Extracted main text — title through conclusion, appendix excluded. This is what our citation measures are computed over, published so the extraction can be checked by eye.
58,595 characters · 10 sections · 0 citation commands
Chain effects of clean water: The Mills--Reincke phenomenon in early twentieth-century Japan
Waterborne diseases caused considerable losses of human capital throughout the twentieth century (Preston and van de Walle 1978; Evans 1987; Troesken 2004). Previous studies have therefore regarded the implementation of water purification systems as being responsible for the huge improvements in public health. The pioneering and influential study by Cutler and Miller (2005) showed that the development and application of water purification technologies was responsible for roughly 40% of the decline in the mortality rate from 1900 to 1940. Subsequent studies also found the similar improving effects of clean water both in European and in Asian countries (Macassa et al. 2006; Jaadla and Puur 2016; Ogasawara and Matsushita 2018; Peltola and Saaritsa 2019).\footnote{Alsan and Goldin (2019) and Kesztenbaum and Rosenthal (2017) showed the importance of sewers in reducing the mortality rate in Massachusetts and Paris, respectively.} However, the recent study by Anderson et al. (2018) found more moderate impacts of water purification technology on crude and infant mortality rates than those shown by Cutler and Miller (2005). Brown and Guinnane (2018) also provided contradictory views on these improving effects of safe water in Bavaria between 1825 and 1910.
The present study seeks to contribute to this debate by adding new evidence on the improving effects of modern water-supply systems on the historical decline in the mortality rate. While a large number of previous studies have shown the instantaneous effects of clean water on mortality rates, we try to show the chain (i.e., persistent) effects of safe water on mortality rates by testing the Mills--Reincke phenomenon. The Mills--Reincke phenomenon is an epidemiological proposition arguing that the purification of polluted water could not only reduce deaths due to typhoid in the initial stage, but also those due to other infectious diseases in the later stage (Sedgwick and MacNutt 1910; Evans 1987). Epidemiological case studies have provided evidence of this phenomenon in large cities in France and the United States (Preston and van de Walle 1978; Crimmins and Condran 1983). Economic studies of this phenomenon are more scarce; however, the work by Ferrie and Troesken (2008) showed that for every death from typhoid fever prevented by water purification, three or more deaths from other causes were also prevented in Chicago from 1850 to 1925. While these studies have predominantly focused on the cases of a few large cities, this study is the first to use a more comprehensive dataset covering multiple cities in prewar Japan.
We find that eliminating typhoid fever infection decreased the risk of non-waterborne diseases. Our estimates show that for each additional death due to typhoid, there were approximately two deaths due to tuberculosis and pneumonia; this magnitude is indeed greater than that observed previously in Chicago (Ferrie and Troesken 2008). The composition of the cause-specific deaths suggests that the national malady (kokuminby\=o), especially tuberculosis, was more likely to be improved by the eradication of typhoid fever. This finding not only supports the evidence provided by Ferrie and Troesken but also adds more comprehensive evidence on the chain effects of clean water in industrializing Japan.
This study contributes to the broader literature in the following two ways. First, we complement the abovementioned discussion on the role of water purification technology in mitigating historical mortality declines (Anderson et al. 2018; Cutler and Miller 2019). The results of this study indeed support the evidence provided by previous studies of the important role of clean-water technology (e.g., Alsan and Goldin 2019). Second, we explore the validity of the Mills--Reincke phenomenon using comprehensive city-level mortality data from the early twentieth century. Expanding Ferrie and Troesken's (2008) strategy to capture the mechanisms responsible for the appearance of the phenomenon, we find that the chain effects of clean water were still considered to be substantial.\footnote{Our cost/benefit analysis in (ref) also indicates that the contributions of clean-water technologies to historical declines in mortality have been sufficiently large to exceed the costs of installation, as demonstrated by Cutler and Miller (2005) and Ferrie and Troesken (2008).} This study highlights the importance of investigating not only the instantaneous effects but also the subsequent chain effects of clean water when we discuss the impacts of the implementation of water purification systems on historical mortality declines.
The rest of the paper proceeds as follows: Section (ref) provides an overview of mortality rate trends in pre-war Japan, and also illustrates the features of the Mills--Reincke phenomenon. Section (ref) describes the data that we used. Section (ref) explains our empirical strategy and presents the main results. Section (ref) concludes.
