systems create another barrier by providing clean water to the households, lowering the risk of person→ person contamination. Together, the systems create multiple barriers and are expected to be complementary to each other rather than substitutes (Alsan & Goldin, 2015). Whether or not this was the case in the Swedish context will also be addressed in this study.

3 Data and methods

The data used in this paper has been obtained through the digitization of several historical publications, providing annual information for all Swedish cities on a set of essential variables. The outcome variables

10

examined in the analysis represent various measurements of mortality, including the crude and infant mortality rate. Whereas both outcomes display starkly improving health over time, the ability to examine cities separately allows for interesting heterogeneities to emerge. As displayed in Figure 2, differences between cities representing the 10th and 90th percentile, respectively, in terms of the infant mortality rate (IMR) is substantial (the 10th and 90th percentile being represented by the upper and lower edges of the grey area around the black median line). In 1875, around a median IMR of 167 deaths per 1,000 live births, the cities representing the 10th and 90th percentile, respectively, are characterized by about 100 infant deaths less/more. Over the time period examined, the IMR is characterized by an almost linear decline, to a median

• f about 50 in 1930, but with an increasing relative difference between high and low IMR cities. A similar

mortality decline is displayed by the crude mortality rate (CDR), with a median decline of about 50 percent

• ver the time period, from 22.2 to 11.1 deaths per 1,000 (Figure 10, Appendix). The CDR, however, appears

to be considerably less context specific, as the relative difference between high and low mortality cities is declining until the turn of the century, thereafter remaining constant. Figure 2 here Of potentially even greater relevance is the information on cause specific mortality, which allows for distinguishing between causes of death depending on whether they can be expected to be directly linked to the provision of piped water and sewage. Here, we separate between two main causes of death, both sharing a common declining trend over time, but where the provision of water and sewage only plausibly can be directly linked to the decline of one. We distinguish between deaths due to waterborne diseases (typhoid fever, dysentery, general gastrointestinal diseases, cholera, polio) and due to airborne diseases (measles, smallpox, tuberculosis, whooping cough), where only the former is believed to be causally linked to water and sewage access. Figure 3 shows how both causes of death changes over the time period examined. Approximately one third of all deaths occurring in Swedish cities in 1875 were due to waterborne or airborne diseases, to be compared to only 15 percent in 1930. Whereas both causes of death are characterized by a commonly declining trend, the time period examined represents the virtual eradication of mortality due to waterborne disease. Airborne disease mortality clearly remained more resilient, remaining at around 1 per thousand even at the end of the period. Figure 3 here The aim of this paper is to understand to what degree the expanding provision of water and sewage services were responsible for the urban mortality decline. Annual data on the city level has been digitized, allowing for the identification of the year during which the inhabitants of a city were able to benefit from these

• services. Key variables are thereby represented by dichotomous variables, indicating whether piped water

and/or sewage services were provided in city i during year t. The main source for this information, yearly publications from the Svenska kommunaltekniska föreningen (SKF) from the year 1904 and onward, has been supplemented with information from the Contributions to the Official Statistics of Sweden (BISOS) as

11

well as with Lindman (1911). The latter source is of particular importance, as it covers the time period before 1901, offering a written account by the health board of every Swedish city. The agreement across the different sources as regards the provision of piped water is very high, with the exception being the occasional

• ne-year discrepancy in terms of when a city’s inhabitants started enjoying access to piped water. This is also

in line with the list of priorities outlined in the Hälsovårdsstadgan of 1874, clearly outlining the necessity for such services being provided to Swedish urban areas. The definition of the provision of sewage emerges as somewhat less precise, possibly also linked to this type of intervention being less precisely defined. Whereas the timing of the provision of sewage has been confirmed across all available sources, it appears that what is labeled as sewage at least during its early days sometimes refers to piping intending to provide drainage and divert liquid waste material from the dwelling. In fact, the introduction of the WC was initially met with considerable skepticism by the authorities and in many cities remained prohibited until into the 20th century. Figure 4 here Figure 4 shows the expansion of water and sewage provision across all Swedish cities during the time period examined in the paper. Clearly, the expansion of both services were occurring largely simultaneously, with a coverage increasing from around ten percent of the cities to about ninety percent. An interesting observation is that the expansion of these services appears to slow down towards the end of the 1920s, with cities that have yet to provide their inhabitants with water and sewage failing to do so before the end of the period

• examined. Since the presence of water and sewage systems for most time periods was either both or none,

creating two dummy variables of these would also create problems of distinguishing between the effects – which is why a combined categorical variable is used. Based on a method by Gabadinho et al. (2011), four city clusters of water/sewage implementation sequencing1 have been identified, covering all cities in the

• sample. As can be seen in Figure 5, the sequences where cities have either water or sewage are short and

transient in all groups except the group named “Slow adopters”, in which the cities keep only sewage for a longer period before installing also a water system. Based on this, it was decided that the implementation will be mainly modelled using a dichotomous indicator of “no water, no sewage” or “both”, since the interpretation of the transient stages as coefficients would not be straightforward. Models using one of the sequence clusters, to address the question of whether the water and sewage systems are complementary or substitutes, will be presented in the Results section. Figure 5 here A fundamental assumption for the analysis is that between-city differences in the timing of implementation

• f water and sewage services can be treated as a quasi-experiment, thereby allowing for the results to be

interpreted in causal terms. To this end, we rely on that the timing of implementation of water or sewage services in a city can be considered as “plausibly” exogenous and determined by arbitrary events in a city (Cutler & Miller, 2005). This would mean that, for example, city A and city B might be alike and have

1 The clusters can be explained as groups of cities with similar behaviour when it comes to how much time is spent in each category before upgrading to the next.

