The phenomenon of summer diarrhea and its waning, 1910–1930

Abstract

During the first two decades of the 20^th century, diarrheal deaths among American infants and children surged every summer. Although we still do not know what pathogen (or pathogens) caused this phenomenon, the consensus view is that it was eventually controlled through public health efforts at the municipal level. Using data from 26 major American cities for the period 1910–1930, we document the phenomenon of summer diarrhea and explore its dissipation. We find that water filtration is associated with a 15 percent reduction in diarrheal mortality among children under the age of two during the non-summer months, but does not seem to have had an effect on diarrheal mortality during the summer. In general, we find little evidence to suggest that public health interventions undertaken at the municipal level contributed to the dissipation of summer diarrhea.

1. Introduction

Summer diarrhea (also known as “cholera infantum ” or the “disease of the season ”) was first described in the medical literature by Benjamin Rush (1794). According to Rush (1794, p. 160), it would sometimes begin “with a diarrhoea, which can last for several days without any other symptom...but it more frequently comes on with a violent vomiting and purging, and a high fever. ” He argued that it was likely caused by the “discharge of bile which generally introduces the disease…and exacerbations of the fever which accompanies it…” (Rush 1794, p. 163).

By the turn of the 20^th century, physicians had gained a better understanding of how gastronomical diseases are contracted but still had not discovered why thousands of American infants and children were dying of diarrhea every summer. In fact, to this day, we can only speculate as to precisely what pathogen (or pathogens) caused the phenomenon (Meckel 1990; Thompson and Keeling 2012).

Economists and historians have argued that summer diarrhea was controlled through the combined effects of public health interventions (Meckel 1990; Fishback et al. 2011). Specifically, they point to municipal-level efforts during the early 1900s to purify water and milk supplies. Because breastfeeding was typically discontinued at an early age (Wolf 2003), and because there is anecdotal evidence that children were at the greatest risk of dying from diarrhea in their “second summer ” (Condie 1858, p. 224; Moss 1903; Southworth 1904), focusing on these efforts is not without justification.¹

We begin our study by documenting the phenomenon of summer diarrhea using newly transcribed mortality data from 26 major American cities for the period 1910–1930.² These data, which were originally published by the U.S. Bureau of the Census, include monthly counts of diarrheal deaths among children under the age of two. We show that 21,000 children under the age of two died of diarrhea in 1910, and 14,000 of these deaths occurred in the months of June-September. By the end of the period under study, the phenomenon of summer diarrhea had largely dissipated. Fewer than 4,000 children under the age of two died of diarrhea in 1930, and fewer than 1,500 of these deaths occurred in the summer.³

Next, we build upon the work of Anderson et al. (forthcoming), who, using data on 25 major American cities for the period 1900–1940, found that filtering the water supply was associated with an 11–12 percent reduction in infant mortality. Our interest is in whether filtration and other public health efforts at the municipal level contributed to the dissipation of summer diarrhea among children under the age of two. We find that the construction of a water filtration plant is associated with a 15 percent reduction in diarrheal mortality during non-summer months, consistent with the hypothesis that transmission occurred through contaminated water. Surprisingly, there is little evidence that filtration reduced diarrheal mortality during the hottest 4 months of the year (June-September), which suggests that transmission was through contaminated food or person-to-person contact, although efforts to purify the milk supply do not appear to have affected diarrheal mortality during the summer either. In fact, we find little evidence that public health interventions can explain the waning of summer diarrhea. We conclude that the phenomenon of summer diarrhea was serious and widespread, but, contrary to the consensus view, its dissipation was not due to municipal-level efforts to purify the milk or water supplies or treat sewage. It is possible that improvements in the refrigeration chain, better nutrition, or some combination of these and other factors may have contributed to the dramatic reduction in diarrheal deaths among American infants and children during the summer months.

2. Background

Today, diarrheal disease is the second-leading cause of death among children under the age of 5 (Liu et al. 2015). Between 500,000 and 800,000 children under the age of 5 die of diarrhea every year and most of these deaths occur in developing countries (Liu et al. 2015; Kovacs et al. 2015). A wide variety of bacteria, parasites and viruses cause diarrhea and other symptoms of gastroenteritis (Hodges and Gill, 2010). Infection is usually through contaminated food or water, or person-to-person contact (Pawlowski et al. 2009). In temperate climates, bacterial infections are more common during the summer (Ramos-Alvarez and Sabin 1958; Fletcher et al. 2013), while viral gastroenteritis is more common during the winter (Pawlowski et al. 2009). In tropical climates, the incidence of diarrhea peaks during the rainy season (Zhang et al. 2010; Phung et al. 2015; Xu et al. 2015; Kulinkina et al. 2016; Muluken et al. 2017).⁴

Over 70 percent of total deaths from diarrhea occur among children under of the age of two (Walker et al. 2013). Susceptibility is highest at 6–11 months, presumably because exclusive breastfeeding protects against infection and crawling brings children into contact with human and/or animal feces (Walker et al. 2013; Mduduzi et al. 2015). Death is typically caused by dehydration and loss of electrolytes (King et al. 2003), although malnutrition is often a contributing factor (Baqui and Ahmed 2006).