In Japan, modern waterworks were first constructed in Yokohama city in 1887. Subsequently, the port cities of Hakodate and Nagasaki introduced water-supply systems in 1888. Modern waterworks, however, were installed only in the most highly populated cities and open ports at the beginning because the scope of the governmental subsidy was limited (Japan Water Works Association 1967). The Waterworks Ordinance enacted in 1890 required the construction cost of waterworks to be covered by public funds and thus a subsidy from the national government was necessary for municipalities to introduce modern water-supply systems. Therefore, installations and expansions of modern waterworks only quickly spread after 1918 and 1920 when the government twice expanded the scope of the subsidy. Indeed, modern systems were installed in only seven locations in 1900, rising to 55 in 1921 and 345 in 1940 (Japan Water Works Association 1967). Likewise, Fig. (ref) shows that the number of water taps per 100 households in cities increased from 10.63% in 1922 to 35.45% in 1938.
The most important difference between old and modern systems is whether they had water purification technologies. Old waterworks simply ran water from rivers and springs to urban areas without any clarification facilities. Moreover, their wooden pipes had decayed over a long time (Japan Water Works Association 1967). Thus, the water supplied through the old waterworks was incredibly polluted and carried pathogens. For instance, the University of Tokyo and Sanitary Bureau of the Home Department conducted a survey of water quality in Tokyo around 1880 and concluded that drinking water in urban areas was dirty like “thinned urine” (Bureau of Waterworks, Tokyo Metropolitan Government 1999, pp.6--7). By contrast, modern waterworks used cast iron pipes and water purified by filtration (and chlorination in some cities) in accordance with water quality standards that are still applicable today (Japan Water Works Association 1967). These improvements in water-supply systems made tap water sufficiently clean to prevent waterborne infections (Ogasawara and Matsushita 2018).
Figure (ref) illustrates the death rates from typhoid fever, a waterborne disease, in 1921--1938 as well as the coverage of tap water. The typhoid death rate indeed decreased rapidly as the use of water-supply systems spread. However, it should be noted that a downward trend in mortality rates was also observed for deaths due to non-waterborne disease as shown in Fig. (ref). This means that improvements in both waterborne and non-waterborne risks of death contributed to the declines in mortality, which invites the question, “Were water-supply systems responsible for the reductions in non-waterborne mortality?” If one considers the Mills--Reincke phenomenon, then the mortality transitions that were observed in Japan provide interesting new evidence.
Figure (ref) shows rates of Milles--Reincke effect-associated (MRE) deaths, which are defined as deaths from tuberculosis, pneumonia, bronchitis, meningitis, heart disease. These causes of death are likely to be subject to the Mills--Reincke effect as discussed in detail in Subsection (ref). Rates of death are reported per 1,000 people and are for major cities. The MRE death rate dropped from 6.46 in 1921 to 4.49 in 1938, which is a 30.50% reduction. On the other hand, Fig. (ref) also shows rates of non-MRE deaths, which are defined as deaths from causes other than the MRE and waterborne deaths. The rates of non-MRE deaths in 1921 and 1938 were 12.01 and 8.98 per 1,000 people, respectively, which amounts to a 25.23% reduction. Clearly, the MRE death rate declined more rapidly than the non-MRE death rate. This divergence strongly suggests that the Mills--Reincke phenomenon was observed in Japan, as previously reported for European countries. Therefore, to understand the entirety of the effects that improving water quality had on mortality, we investigated the chain of effects in terms of the Mills--Reincke phenomenon.