12

similar preconditions for implementing clean water technology, but city A might be 10 years slower in implementing because of a different political process. Assuming that this political process does not vary with the current variable of interest (mortality), this difference in timing can be exploited as a quasi-experiment2. Table 1, below, attempts to shed light on this potential process, by displaying median mortality rates for cities depending on whether they implement piped water services during a subsequent period. The death rates are calculated for cities which have not yet implemented water services at the beginning of the time period in question, displaying the median death rate during the preceding decade, depending on whether piped water was provided during the period. The figures should thus provide an indication as to whether cities that invested in piped water early or late were characterized by large differences in mortality rates prior to this intervention. Table 1 here The results show that mortality due to waterborne disease was higher in cities who were early providers of piped water, as indicated by the columns for the time period 1880-1889. At the same time, overall mortality (CDR) as well as infant mortality (IMR) were roughly the same in providing / non-providing cities. Moving forward in time, however, the difference in waterborne mortality switches from positive to negative, and does now show any clear signs of being systematic. A similar lack of a systematically higher (or lower) CDR and IMR among the cities that implement piped water services during subsequent periods reinforces the idea that it remains difficult to link mortality differences between cities to their propensity to invest in piped water. Another method of potential use in investigating the exogeneity of here examined public investments is represented by examining the time from decision to implementation of water provision. As argued in previous research from other contexts, due to the frequently time consuming process of implementation, the timing when a city’s inhabitants actually were able to start enjoying the services may be considered as exogenously determined. In the Swedish case, the median time elapsed from when the decision that water pipes were to be laid was made and until households first were provided with water was two years. This number is obtained from Lindman (1911), whose written account of the work of the health boards of Swedish cities in certain cases (n=40) allows for the identification of both the timing of decision and

• implementation. Whereas the median time until implementation was short, with a mode value of two years,

in 20 percent of the cases for which this information is available, this exceeds four years. Another possible threat to the validity of the study design is linked to the possibility that news of the construction of water and sewage technologies affected migration flows, affecting the demographic composition of the population. If this resulted in an increased inflow of individuals of better health than the

2 A problem with having non-random variation in timing of treatment would be that a city might implement a clean water system in the midst of a waterborne disease spike. The variation captured by the treatment timing might then be capturing a “process of mean disease reversion”, and not a variation associated with the implementation of the system (Cutler & Miller, 2005, p.10).

13

existing population, finding a beneficial health effect from examined investments will represent an

• verestimated effect, partly attributable to the improved health stock of the population. Figure 6 attempts to

illustrate to what extent the completion of piped water services across Swedish cities influenced migration

• flows. Each dot on the graph represents a separate city, and the horizontal (vertical) axis indicates the mean

population growth due to migration over the five years prior to (after) piped water services had been

• implemented. The diagonal line across the graph represents unchanged pre- and post intervention migration

rates, and observations above the line consequently represent cities where the population growth due to migration was higher during the five years subsequent to the implementation of water services, compared to the immediately preceding five years. The figure displays considerable variation, with the existence of cities experiencing increased as well as decreased in-migration during the period after the piped water services were implemented. Consequently, there is no evidence suggesting that the provision of such services resulting in dramatically changing migratory flows, neither into nor out of the cities having made these investments. Figure 6 here The initial analysis will focus on the mortality effects from the provision of basic water and/or sewage services to Swedish cities. Over the time period examined, the data suggests that the cities rapidly extended the provision of these services to cover the majority of the households under their jurisdiction. Table 2 displays the mean share of households being covered by the water and sewage services, by time since

• implementation. Already within two years, about two thirds of households in cities having made these

investments were covered, implying that it appears reasonable to attribute a decline in mortality taking place with the implementation of water and sewage services to improved and/or increased access to water and waste disposal. Table 2 here Apart from gradually expanding the coverage of the laid down water and/or sewage pipes, over a somewhat more extended perspective, cities also invested in different methods to be used in order to provide the inhabitants with high quality services. While some cities simply distributed the water directly from the water source, other cities made the effort of processing and purifying the water prior to distribution. These processing methods range from aeration and sand filters, to the usage of more sophisticated filtration methods and the addition of chemicals such as chalk and chlorine. The debate concerning water quality and how this is measured developed from the miasma theory, to a focus on chemical composition of the water, and finally to a focus on the quantity of coli bacteria (Petersson, 2005). This scientific development seems to have had an impact on the water processing methods implemented in cities over time. Apart from investigating the baseline influence from water and sewage provision, the analysis will also exploit aforementioned changes in water processing methods, displayed in Figure 7. The data on water processing in Swedish cities is only available for the latter part of the study period (beginning in 1901).

14

As seen in Figure 7, around 25% of the cities used basic water processing (filtration/aeration) at the start of the data collection period. It is possible that many cities were processing water at earlier stages of the period (which is suggested by Lindman, 1911), but there are not enough sources to motivate a general backward extrapolation of this data. The water processing variable will therefore be modelled only on the later time

• period. Also, the fact that some cities could exploit a water source consisting of high-quality naturally

filtrated groundwater (here referred to as natural filters) will be addressed in these models (as these cities would not need artificial filtration and might bias the filtration variable downwards). In about 75% of the cities – and representing an increasing share until at least the turn of the century – no active water processing methods were used in order to rid the water of harmful substances or otherwise unwanted characteristics (not including natural filtration). Various forms of filtration clearly represent the most common form of processing, spanning from a generic definition of “filtration” to more detailed descriptions, such as sand or clay filtration. The use of advanced processing methods did not take off until well into the 20th century, however, expanding to cover about 15 percent of the cities by 1930. This represents an interesting feature, as Swedish cities appear to have been comparatively late at adopting advanced water processing methods. Indeed, the U.S. cities analyzed by Cutler and Miller (2006) were characterized by the implementation of processing through chlorination at least a decade earlier than in