2.1. Previous Research on Summer Diarrhea

In the decades leading up to its dissipation, summer diarrhea received a great deal of attention from physicians, who described its symptoms, noted that its victims were often born in crowded tenement housing districts, and proposed various causes.⁵ One school of thought held that exposure to summer heat was directly responsible for the annual wave of diarrheal deaths among infants and children (Miller 1879; Schereschewsky 1913), while another held that overfeeding was the cause (Burg 1902; Tilden 1909). Even among physicians who believed that summer diarrhea was caused by bacteria, there were several competing theories as to the mode of transmission (Zahorsky 1913, 1918). For instance, Hewitt (1910) and Youmans and Youmans (1922) argued that houseflies were the principal vector of transmission, a possibility that cannot be dismissed out of hand (Levine and Levine 1991; Förster et al. 2007).⁶

Since its dissipation, only a handful of studies have examined the phenomenon of summer diarrhea. Cheney (1984) focused on the experience of Philadelphia during the period 1869–1921, while Condran (1988) focused on New York City during the period 1870–1919. These authors noted that infant mortality spiked every summer through the early 1900s, due principally to diarrheal diseases. By the second decade of the 20^th century, the phenomenon of summer diarrhea had begun to wane in both Philadelphia and New York City, which Cheney (1984) and Condran (1988) attributed to the purification of municipal milk supplies.

Certainly, many contemporary physicians and public health experts were convinced that purifying the milk supply was key to reducing diarrheal mortality during the summer.⁷ The refrigeration chain was still missing important links during this period (Rees 2013), and bacteria such as E. coli and Shigella grow rapidly in warm, moist summer conditions (Winfield and Groisman 2003; Noor 2013; Girma 2015). It is, however, difficult to rule out the possibility that other factors, including efforts to purify the water supply, contributed to the observed reduction.

Condran and Lentzner (2004) used data from Chicago, New Orleans, and New York for the period 1870–1917 to document the phenomenon of summer diarrhea and its waning. These authors found that excess mortality during the summer months fell gradually after the turn of the 20^th century, but noted that identifying the cause of this phenomenon is made exceedingly difficult by the large number of public health interventions that were undertaken at the municipal level. They concluded that better sanitation and efforts to purify the milk and water supplies likely contributed to the decline in summer diarrhea but wrote that no single factor is “sufficient to understand either the poor life chances of infants in nineteenth-century cities or the improvements in those chances…” (Condran and Lentzner 2004, p. 352).

3. Summer Diarrhea, 1910–1930

Our focus throughout is on diarrheal deaths among children under the age of two in major American cities, defined as the 26 most populous cities as of 1910. City-level counts of diarrheal deaths are available by month for the period 1910–1930 from Mortality Statistics, which was published annually by the U.S. Census Bureau.⁸ These counts, which are listed under the title “diarrhea and enteritis ”, include deaths due to cholera infantum, colitis, diarrhea, enteritis, enterocolitis, gastroenteritis, summer complaint⁹, and other similar causes (United States Bureau of the Census 1910).

In 1910, there were 21,101 diarrheal deaths among children under the age of two in the 26 most populous American cities (Figure 1), accounting for 30 percent of total mortality in this age group.¹⁰ Two-thirds of these deaths occurred in the months of June-September. Between 1910 and 1930, diarrheal mortality fell by 83 percent, to 3,513. The reduction in diarrheal deaths during the months of June-September was even more pronounced: only 1,482 children under the age of two died from diarrhea in the summer of 1930, a reduction of almost 90 percent as compared to the summer of 1910.¹¹

Figure 1. Diarrheal Mortality among Children Under the Age of Two.

Notes: Based on data from Mortality Statistics for the 26 cities under study, published by the U.S. Census Bureau. The summer months are defined as June-September.