Typhoid fever, which is an oral infectious disease, is caused by the bacterium Salmonella enterica serotype Typhi (S. Typhi) (Huang and DuPont 2005). The onset of symptoms is marked by fever and malaise. Patients often have a fever, a dull frontal headache, nausea, a dry cough, a coated tongue, splenomegaly, relative bradycardia, constipation, diarrhea, and any other symptoms, but have few physical signs (Parry et al. 2002, p.1774). Typhoid fever often does not threaten human life directly and the case-fatality rate is reported to be approximately 10--20% (World Health Organization 2011, p.1). This relatively low case-fatality rate likely understates its importance. The most important feature of typhoid fever is that it usually causes non-typhoid diseases or worsens chronic diseases. Therefore, it has often been observed that, as water quality improved in Western nations during the late nineteenth and early twentieth centuries, there were reductions in both typhoid and non-typhoid death rates. The negative correlation between the purification of polluted water supplies and non-typhoid death rates is called the Mills--Reincke phenomenon because it was discovered independently by Hiram F. Mills and J. J. Reincke in 1893--1894. \footnote{See Appendix (ref) for finer details of the typhoid fever and a brief overview of the epidemiological studies of this phenomenon that were performed in the early twentieth century.}
The main factor that could cause the Mills--Reincke phenomenon is considered to be complications from typhoid fever. Numerous extra-intestinal complications can occur with S. Typhi infection, including the involvement of the central nervous, cardiovascular, pulmonary, bone and joints, hepatobiliary, and genitourinary systems (Huang and DuPont 2005). To confirm that deaths from complications accounted for some proportion of deaths that were not due to typhoid fever itself, we compiled a unique survey report, in which the frequencies of deaths from complications of typhoid fever were recorded. In this survey, complications were investigated for $1,214$ patients diagnosed with typhoid fever and hospitalized in Komagome Hospital in Tokyo during 1932--1933 (Tokyo City Office 1935). According to this survey, $96.21$% (i.e., $1,168$) of patients had complications during the survey period. \footnote{Table (ref) in Appendix (ref) shows the numbers of complication patients by disease category.} Table (ref) summarizes the frequencies of deaths due to typhoid fever or complications among typhoid patients. Of the $1,214$ patients with typhoid fever, $92$ ($7.58$%) had deaths attributed to the complications, implying that if patients with typhoid fever were affected with other diseases, approximately $7.6$% of them might have died of these complications. Moreover, since the total number of deaths among the patients was reportedly $172$, the number of deaths due to complications per 100 total deaths is estimated to be $53.49$% ($92$/$172$). This implies that more than half of the deaths in patients with typhoid fever were ultimately due to non-typhoid diseases.
Among the complications, the shares of deaths from enterorrhagia and perforation (which are considered to be direct consequences of typhoid fever) were relatively high ($20.35$% and $5.81$%, respectively). The more important fact is that pneumonia, beriberi, and meningitis also accounted for relatively large proportions of the deaths ($13.95$%, $6.4$%, and $3.49$% of the $172$ typhoid-related deaths, respectively). In addition, mumps and bronchitis were also listed as complications that had fatal consequences. However, they accounted for only $1.16$% and $1.74$% of the total deaths, respectively. The above findings suggest that improvements in the typhoid death rate are more likely to reduce the death rates from mumps, bronchitis, pneumonia, beriberi, and meningitis, as well as digestive diseases such as enterorrhagia and perforation.
The abovementioned findings imply that various complications were derived from infections with typhoid fever. These complications caused additional deaths from non-typhoid causes, such as pulmonary and cardiovascular system diseases, as well as digestive system diseases.
We hypothesized that the improvement of water-supply systems reduced both waterborne and non-waterborne deaths by preventing typhoid fever. To confirm this hypothesis by empirical analyses, the present study used different panel datasets for our two main dependent variables: the non-typhoid death rate and the MRE death rate. The characteristics of the study samples are as follows: To estimate the effects on the non-typhoid death rate, we compiled panel data from 108 cities pertaining to the period between 1922 and 1940, including almost all of the cities interspersed across the whole of Japan. In fact, approximately 92.38% of the city-dwelling Japanese citizens were included in our study cohort during the study period. Regarding the MRE death rate, the number of deaths from each cause is not described for every city. Nevertheless, we could compile panel data on cause-specific deaths in 26 cities with populations greater than 100,000 between 1922 and 1936. \footnote{Cities with populations greater than 50,000 are included for 1922.}
In our hypothesis, typhoid fever was the key factor contributing to the Mills--Reincke phenomenon. We use the death rate from typhoid fever as an independent variable that indexes typhoid epidemic and severity level. The typhoid death rate (Typhoid) is defined as the number of deaths from typhoid fever per 1,000 people. In addition, we use the typhoid incidence rate (TIR) as an alternative measure of epidemic level. This is defined as the number of cases of typhoid fever per 1,000 people. However, the measurement error of typhoid cases was larger than that of typhoid deaths. \footnote{For example, some people who were affected by typhoid fever masked their disease (Doi 1925).} Thus, we take the death rate as primary measure in our estimates. The data on the number of typhoid deaths and cases are obtained from statistical reports: Nihonteikoku shiin tokei, Josuido tokei, Eiseikyoku nenpo, and Eisei nenpo.