• Sweden. As is noted by Lindman (1911), no cities during the study period attempted to purify sewage water

– at best, the sewage water was transported to an outlet outside of the city using sewage pipes. It is only from the late 1940s and onwards where sewage processing methods such as chlorination are starting to be used. Figure 7 here The argument about sequencing, as was made above, can also be made for the water processing categories (these are categorized “No processing”, “Simple processing”, containing slow filtering and aeration, and “Advanced processing”, containing chlorination and mechanical filtering). For these water processing categories, clusters of cities were created based on the implementation sequences 3, and these can be seen in Figure 8. These clusters are similar in that there are late adopters (this group consists of almost the same cities as for the previous clustering), but apart from that, there is heterogeneity in the sequence of water processing within the clusters in Figure 5. From Figure 8, it is apparent that most cities were late in adopting advanced processing methods, and that in two of the groups, this adoption was preceded with a longer period

• f simple processing.

Figure 8 here The analysis will rely on a city-fixed effects framework, making use of a Maximum Likelihood (ML) estimator and taking equation (1), below, as a point of departure. Comparable previous research such as Cutler & Miller (2005) and Alsan & Goldin (2015) uses linear fixed-effects models with a logarithmic

3 Similarly to the clusters in Figure 5, models using one of the clusters in Figure 8 have been run and will be addressed in the Results section.

15

dependent variable (mortality rate), but the mortality rate distribution of this sample is heavily left-skewed (around 20% of the city-year observations have a 0 mortality count), which makes logarithmic transformation inadequate. The presence of 0-values in the mortality counts gives an indication of the broad representation in the data, since many of these 0 values naturally come from smaller cities (on which previous research is scarce), but it also calls for a different estimator. In presence of excess zeroes and an

• verdispersion4 in the data, as is the case here, a negative binomial distribution is considered more

appropriate than a Poisson distribution (Allison, 2009). Furthermore, since the data is a panel, the incidental parameters problem5 is sometimes mentioned as an issue when individual fixed effects are used (Angrist & Pischke, 2008, p.224). A proposed solution is to use a conditional negative binomial estimator (as proposed by Haussman, Hall & Griliches, 1984), but it has been argued that this type of estimator does not behave as can be expected of a fixed effects model (Allison & Waterman, 2002, p.265). Allison (2009, chap.4, pp. 14– 15) shows that using an unconditional negative binomial estimator in combination with outer product gradient correction yields results that do not suffer from incidental parameters bias, and this is the estimator that is used for this model. The main advantage of using a fixed-effects strategy is that it, in theory, accounts for all other major changes that did not vary across areas and over time in exactly the same way that clean water technologies did. The

• utcome variable is measured for city i in year t, and ranges from cause-specific waterborne disease

mortality counts to general mortality (all deaths) and infant mortality. Clearly, the expectation is that the strongest effect should be found for waterborne disease mortality, as this mortality measurement should be characterized by the strongest link between exposure to contagion and here examined interventions. The link to general mortality and infant mortality emerges as being likely, albeit somewhat less clear, due to the relatively small role played by waterborne disease mortality for respective measurement. Lastly, the link between piped water and sewage provision and airborne disease mortality can be questioned, due to such diseases being spread through different (and largely unrelated) mechanisms. This outcome is mainly included as a sensitivity test, although the risk of succumbing to airborne diseases, such as tuberculosis has been suggested to be substantially elevated among individuals whose defenses are weakened due to having previously experienced diarrhea or some other waterborne disease. The outcome variable yit is the predicted number of deaths within each mortality category. The key independent variable is a categorical indicator of a city i:s access to water and/or sewage provision in year t, which is captured by the β1 coefficient. In addition, the analyses include a vector of control variables (Zit),

4 The overdispersion is defined as the conditional (on the independent variable of interest) variance of the dependent variable being larger than its mean. 5 The incidental parameters problem refers the situation where the individual fixed effects also increase the total number of parameters in the model, and in an ML setup, this can bias the standard error downwards (Winkelmann, 2013, p.174).

16

including the city’s population6 and a variable capturing the share of factory workers, intended to serve as an approximation of a city’s degree of industrialization. Timing of other public health interventions that might affect mortality, such as food inspections and building of a hospital in the city, is also included as controls. All variables used in the analysis are presented in Table 6, Appendix. Lastly, the models include city and year fixed effects, modeled through dummy variables, capturing city and year specific characteristics, as well as a city and year (as a continuous variable) interaction effect. The latter allows for the capturing of a city specific mortality trend, as cities could differ, not only in terms of their baseline mortality regime, but also in how this changed over time. Consequently, aforementioned time and city fixed effects are necessary in order to being able to tease out the potential influence from water and sewage provision on a city’s mortality. Naturally, since a city-fixed effects setup is used, cities without variation in treatment can not be estimated:

• f all the 114 cities, 19 were treated before the period7 starts (here included the two biggest cities Stockholm

and Göteborg), and 11 cities were never treated during the period. The final sample of cities, also excluding three cities that changed administrative boundaries (resulting in an unnatural population variation), consists

• f 84 cities and 4474 observations.