Figure 2 shows diarrheal deaths by month among children under the age of two per 100,000 population.^{12, 13} It is clear from this figure that seasonality waned considerably during the period under study. For instance, the diarrhea mortality rate was 3.8 in January of 1910 but reached as high as 29.1 in July of that same year, a peak-to-trough ratio of 7.7. The peak-to-trough ratio had fallen to 3.5 by 1920, and it had fallen to 2.6 by 1930.¹⁴

Figure 2. Monthly Diarrheal Mortality Among Children Under the Age of Two per 100,000 Population.

Notes: Based on data from Mortality Statistics for the 26 cities under study, published by the U.S. Census Bureau. The shaded vertical bars indicate the summer months (June-September).

4. Public Health Efforts and the Waning of Summer Diarrhea

Economists and historians have long asserted that public health efforts undertaken at the municipal level caused the waning of summer diarrhea (Cheney 1984; Condran 1988; Meckel 1990; Fishback et al. 2011). For instance, Cheney (1984) and Condran (1988) asserted that efforts to purify milk supplies caused its waning, while Meckel (1990) and Fishback et al. (2011) pointed to the combined effects of purifying milk and water supplies.¹⁵

The best evidence on public health interventions and diarrhea mortality in American cities at the turn of the 20^th century comes from Anderson et al. (forthcoming) and Alsan and Goldin (2019). Using data on 25 major American cities for the period 1900–1940, Anderson et al. (forthcoming) found that water filtration was associated with an 11–12 percent reduction in infant mortality and a (statistically insignificant) 15 percent reduction in diarrheal deaths among children under the age of two. Alsan and Goldin (2019) analyzed data on 60 Massachusetts municipalities from the period 1880–1920. Instead of filtering or chlorinating their water supply, these municipalities built a series of “impounding reservoirs in which spring floodwaters were stored ” (Alsan and Goldin 2019, p. 595). Alsan and Goldin (2019) found that the provision of clean water combined with access to a regional sewerage system was associated with a 23 percent reduction in mortality among children under the age of 5. Over half of this reduction was due to fewer deaths from gastrointestinal diseases, a category that included diarrhea.¹⁶

The remainder of our study builds upon the research described above by exploring the relationship between municipal public health interventions and the phenomenon of summer diarrhea during the period 1910–1930. We begin our exploration of this relationship by estimating the following regression:

$$\text{ln}({\mathit{Diarrhea}}_{\mathit{ct}})={\beta }_0+{\beta }_1{\mathrm{Filtration}}_{\mathrm{ci}}\times {\mathit{Non\; -\; Summer}}_t+{\beta }_2{\mathit{Filtration}}_{\mathit{ct}}\times {\mathit{Summer}}_t+{\beta }_3{\mathit{Chlorination}}_{\mathit{ct}}\times {\mathit{Non\; -\; Summer}}_t+{\beta }_4{\mathit{Chlorination}}_{\mathit{ct}}\times {\mathit{Summer}}_t+X_{\mathit{ct}}^{\prime }{\beta }_5+{\theta }_c+{\delta }_t+{\epsilon }_{\mathit{ct}},$$

where Diarrhea is equal to diarrheal deaths per 100,000 population among children under the age of two in city c and month t, where t = 1…252.¹⁷ Filtration is an indicator for whether a water filtration plant was in operation and Chlorination is an indicator for whether the water supply was chemically treated.¹⁸ These indicators are interacted with two mutually exclusive season dummies: Summer, which is equal to 1 for the months of June-September and equal to zero for the non-summer months; and Non-Summer, which is equal to 1 for the months of October-May and equal to zero for the summer months.

Demographic controls, based on information from the 1910, 1920, and 1930 Censuses (and linearly interpolated for intercensal months), are represented by the vector X_ct and are listed in Appendix Table 1, along with descriptive statistics and definitions. City-level characteristics include the natural log of population and percentages of the population by gender, race, foreign-born status, and age group. City and month-by-year fixed effects are represented by the terms θ_c and δ_t, respectively. The city fixed effects control for determinants of diarrheal mortality that were constant over time, and the month-by-year fixed effects control for common shocks. All regressions are weighted by city populations and standard errors are corrected for clustering at the city level (Bertrand et al. 2004).¹⁹

Our primary interest is in the parameters β₁ through β₄. β₁ represents the effect of filtration on diarrheal mortality in the non-summer months, while β₂ represents the effect of filtration in the months of June-September; similarly, β₃ represents the effect of chlorination on diarrheal mortality in the non-summer months, while β₄ represents the effect of chlorination in the months of June-September. Based on the results of previous studies (Knutsson 2017; Bhalotra et al. 2018; Anderson et al. forthcoming) and contemporary accounts (Elms 1914; Jennings 1918; Morse and Hall 1919), we expect the estimates of β₁ through β₄ to be negative. Because filtration and chlorination are more effective at eliminating bacteria than viruses (Centers for Disease Control 2012; Hijnen et al. 2004; Jenkins et al. 2011) and because bacterial infections are most common during the summer (Ramos-Alvarez and Sabin 1958; Fletcher et al. 2013), we expect ${\widehat{\beta }}_2-{\widehat{\beta }}_1<0$ and ${\widehat{\beta }}_4-{\widehat{\beta }}_3<0$.²⁰