We use the following measures of mortality rates as the dependent variables. The non-typhoid death rate (Non-typhoid) is defined as the number of deaths from causes other than typhoid fever per 1,000 people. Since this measure was also used in Ferrie and Troesken (2008), we are able to compare the results of the present study with their previous research. However, the non-typhoid death rate includes deaths that could not possibly be considered to have been affected by typhoid, such as drowning and freezing deaths. This might increase the gap between the underlying truth and our interpretation of estimated results.
Therefore, we also use the MRE death rate (MRE death). We define MRE deaths as deaths from tuberculosis, pneumonia, bronchitis, meningitis, and heart disease based on medical and historical evidence. These diseases involve the central nervous, cardiovascular, and pulmonary systems, which can occur with S. Typhi infection (Huang and DuPont 2005). \footnote{We categorize tuberculosis, which was known as one of the two major national maladies in Japan, as MRE deaths; it was not listed as showing complications by Huang and DuPont (2005), as tuberculosis is rare in modern developed countries because of the development of effective drugs (Zumla et al. 2013).} In addition, they were frequently observed as complications of typhoid fever in pre-war Japan. If our hypothesis is correct, the relationship between the MRE death rate and the typhoid death rate should be significantly positive.
Furthermore, to investigate the Mills--Reincke phenomenon in more detail, we estimate effects on cause-specific death rates. As a placebo test, our main estimates also use the death rate for causes that were not categorized as MRE death. We expect that preventing typhoid fever would contribute to decreases in deaths that met the definition of MRE death, but not affect other deaths, namely those due to scarlet fever, smallpox, diphtheria, whooping cough, measles, influenza, beriberi, and nephritis. These tracheal and renal system diseases and nutritional deficiency disease are not considered to be complications of typhoid fever in medical studies (see Huang and DuPont 2005). Our main sources for the data on the cause-specific deaths are Nihonteikoku shiin tokei and Shiin tokei. The materials are based on official statistics after the first national census conducted in 1920 and the statistics are known to be sufficiently accurate (see Ito 1987).
We included control variables representing demographic, socioeconomic, and meteorological characteristics. The demographic variables included the number of citizens, shares of various age groups, and the sex ratios of the age groups. For socioeconomic variables, we included the size of the financial budget per capita, the number of doctors per 100 people, and the proportion of industrial workers in the population. For meteorological variables, we included the annual mean temperature, annual mean humidity, and annual mean actual sunshine duration. These variables could control for spatial heterogeneity in the risk of infectious diseases (Ni et al. 2014; World Health Organization 2009). In addition, to control for the the consequences of the Great Kant\=o Earthquake, we included an indicator variable that takes the value one for both Tokyo and Yokohama in 1923. (ref) provides the finer details of our data, definitions of the variables, and data sources.
To capture the size and impact of the Mills--Reincke phenomenon in early twentieth-century Japan, we employed the following city fixed effects approach as our identification strategy. Our baseline model is given by
where $i$ indexes cities from 1 to 108 and $t$ indexes years from 1922 to 1940. The variable $\textit{Typhoid}_{it}$ is the typhoid death rate, $\textit{Non--typhoid}_{it}$ is the non-typhoid death rate, $\mathbf{x}'_{it}$ is a vector of city characteristics, and $e_{it}$ is a random error term. $v_{i}$ and $u_{t}$ represent city and year fixed effects, respectively. $t\gamma_{i}$ indicates a city-specific linear time trend. The coefficient $\delta$ is our parameter of interest, and its estimate $\hat{\delta}$ measures the impact of typhoid fever on the non-typhoid death rate.