4 Results

For comparative purposes, results from otherwise identical model specifications but estimated on different

• utcome variables are presented jointly. All models include the full set of control variables, as well as the

city and year fixed effects, and the city and year (continuous) interaction term. Table 3 shows the key output from the basic model specification for the whole period (years 1875-1930), estimating the influence of (any) provision of piped water and/or sewage services. Incident Rate Ratio coefficients are used for presentation, and these can be interpreted as the relative risk of an event occurring, where 1 is no effect (compared to the reference category, for categorical variables). Table 3 here Model 1 arguably represents the most relevant specification for the purposes of this paper, with waterborne disease mortality as the dependent variable. As expected, the provision of piped water and/or sewage appears to have exercised a negative influence on waterborne mortality, amounting to an approximately nine percent reduction (-0.087) for the whole period, with statistical significance at the five percent level. Proceeding to the remaining outcome variables, assumed to be less strongly affected by the investments into piped water and/or sewage, the results largely confirm this to be the case. The provision of water and/or sewage shows a five/six percent reduction general mortality and infant mortality, significant at the one and five percent level,

• respectively. The coefficient for airborne mortality is negative but close to no effect, and also statistically

insignificant.

6 The population is included in the model as an offset, which in the ML setup denotes the maximum number of events (in this case, deaths) than can occur. In the case of infant mortality, the exposure variable is number of births. 7 Some cities were treated after the calendar period starts (>1875) but were not included in the official mortality data until after they were treated – these cities have also been excluded.

17

The influence of different processing methods used to rid the water being provided from bacteria and other unwanted substances are examined in models 5-12, presented in Table 4. These models distinguish between whether the water was provided without any type of processing, as well as between comparatively rudimentary and more advanced processing methods. Since the data on water processing is available from 1901 and onward, the basic models from Table 3 are, for comparison, presented for these later years in models 5-8. For waterborne mortality, the percent reduction from water and/or sewage is approximately 23 now percent, however, this larger coefficient needs to be interpreted considering the lower mortality for this later period (as seen in Figure 3). Also, this coefficient is now only significant on the ten percent level. For the other three mortality outcomes, the coefficients are no longer statistically significant. The coefficients for water processing methods on mortality can be seen in models 9-12. As regards waterborne disease mortality, Model 9 confirms that already when the water was provided without any prior processing, the mortality effect was substantial (-16.2%), but significant only at the ten percent level. As expected, the use of processing methods, resulting in the provision of water containing fewer contaminants, is associated with an even larger mortality reduction. This is in particular the case when advanced processing methods were used, resulting in an almost 30 percent reduction in waterborne disease mortality. It should, however, be noted here that the baseline mortality when advanced methods were implemented had already declined considerably, why the reduction in absolute terms most likely did not exceed the initial effect of any water services being provided. Also, a test shows that the different water processing methods are not statistically significantly different from each other other than on a 20% significance level, so the differences between these categories should not be over-interpreted. Furthermore, as noted earlier, there are sequential differences within the timing of implementation of water processing categories, which will be discussed in the Results section. Table 4 here Turning to the remaining outcome variables, the results fail to suggest as strong a link between here investigated public health interventions and other measures of mortality. The IRR coefficients for general mortality and infant mortality changes from water processing are mostly close to 1 and statistically

• insignificant. The effect on airborne disease mortality is, interestingly, positive and significant on the ten

percent level for two of the categories, however, there seems to be no systematic pattern between the different water processing categories. Regarding heterogeneous treatment effects, several models have been executed using different interactions between city population size (at timing of treatment) and water/sewage implementation categories, but no interesting results have been found.

4.1 Sequence of implementation

The potential problems in interpreting the separate water/sewage implementation categories have been discussed above, but it might still be interesting to test whether having one or the other is different from having both – that is, whether the complement or substitute each other. As seen in Figure 5, the period of

18

cities having only a sewage system installed in, except for the cities in the “Slow adopters” cluster, mostly a very short period, and would therefore make an interpretation of discrete water- or sewage coefficients

• difficult. As a separate analysis, models8 have been run on a sample restricted to cities in the “Slow adapters”

cluster, and the results from this can be seen in Table 5. Model 13 (waterborne disease mortality) clearly shows a gradient between having only sewage, and both sewage and water. A test shows that the two categories are statistically significant from each other, which can be interpreted as an indication of a gradient effect from no system → only a sewage system → both a sewage and water system. This result suggests that sewage and water systems complement each other9. As seen in Model 14, there is also a gradient for the general mortality outcome, and as expected, this is of lower magnitude. Model 16 also shows a negative coefficient for the airborne disease mortality, but without any strong gradient between the implementation (also, the two categories are not statistically significant from each other). In a similar manner as with the water/sewage systems in general, cities spent unequal amounts of time in different states of implementation of water processing, and the sequence clusters for this has been shown in Figure 8 - the cluster “Incremental implementers” has been modelled in a similar way as the “Slow adapters” above, but no statistically significant results were found10 Table 5 here

4.2 Sensitivity analyses

In the analysis, the provided water has been considered as unprocessed unless the city has been using a constructed processing infrastructure. Consequently, cities that have provided their inhabitants with water that has been filtrated naturally through a ridge or similar has been considered to belong to the no processing

• category. To the extent that natural means of filtration are as efficient as artificial methods of filtration, the

coefficient for unprocessed water might be biased downwards, which is why models that classify natural filtration as a simple filter have been run, and the results can be seen in Table 8 (Models 21-24, to be compared with models 9-12, Table 4). The coefficient for simple filtration (including the natural filters) does not change in models 21-24 compared to 9-12 in a way that would suggest that the results are biased by the natural filtration. Table 8 here Serial correlation in the dependent variable, in this case mortality, is sometimes handled using a lagged variable (yt-x), as is used by Cutler & Miller (2005). However, it is argued by Angrist & Pischke (2008, p.245) that this creates (in a fixed-effects panel setup) a serial correlation between the residual at time t and the

8 These models treat the included cities as comparable and do not include city fixed-effects (because of the reduced sample size in this restricted sample), but year fixed effects and all other controls. 9 A model that includes the “Only water” category was deemed inappropriate using this data, since the time spent in this state is most often very short in the Swedish cities. This means that the complementarity suggested here only is one- way, since a true complementarity would require also a model using “Only water”. 10 These results can be seen in Table 7, Appendix.