4.1. Main Results

Estimates of β₁ - β₄ are reported in the first column of Table 1. The estimated effects of filtration and chlorination in the non-summer months are negative, but not statically significant at conventional levels. Their estimated effects for the summer months are positive but, again, are statistically indistinguishable from zero.

Table 1.

Public Health Interventions and Summer Diarrhea, 1910–1930

	(1) ln(Diarrhea)	(2)	(3)	(4)
Filtration × Non-Summer	−.092 (.075)	−.111 (.068)	−.100 (.060)	−.161*** (.047)
Filtration × Summer	.069 (.128)	.103 (.142)	.088 (.134)	.029 (.108)
Chlorination × Non-Summer	−.014 (.079)	−.013 (.071)	−.043 (.074)	−.029 (.062)
Chlorination × Summer	.055 (.060)	.047 (.056)	.031 (.052)	.052 (.056)
Clean Water Project × Non-Summer	…	.021 (.084)	.016 (.103)	.120 (.131)
Clean Water Project × Summer	…	.128 (.113)	.133 (.125)	.236 (.156)
Sewage Treated × Non- Summer	…	.022 (.089)	.031 (.082)	.034 (.074)
Sewage Treated × Summer	…	.026 (.128)	.049 (.117)	.051 (.077)
Bacteriological Standard × Non-Summer	…	…	.219*** (.062)	.111* (.060)
Bacteriological Standard × Summer	…	…	.077 (.082)	−.024 (.074)
City-specific linear trends	No	No	No	Yes
N	6,552	6,552	6,552	6,552
R2	.840	.841	.844	.858

^*Statistically significant at 10% level;

^**at 5% level;

^***at 1% level.

Notes: Based on annual data from Mortality Statistics for the period 1910–1930, published by the U.S. Census Bureau. Each column represents the results from a separate OLS regression. The dependent variable is equal to the natural log of the number of diarrheal deaths among children under the age of two per 100,000 population in city c and month t. Controls include the city characteristics listed in Appendix Table 1, city fixed effects, and month-by-year fixed effects. Regressions are weighted by city population. Standard errors, corrected for clustering at the city level, are in parentheses.

In the second column of Table 1, we introduce two additional municipal-level public health interventions, both of which are interacted with the season indicators. The first, Clean Water Project, is equal to 1 if a new aqueduct or underground tunnel was built to deliver clean water to the municipality and is equal to zero otherwise.²¹ The second, Sewage Treated, is an indicator for whether the city treated its sewage before discharging it into local waterways.²² With their inclusion, water filtration is associated with an 11 percent reduction in diarrheal mortality during the non-summer months (e ^−.111 − 1 = −.105) and an 11 percent increase during the summer months (e^.103 − 1 =.108). It should be noted, however, that neither of these estimates is statistically significant. Likewise, the estimated coefficients of the Clean Water Project and Sewage Treated interactions are statistically indistinguishable from zero.

In the third column of Table 1, we introduce an indicator for whether city c required milk sold within its limits to meet a strict bacteriological standard. During the period 1910–1930, 15 cities passed ordinances requiring that milk sold within their limits meet a bacteriological standard (Appendix Table 5). Because such ordinances were difficult to meet without resorting to pasteurization (Meckel 1990, pp. 88–89), they are often referred to as “pasteurization ordinances ” (Harding 1917, p. 57; Swinford 2016, p. 254; Komisarow 2017, p. 131). For instance, New York passed an ordinance, effective on January 1, 1912, requiring that “certified ” raw milk have less than 30,000 bacteria per cubic centimeter, and that “inspected ” raw milk have less than 60,000 bacteria per cubic centimeter.²³

Anderson et al. (forthcoming) found little evidence that setting a bacteriological standard for milk reduced infant mortality. Our results are consistent with theirs. Specifically, the estimated coefficient of the bacteriological standard indicator interacted with Non-Summer is actually positive and statistically significant at the.05 level. The estimated coefficient of the interaction between Bacteriological Standard and the summer indicator is also positive but smaller in magnitude and not sufficiently precise to reject the null.