Table (ref) reports our results for the non-typhoid death rate estimated by Eq. ((ref)). In all specifications, we controlled for the consequences of the Great Kant\=o Earthquake. The result in columns (3) and (7) report the estimates from our baseline specification, in which we controlled for all characteristics. Evidently, the estimates are stable across the different specifications and different measures. In columns (1)--(4), all estimated coefficients of Typhoid are significantly positive. This result suggests that typhoid fever caused complications which led to death. Moreover, the coefficient of $\textit{Non-typhoid}_{t-1}$, which is insignificant in column (4), implies that typhoid deaths were not affected by other death previous year.
To understand the contributions of the present study, we should compare our results with the previous literature. Columns (1)--(4) show that one additional typhoid death per 1,000 people lead to increases in deaths from other causes by 0.994, 1.013, 1.033, and 1.092 per 1,000 people, respectively. On the other hand, in the case of Chicago from 1850 to 1925, the feasible estimate of the Mills--Reincke effect was between 4 and 7 (Ferrie and Troesken 2008, p.13). Our estimates are obviously much smaller than those presented in the previous study. A possible explanation of this difference is competing risks. If someone died from typhoid fever, the cause of death was understandably typhoid fever rather than any other disease. Thus, an increase in typhoid deaths among typhoid cases would potentially decrease the number of deaths due to complications. In pre-war Japan, the case-fatality rate of typhoid fever was roughly 20%, whereas the rate was 5--10% in the United States in 1850--1925 (Ferrie and Troesken 2008, p.7). \footnote{Typhoid case-fatality rate is generally approximately 10--20% (World Health Organization 2011, p.1),} This higher case-fatality rate might lower the Mills--Reincke effect in Japan in comparison with the United States.
In columns (5)--(8), the estimated coefficients are 0.170, 0.172, 0.175, and 0.206, respectively. Clearly, all of the results remained significantly positive, suggesting that our main finding is not sensitive to the definition of the key variable that was used to measure typhoid epidemic level. Moreover, it is noteworthy that the result for incidence rate imply a mechanism for the Mills--Reincke phenomenon. The estimated coefficient in column (7) means that for one additional typhoid-infected person there were 0.175 additional non-typhoid deaths. Further, since the average typhoid case-fatality rate in our sample is 19.13%, the result in column (3) suggests that one additional incidence case of typhoid fever per 1,000 people increased non-typhoid death rates by 0.198‰ (i.e., 0.1913 $\times$ 1.033). Therefore, the impacts of the increase in the typhoid death rate were likely to be larger than the impacts of the increase in the incidence rate. This implies that complications were caused by serious cases of typhoid, which are reflected by the death rate, rather than by mild cases. In other words, it is possible that relatively more non-typhoid deaths were prevented by decreases in the more serious forms of typhoid.
Since the non-typhoid death rate includes some noise such as non-MRE deaths, we investigate the effects related to the Mills--Reincke phenomenon in detail using cause-specific death rates. The specification of our fixed effects model is given by
where $i$ indexes cities from 1 to 26 and $t$ indexes years from 1922 to 1936. The variable $y_{it}$ is the dependent variable, such as the MRE death rate or the cause-specific death rate. The other variables are defined as in Eq. ((ref)).
We further employed the fixed effects two-stage least squares (FE-2SLS) approach using the number of water taps per 100 households (Water) as an instrumental variable to address a possible omitted variable problem when controlling for fixed effects. \footnote{ Although the modern sewage system is also an effective public health facility (Alsan and Goldin 2019), the coverage of sewage systems among citizens continued to be low in interwar Japan; for instance, the share of households with flushable toilets in 1935 was only 0.16% (Ogasawara and Matsushita 2019). Since the traditional system of circulating human waste from urban to rural areas was well developed before the introduction of chemical fertilizers, urban households had little need to be connected to the sewage system for waste disposal for a long time. Maeda (2008 pp. 67--70) suggested that this feature of traditional disposal systems led to delays in the installation of a flushing toilet.} Our specification of FE-2SLS model is then given by
where $i$ indexes cities from 1 to 26 and $t$ indexes years from 1922 to 1936. As defined in Eq. ((ref)), the variable $\mathbf{x}'_{it}$ is a vector of city characteristics, $\nu_{i}$ ($v_{i}$) and $\mu_{i}$ ($u_{t}$) represent city and year fixed effects, and $t\rho_{i}$ ($t\gamma_{i}$) indicates a city-specific linear time trend. In Eq. ((ref)) of the first stage, $\textit{Typhoid}_{it}$ is the typhoid death rate, $\textit{Water}_{it}$ is the tap water coverage, and $\epsilon_{it}$ is a random error term. In Eq. ((ref)) of the second stage, we used the fitted value $\widehat{\text{\textit{Typhoid}}}_{it}$ instead of $\textit{Typhoid}_{it}$ to identify the effect of typhoid fever instrumented by $\textit{Water}_{it}$. The variable $\textit{y}_{it}$ is the dependent variable: the MRE death rate and each of the cause-specific death rates. \footnote{The Non-typhoid variable is likely to correlate with the popularization of tap water because it includes deaths due to waterborne diseases, such as dysentery and diarrhea. Therefore, we employed the FE-2SLS model for only the MRE and the cause-specific death rates. } $e_{it}$ is a random error term.