19

dependent variable at time t-1 (rendering an inconsistent estimator), which is why it was not included in the main models of this paper. A sensitivity model including the (yt-1) variable was nevertheless executed, and the results did not change in any noticeable way.

5 Conclusions

Over the period examined in this paper, 1875 to 1930, the urban mortality penalty in Sweden disappeared, partly attributable to the virtual eradication of mortality due to waterborne diseases, which was primarily an urban phenomenon. This represents an impressive feat, as it allowed the overall crude mortality rate to reach a level close to today’s levels. The period was characterized by a range of public health interventions, of which the decisions made across Swedish cities to provide their inhabitants with piped water and sewage services emerge as one of the more comprehensive as well as costly. Using annual data aggregated at the city level, the aim of this paper has been to analyze the role played by aforementioned investments in the urban mortality decline. The dataset, which contains a full national range of cities from small to larger, is a unique source that allows for an analysis of water and sewage systems on a wider scale, within the same national context. Examining various measurements of mortality, the paper has demonstrated a important influence from the provision of piped water in general, and - in particular – when in combination with advanced processing

• methods. More specifically, waterborne disease mortality declined by about ten percent as a result of piped

water and/or sewage – regardless of processing methods – being provided. The magnitude of the results is, however, moderate in comparison with previous research on larger cities, which is in line with expectations both according to theory and previous research on the Swedish urban mortality decline. The corresponding mortality decrease from the use of advanced processing methods amounts to almost 30 percent, but this effect needs to be interpreted with their attached low average mortality in mind. A specific analysis of the cities that had a slow incremental implementation of first a sewage system, then both water and sewage systems, suggests that there are complementary gains from having both systems implemented. A similar analysis on water processing methods showed a similar, but statistically insignificant, gradient. The analysis is complemented by the examination of alternate indicators of mortality, believed to be less strongly connected to the expansion of piped water and sewage services. Indeed, we find less pronounced effects on crude as well as infant mortality, although still statistically significant. Again, in comparison, similar studies on the US (Alsan & Goldin, 2015; Cutler & Miller, 2005) find effects of higher magnitudes between these interventions and this outcome. Turning to mortality due to airborne disease, the results are as expected and – in our view - corroborate the findings obtained for waterborne mortality. As these types of disease typically are transmitted through different mechanisms, had we found identical (or similar) results for both types of mortality outcomes, this

20

would have indicated the existence of an omitted variable, strongly correlated with water and sewage investments and exercising a beneficial health effect overall.

6 References

Aaby, P. (1992). Lessons for the Past: Third World Evidence and the Reinterpretation of Developed World Mortality Declines, Health Transition Review, pp.155–183. Alirol, E., Getaz, L., Stoll, B., Chappuis, F. & Loutan, L. (2011). Urbanisation and Infectious Diseases in a Globalised World, The Lancet infectious diseases, vol. 11, no. 2, pp.131–141. Allison, P. D. (2009). Fixed Effects Regression Models, Vol. 160, SAGE publications. Allison, P. D. & Waterman, R. P. (2002). 7. Fixed-Effects Negative Binomial Regression Models, Sociological methodology, vol. 32, no. 1, pp.247–265. Alsan, M. & Goldin, C. (2015). Watersheds in Infant Mortality: The Role of Effective Water and Sewerage Infrastructure, 1880 to 1915, National Bureau of Economic Research, Available Online: http://www.nber.org/papers/w21263 [Accessed 14 November 2016]. Angrist, J. D. & Pischke, J.-S. (2008). Mostly Harmless Econometrics: An Empiricist’s Companion, [e-book] Princeton university press, Available Online: https://www.google.com/books? hl=sv&lr=&id=ztXL21Xd8v8C&oi=fnd&pg=PR8&dq=mostly+harmless+econometrics&ots=Uia8Z A5LzW&sig=jHJtNre0uyqcJ4LgWIurxtr9fBc [Accessed 21 February 2017]. Beach, B., Ferrie, J., Saavedra, M. & Troesken, W. (2016). Typhoid Fever, Water Quality, and Human Capital Formation, The Journal of Economic History, vol. 76, no. 01, pp.41–75. Beemer, J. K., Anderton, D. L. & Leonard, S. H. (2005). Sewers in the City: A Case Study of Individual- Level Mortality and Public Health Initiatives in Northampton, Massachusetts, at the Turn of the Century, Journal of the history of medicine and allied sciences, vol. 60, no. 1, pp.42–72. Berglund, B. & Palm, L. A. (2005). Borås Stads Historia: Omstukturering Och Anpassning 1920-2000: 1920- 2000, Historiska Media. Bjur, H. (1988). Vattenbyggnadskonst I Göteborg under 200 \a Ar, Göteborgs VA-verk. Brockmann, D., David, V. & Gallardo, A. M. (2009). Human Mobility and Spatial Disease Dynamics, Reviews of nonlinear dynamics and complexity, vol. 2, pp.1–24. Burström, B., Macassa, G., Öberg, L., Bernhardt, E. & Smedman, L. (2005). Equitable Child Health Interventions: The Impact of Improved Water and Sanitation on Inequalities in Child Mortality in Stockholm, 1878 to 1925, American Journal of Public Health, vol. 95, no. 2, pp.208–216. Cain, L. & Rotella, E. (2001). Death and Spending: Urban Mortality and Municipal Expenditure on Sanitation, in Annales de Démographie Historique, 2001, Belin, pp.139–154, Available Online: http://www.cairn.info/resume.php?ID_ARTICLE=ADH_101_0139 [Accessed 14 November 2016]. Castensson, R., Löwgren, M. & Sundin, J. (1988). Urban Water Supply and Improvement of Health Conditions, Society, health and population during the demographic transition, pp.273–298. Cutler, D. & Miller, G. (2005). The Role of Public Health Improvements in Health Advances: The Twentieth-