The instrument Water is considered to satisfy the exclusion restriction for the following reasons. First, both the timing of installation and the transition of tap water coverage were exogenous owing to the characteristics of the natural environment and unpredictable events. Installations of modern water-supply systems were influenced by geographical characteristics such as a warm climate, light rainfall, and shallow rivers; natural disasters such as droughts, floods, and wildfires; political issues such as water rights, residents' opposition, delayed land acquisitions, and conflict between city councils; and increases in the import value of cast iron pipes. In addition, the expansion of the number of water taps was influenced by changes in the natural state of water sources, outbreak of war, and consolidation of municipalities as well as the factors described above. \footnote{See Ogasawara and Matsushita (2018) for finer details of the uncertainties of modern water-supply projects in early twentieth-century Japan.} Therefore, we can exploit the exogenous variation over time using tap water coverage rather than the timing of the installation of waterworks.
Second, there was no direct relationship between tap water coverage and non-waterborne death rates. To confirm that modern water-supply systems did not directly decrease non-waterborne diseases, we estimate the effects of tap water coverage on death rates from non-waterborne diseases, namely tuberculosis, pneumonia, bronchitis, meningitis, heart disease, scarlet fever, smallpox, diphtheria, whooping cough, measles, influenza, beriberi, and nephritis. Our specification is similar to that in Eq. ((ref)) but we use Water instead of Typhoid as the key independent variable. All the demographic, socioeconomic, and meteorological variables, which are important factors in deaths because of typhoid fever as well as other diseases, are included as control variables. Table (ref) reports the estimation results. The finding that the estimated coefficients of Water are insignificant for all death rates regardless of MRE or non-MRE deaths indicates that waterworks had no direct effects.
Third, modern waterworks improved water quality but did not dramatically change the time taken to access water, which could be used for something else to improve health if it was not needed. Although the ground water was occasionally contaminated, Tokyo city had 187,169 wells, that is, 3.27 per 100 square meters in 1939 (Bureau of Waterworks, Tokyo Metropolitan Government 1999, p.193; Tokyo Institute for Municipal Research 1939, p.14). In addition, many cities had old waterworks that drew water from rivers and springs to residential areas without any purification systems before the installation of modern waterworks (Japan Water Works Association 1967). These facts indicate that access to water would have been easy in Japan even if no modern waterworks had been installed. These evidences support the validity of our instrument if we control for the appropriate city characteristics.
Table (ref) reports the results for MRE deaths and non-MRE deaths in Panel A. \footnote{Robustness checks and cost-benefit analysis are reported in (ref) and D. } The results of FE were estimated by our baseline specification, as shown in column (3) of Table (ref). The estimated coefficient for the MRE death rate by fixed effects model is positive and statistically significant, which is consistent with the medical and historical evidences. The coefficient is 0.742 and thus implies that, for one additional typhoid death per 1,000 people, there were 0.742 additional MRE deaths per 1,000 people. Since this value is approximately equal to three-quarters of the coefficient reported in column (3) of Table (ref), it is likely that the impact on MRE deaths accounted for a large proportion of the impact on non-typhoid deaths, although we should take into account the difference in samples between the analyses.