21

Century United States, Demography, vol. 42, no. 1, pp.1–22. Duffy, J. (1992). The Sanitarians: A History of American Public Health, University of Illinois Press. Duflo, E., Greenstone, M., Guiteras, R. & Clasen, T. (2015). Toilets Can Work: Short and Medium Run Health Impacts of Addressing Complementarities and Externalities in Water and Sanitation, National Bureau of Economic Research, Available Online: http://www.nber.org/papers/w21521 [Accessed 2 May 2017]. Easterlin, R. A. (1999). How Beneficent Is the Market? A Look at the Modern History of Mortality, European Review of Economic History, vol. 3, no. 3, pp.257–294. Edvinsson, S. (1992). Den Osunda Staden: Sociala Skillnader I Dödlighet I 1800-Talets Sundsvall, [e- journal], Available Online: http://www.diva-portal.org/smash/record.jsf?pid=diva2:435711 [Accessed 14 November 2016]. Ferrie, J. P. & Troesken, W. (2008). Water and Chicago’s Mortality Transition, 1850–1925, Explorations in Economic History, vol. 45, no. 1, pp.1–16. Fewtrell, L., Kaufmann, R. B., Kay, D., Enanoria, W., Haller, L. & Colford, J. M. (2005). Water, Sanitation, and Hygiene Interventions to Reduce Diarrhoea in Less Developed Countries: A Systematic Review and Meta-Analysis, The Lancet infectious diseases, vol. 5, no. 1, pp.42–52. Fogel, R. W. (1986). Nutrition and the Decline in Mortality since 1700: Some Preliminary Findings, in Long- Term Factors in American Economic Growth, University of Chicago Press, pp.439–556. Fridlizius, G. (1985). The Mortality Decline in the First Phase of the Demographic Transition: Swedish Experiences, [e-journal], Available Online: http://www.popline.org/node/415951 [Accessed 12 September 2016]. Gabadinho, A., Ritschard, G., Mueller, N. S. & Studer, M. (2011). Analyzing and Visualizing State Sequences in R with TraMineR, Journal of Statistical Software, vol. 40, no. 4, pp.1–37. Galor, O. (2012). The Demographic Transition: Causes and Consequences, Cliometrica, vol. 6, no. 1, pp.1– 28. Haussman, J., Hall, B. & Griliches, Z. (1984). Economic Models for Count Data with an Application to the Patents-R&D Relathionship, Econometrica, vol. 52, pp.909–938. Kesztenbaum, L. & Rosenthal, J.-L. (2014). Income versus Sanitation; Mortality Decline in Paris, 1880- 1914, [e-journal], Available Online: https://halshs.archives-ouvertes.fr/halshs-01018594/ [Accessed 14 November 2016]. Lee, R. (2003). The Demographic Transition: Three Centuries of Fundamental Change, The Journal of Economic Perspectives, vol. 17, no. 4, pp.167–190. Lindman, C. (1911). Sundhets-Och Befolkningsförhållanden I Sveriges Städer 1851-1909. 2, Tabeller Och Diagram, Nord. bokh. McCarthy, M. P. (1987). Typhoid and the Politics of Public Health in Nineteenth-Century Philadelphia, Vol. 179, American Philosophical Society. McKeown, T., Brown, R. G. & Record, R. G. (1972). An Interpretation of the Modern Rise of Population in Europe, Population studies, vol. 26, no. 3, pp.345–382. McKeown, T. & Record, R. G. (1962). Reasons for the Decline of Mortality in England and Wales during the

22

Nineteenth Century, Population studies, vol. 16, no. 2, pp.94–122. Molitoris, J. (2015). Life and Death in the City: Demography and Living Standards during Stockholm’s Industrialization, Vol. 73, [e-book] Lund University, Available Online: http://lup.lub.lu.se/record/8045791 [Accessed 14 November 2016]. Nilsson, H. (1994). Mot Bättre Hälsa: Dödlighet Och Hälsoarbete I Linköping 1860-1894, [e-journal], Available Online: http://www.diva-portal.org/smash/record.jsf?pid=diva2:255928 [Accessed 14 November 2016]. Nilsson, L. (2013). Städerna 1860–1920, in 150 År Av Självstyrelse - Kommuner Och Landsting I Förändring, SKL, Sveriges Kommuner och Landsting. Omran, A. R. (1971). The Epidemiologic Transition: A Theory of the Epidemiology of Population Change, The Milbank memorial fund quarterly, vol. 49, no. 4, pp.509–538. Petersson, F. (2005). Vattnets Vägar - Från Vik till Innergård På Södermalm 1880-1920, Center for Health Equity Studies (Chess). PHAS, P. H. A. of S. (2005). Svenska Folkets Hälsa I Historiskt Perspektiv. Razzell, P. E. (1974). ‘An Interpretation of the Modern Rise of Population in Europe’—A Critique, Population Studies, vol. 28, no. 1, pp.5–17. Reher, D. S. (2001). In Search of the ‘urban Penalty’: Exploring Urban and Rural Mortality Patterns in Spain during the Demographic Transition, International Journal of Population Geography, vol. 7, no. 2, pp.105–127.

• SCB. (1969). Del 1. Befolkning. Andra Upplagan. 1720–1967, Stockholm: SCB.

Schön, L. (2000). En Modern Svensk Ekonomisk Historia: Tillväxt Och Omvandling under Två Sekel [A Modern Economic History of Sweden: Growth and Transformation during Two centuries](3rd Swedish Ed.), Stockholm, Sweden: SNS förlag. Singh, P. P. & Sharma, V. (2014). Water and Health, [e-book] Springer, Available Online: http://link.springer.com/content/pdf/10.1007/978-81-322-1029-0.pdf [Accessed 31 March 2017]. Soares, R. R. (2007). On the Determinants of Mortality Reductions in the Developing World, Population and Development Review, vol. 33, no. 2, pp.247–287. Söndags-Nisse. (1866). En Kloak – I Stockholm, Söndags-Nisse – Illustreradt Veckoblad för Skämt, Humor

• ch Satir, nr 34.

Sundin, J., Hogstedt, C., Lindberg, J. & Moberg, H. (2005). Svenska Folkets Hälsa I Historiskt Perspektiv,

• Vol. 2005, [e-book] Statens folkhälsoinstitut Stockholm, Available Online:

http://www.folkhalsomyndigheten.se/pagefiles/21487/r2005-08-svensk-folkhalsohistoria.pdf [Accessed 1 April 2017]. Sundin, J. & Willner, S. (2007). Social Change and Health in Sweden: 250 Years of Politics and Practice, [e- book] Swedish National Institute of Public Health, Available Online: http://www.diva- portal.org/smash/record.jsf?pid=diva2:17729. Tien, J. H. & Earn, D. J. (2010). Multiple Transmission Pathways and Disease Dynamics in a Waterborne Pathogen Model, Bulletin of mathematical biology, vol. 72, no. 6, pp.1506–1533. Troesken, W. (2001). Race, Disease, and the Provision of Water in American Cities, 1889–1921, The Journal

23

• f Economic History, vol. 61, no. 03, pp.750–776.

Van Poppel, F. & Van der Heijden, C. (1997). The Effects of Water Supply on Infant and Childhood Mortality: A Review of Historical Evidence, Health Transition Review, pp.113–148.

• WHO. (2008). Progress on Drinking-Water and Sanitation., [e-book] World Health Organization, Available

Online: https://www.cabdirect.org/cabdirect/abstract/20093082186.

• WHO. (2014). Water Sanitation Health: Burden of Disease and Cost-Effectiveness Estimates, Available

Online: http://www.who.int/water_sanitation_health/diseases/burden/en/. Winkelmann, R. (2013). Econometric Analysis of Count Data, Springer Science & Business Media. Woods, R. I., Watterson, P. A. & Woodward, J. H. (1988). The Causes of Rapid Infant Mortality Decline in England and Wales, 1861–1921 Part I, Population studies, vol. 42, no. 3, pp.343–366.

24

Tables and figures

25 Table 1: Median CDR/IMR/waterborne disease mortality prior to piped water intervention

City obtains piped water during period 1880-1889 1890-1899 1900-1909 1910-1920 1920-1930 Yes No Yes No Yes No Yes No Yes No Median CDR during nearest preceding decade 19,91 20,47 15,80 18,03 16,21 16,44 13,27 14,41 13,98 14,50 Median IMR during nearest preceding decade 147,77 147,64 109,60 113,86 95,43 94,07 59,30 53,08 48,09 54,02 Median waterborne disease DR during nearest preceding decade 2,70 2,19 1,49 1,66 1,12 1,16 0,67 0,55 0,27 0,25 Number of cities 12 56 21 36 18 18 4 14 9 6

26 Table 2: Share of households covered by piped water or sewage services, by years since implementation Water Sewage Years since implementation: 0 58,6 53,2 1 62 65,4 2 69,9 66,7 3 78 72,7 4 77,6 72,3 5 78,6 72,9 6-10 82,3 77,9 11-15 84,9 82,9 16-20 86,9 83,9

Notes: Coefficients presented as Incident Rate Ratios. All models include available control variables, city and year fixed effects, as well as city and (linear) year interaction effect. Standard errors in parentheses.

* p < 0.10, ** p < 0.05, *** p < 0.01

27 Table 3: Negative binomial regression output, years 1870-1930, Models 1-4

(1) (2) (3) (4) Waterborne disease deaths All deaths Infant deaths Airborne disease deaths No water/sewage 1 1 1 1 (.) (.) (.) (.) Water and/or sewage 0.989 (0.0356) (0.0124) (0.0252) (0.0222) Observations 4474 4474 4474 4474 0.913** 0.952*** 0.939**

Notes: Coefficients presented as Incident Rate Ratios. All models include available control variables, city and year fixed effects, as well as city and (linear) year interaction effect. Standard errors in parentheses.

* p < 0.10, ** p < 0.05, *** p < 0.01

Table 4: Negative binomial regression output, years 1901-1930, Models 5-12

(5) (6) (7) (8) (9) (10) (11) (12) All deaths Infant deaths All deaths Infant deaths No water/sewage 1 1 1 1 (.) (.) (.) (.) Water and/or sewage 1.024 1.006 1.038 (0.107) (0.0937) (0.0258) (0.0667) No water/sewage 1 1 1 1 (.) (.) (.) (.) Water, no processing 0.999 1.016 (0.0855) (0.0197) (0.0662) (0.0525) Water, simple processing 1.017 1.055 1.088 (0.0893) (0.0233) (0.0782) (0.0608) Water, advanced processing 1.016 1.098 (0.120) (0.0337) (0.112) (0.0898) Observations 2398 2398 2398 2398 2398 2398 2398 2398 Waterborne disease deaths Airborne disease deaths Waterborne disease deaths Airborne disease deaths 0.773* 0.838* 1.093* 0.772** 0.686** 1.150*

Notes: Coefficients presented as Incident Rate Ratios. All models include available control variables and year fixed

• effects. Standard errors in parentheses.