Our FE results show that while there is no significant relationship between the typhoid death rate and death rates of tuberculosis, bronchitis, meningitis, and heart disease, there is evidence that the incidence of typhoid increased the number of deaths from pneumonia (significant at the 5% level), even when only pneumonia was considered. The coefficient for pneumonia suggests that one additional typhoid death per 1,000 people caused 0.464 pneumonia deaths per 1,000 people. In contrast to MRE deaths, the FE results indicate that typhoid had no significant effect on non-MRE deaths, except for scarlet fever and beriberi. The results correspond with our hypothesis that the relationships between typhoid and non-MRE deaths are negligible.
Panel A of Table (ref) also presents our second stage results of the FE-2SLS model. The specification includes the same control variables as FE model. Obviously, the estimated coefficient for the MRE death rate is positive and significant, which is consistent with the fixed effects estimate. This result suggests that modern water-supply systems reduced MRE deaths by preventing typhoid fever. Our point estimate means that for one additional typhoid death per 1,000 people, there were 2.942 MRE deaths per 1,000 people. The effect in the FE-2SLS regression is approximately four times larger than that in the FE regression.
The coefficients for the tuberculosis, pneumonia, bronchitis, and meningitis death rates are also significantly positive, whereas the coefficients for all the cause-specific death rates categorized as non-MRE deaths are insignificant. This result provides the evidence to support our hypothesis. In contrast to the results that were obtained from the fixed effects model, the estimated coefficients for beriberi and scarlet fever are not significant. A possible interpretation of this result is that these diseases were directly affected by clean water.
In contrast to the comparison with the result for the non-typhoid death rate, our estimated MRE effect on respiratory deaths (i.e., tuberculosis, pneumonia, and bronchitis deaths) via the FE-2SLS model is twice as great as the effect estimated by Ferrie and Troesken (2008, p.11). While they estimated the MRE effect on respiratory deaths in Chicago to be roughly one, our estimate is roughly two. This result is consistent with the rapid decline in the mortality rate in Japanese cities relative to US cities. One possible explanation of this difference is that tuberculosis was more prevalent in pre-war Japan than in Western countries (Johnston 1995; Hunter 2003), suggesting that tuberculosis patients were the potential beneficiary of clean water at that time.
Panel B of Table (ref) shows the first-stage results of the FE-2SLS model. All the estimated coefficients of Water are significantly negative regardless of the number of observations. This result suggests that underidentification due to an irrelevant instrument is less likely to be problematic. Moreover, the $F$-statistic reported in column (1) is more than 10, indicating that we can reject the weak instrument assumption according to the criteria of Staiger and Stock (1997). This finding means that most of the second-stage results reported in Panel A are reliable. However, we cannot reject the weak instrument assumption in columns (2)--(4) of Panel B, implying that the second-stage estimates for scarlet fever, smallpox, and diphtheria shown in Panel A may be biased and should be regarded as the upper bound of the effects.
This study focuses on the potential contributions of amelioration in water quality on health: the effects of the Mills--Reincke phenomenon. We investigate these effects using panel data from Japan during the 1920s and 1930s and our estimates are consistent with the results reported by the related studies described in the Introduction.
Our findings are as follows. First, the Mills--Reincke phenomenon was observed in Japanese cities. In addition, historical records show the many complications of typhoid fever, supporting our hypothesis that typhoid fever was the key factor behind the Mills--Reincke phenomenon. Second, our estimates from the fixed effects models and FE-2SLS models suggest that a decline in typhoid deaths by one per 1,000 people decreased MRE deaths (i.e., tuberculosis, pneumonia, bronchitis, meningitis, and heart disease death rates) by 0.742 and 2.942 per 1,000 people, respectively. We also find that this chain effect on respiratory disease was greater than that observed in the United States.
From a broader view, the results of this study provide evidence to support the efficacy of public health improvements such as modern waterworks. In 2000, 216,510 people were estimated to have died of typhoid fever, especially in developing Asian countries (Crump et al. 2004, p.346; Siddiqui et al. 2006). Therefore, our estimate suggests that if public water purification systems eradicate typhoid fever, an additional 160,650--636,972 people could escape death induced by typhoid.
We wish to thank the participants in the seminars at the Tokyo Institute of Technology for their helpful comments on the paper. The work was supported by the fund for JSPS Research Fellow (Grant Number: 17J03825) and JSPS KAKENHI (Grant Number: 17K03096). There are no conflicts of interest to declare. All errors are our own.
\thispagestyle{empty}