* p < 0.10, ** p < 0.05, *** p < 0.01

29 Table 5: Negative binomial regression output, cities in cluster “Slow Adopters”, Models 13-16

(13) (14) (15) (16) All deaths Infant deaths No water/sewage 1 1 1 1 (.) (.) (.) (.) Sewage, no water 0.911 (0.0599) (0.0232) (0.0553) (0.0406) Water and sewage (0.0651) (0.0273) (0.0651) (0.0515) Observations 1101 1101 1101 1101 Waterborne disease deaths Airborne disease deaths 0.805*** 0.903*** 0.868*** 0.681*** 0.827*** 0.871* 0.844***

30

50 60 70 80 90 100 110 120 130 140 1880 1885 1890 1895 1900 1905 1910 1915 1920 1925 1930

IMR

Rural Urban

Figure 1: Infant mortality in urban and rural areas

31 Figure 2: Median, 10th and 90th percentile infant mortality rate in Swedish cities

32 Figure 3: Average waterborne and airborne disease mortality in Swedish cities

33 Figure 4: Cumulative share of cities with piped water/sewage services

Note: Each cluster contains a set number of cities clustered together using distances between sequences in the water/sewage variable. The y axis shows the share of cities within the cluster per implementation category and year. Clusters have been created based on all years where implementation data is available.

34

Middle fast adopters Early adopters Late adopters Slow adopters

Figure 5: Implementation timing clusters of all cities in sample

Note: Observations above (below) 45 degree line represents cities experiencing higher (lower) migration after piped water services were implemented.

35 Figure 6: Mean migration rate, before and after implementation of piped water.

36 Figure 7: Water processing methods used, cumulative shares

Note: Each cluster contains a set number of cities clustered together using distances between sequences in the water processing variable. The y axis shows the share of cities within the cluster per implementation category and year. Clusters have been created based on all years where implementation data is available.

37

No processors Incremental implementers Simple processors Late adopters

Figure 8: Implementation timing clusters of all cities in sample – water processing

Appendix

38

Variable Mean Min Max Waterborne disease deaths 7,15 0,00 530,00 Total deaths 108,85 0,00 1703,00 Infant deaths 17,29 0,00 325,00 Airborne disease deaths 19,14 0,00 279,00 Factoryworkers/1000 63,58 0,00 311,22 Population 7708,59 450,00 120304,00 Water services provided 0,44 0,00 1,00 Sewage services provided 0,52 0,00 1,00 Water services provided, no processing 0,27 0,00 1,00 Water services provided, simple processing 0,16 0,00 1,00 Water services provided, advanced processing 0,01 0,00 1,00 Milk inspections 0,14 0,00 1,00 Pork inspections 0,27 0,00 1,00 Meat inspections 0,08 0,00 1,00 Has hospital 0,44 0,00 1,00 Observations (years) 4474 Cities 84

Table 6: Sample mean, min and max values

Notes: Coefficients presented as Incident Rate Ratios. All models include available control variables and year fixed

• effects. Standard errors in parentheses.

* p < 0.10, ** p < 0.05, *** p < 0.01

This models suffers from a low sample size because of the sub-sampling, and also the limited time period (>1901) of the variable, which shows in the fact that all coefficients except for airborne mortality11 are insignificant. The coefficients for waterborne deaths and infant deaths show a gradient from no processing → simple processing → advanced processing, but the lack of significance makes it impossible to say that they are statistically different from each other. 11 These coefficients seem to imply an increase in airborne mortality generally in later periods within this cluster of cities, but since they have the same magnitude, the results do not seem interesting within the scope of this analysis.

39 Table 7: Negative binomial regression output, cities in cluster “Incremental implementers”, Models 17-20

(17) (18) (19) (20) All deaths Infant deaths No water/sewage 1 1 1 1 (.) (.) (.) (.) Water, no processing 1.153 0.999 1.090 (0.191) (0.0365) (0.141) (0.0891) Water, simple processing 1.108 1.059 0.924 (0.186) (0.0376) (0.118) (0.0889) Water, advanced processing 0.847 1.051 0.888 (0.221) (0.0547) (0.154) (0.133) Observations 360 360 360 360 Waterborne disease deaths Airborne disease deaths 1.271*** 1.240*** 1.272**

Notes: Coefficients presented as Incident Rate Ratios. All models include available control variables, city and year fixed effects, as well as city and (linear) year interaction effect. Standard errors in parentheses.

* p < 0.10, ** p < 0.05, *** p < 0.01

40 Table 8: Negative binomial regression output, Models 21-24

(21) (22) (23) (24) All deaths Infant deaths No water 1 1 1 1 (.) (.) (.) (.) Water, no proc. 1.020 0.999 (0.0860) (0.0670) (0.0198) (0.0527) Water, simple proc. (incl. natural filter) 1.043 1.025 1.083 (0.0931) (0.0793) (0.0244) (0.0620) Water, advanced proc. 0.988 0.983 1.095 (0.123) (0.108) (0.0362) (0.0886) Observations 2398 2398 2398 2398 Waterborne disease deaths Airborne disease deaths 0.840* 1.097* 0.781** 0.680**

41 Figure 9: Satirical image of water and sewage system in Stockholm, 1866

42 Figure 10: Median, 10th and 90th percentile crude mortality rate in Swedish cities