In the final column of Table 1, we report estimates from a regression model that controls for city-specific linear time trends. These trends are designed to capture smoothly evolving social and demographic factors that could have affected diarrhea mortality among infants and children at the city level.²⁴ With their inclusion, filtration is associated with a statistically significant 15 percent reduction in diarrhea mortality during the non-summer months (e^−.161 − 1 =.149). By contrast, the estimated effect of filtration on diarrhea mortality among children under the age of 2 during the summer months is small, positive and statistically insignificant at conventional levels. Based on the lower bound of the 90 percent confidence interval of this estimate, we can rule out mortality reductions greater than 14 percent during the summer.²⁵

To gauge the contribution of water filtration to the overall decline in non-summer diarrhea mortality from 1910 to 1930, we calculated a predicted diarrhea mortality rate for every non-summer month (and its 90 percent confidence interval) under the as- sumption that no filtration plants were built. The results from this exercise are shown in Appendix Figure 4.²⁶ The solid line represents the actual non-summer diarrhea mortality rate per 100,000 population among children under the age of two and the dashed line represents the predicted rate. Had no filtration plants been built, we predict that the January non-summer diarrhea mortality rate would have fallen from 2.99 to 1.53, or almost 50 percent. In reality, the January non-summer diarrhea mortality rate fell by more than 70 percent, to 0.84.²⁷

4.2. Event-Study Analysis

The negative association between filtration and diarrhea mortality in the non-summer months reported in column (4) of Table 1 is consistent with the hypothesis that transmission occurred through contaminated water. To explore the robustness of this association, we replace the filtration indicator with a series of its leads and lags, each of which is interacted with Non-Summer. The results from this exercise are reported in Panel A of Figure 3. During the first two post-adoption years, water filtration is associated with a 12 percent reduction in non-summer diarrheal mortality, and this effect appears to be fairly stable over time, although the coefficients of the subsequent lags are measured with less precision. Panel B of Figure 3 provides additional evidence that filtration was not effective at reducing diarrheal deaths in the summer. Importantly, neither of these event studies provide evidence that diarrheal mortality was trending differently in treated versus non-treated cities prior to the adoption of filtration technology.

Figure 3. Pre- and Post-Filtration Trends in Diarrheal Mortality among Children Under the Age of Two.

Notes: OLS coefficient estimates (and their 90% confidence intervals) are reported, where the omitted category is one year before treatment. The dependent variable is equal to the natural log of the number of diarrheal deaths among children under the age of two per 100,000 population in city c and month t. Controls include the city characteristics listed in Appendix Table 1, interactions between the remaining public health interventions and the Summer and Non-Summer indicators, city fixed effects, month-by-year fixed effects, and city-specific linear trends. Regressions are weighted by city population. Standard errors are corrected for clustering at the city level.

5. Conclusion

In the United States at the turn of the 20^th century, tens of thousands of American children would die from diarrhea-related disease every summer. One of the principal contributions of this study is to document the scale of this phenomenon. Using newly transcribed diarrheal mortality data for 26 major American cities, we find that summer diarrhea was an important cause of child mortality. In 1910, the year in which mortality counts at the month level first became available, there were over 21,000 diarrheal deaths among children under the age of two in these cities, accounting for 30 percent of total mortality in this age group; two-thirds of diarrheal deaths among children under the age of two occurred in the months of June-September.

The phenomenon of summer diarrhea had largely dissipated by 1930, when only 3,513 children under the age of two died from diarrhea in the cities under study, and only 1,482 of these deaths occurred in the summer months. The precise cause of summer diarrhea was never isolated and the memory of its toll eventually receded.

Economists and historians generally believe that the dissipation of summer diarrhea was due to public health efforts undertaken at the municipal level (Cheney 1984; Condran 1988; Meckel 1990; Fishback et al. 2011). Evidence for this belief, however, is anecdotal or based on only a handful of case studies (Cheney 1984; Condran 1988; Condran and Lentzner. 2004). In addition to documenting the phenomenon of summer diarrhea, we explore whether its waning over the period 1910–1930 was, in fact, related to public health interventions undertaken at the municipal level.

We find no evidence that treating sewage or setting bacteriological standards for milk resulted in the dissipation of summer diarrhea, which is consistent with the results of Anderson et al. (2019a) and Anderson et al. (forthcoming), who concluded that public health efforts at the municipal level were not important drivers of the urban mortality transition. Likewise, we find no evidence that water filtration reduced diarrhea mortality among children under the age of two during the months of June-September, which is perhaps surprising given that water filtration plants built at the turn of the 20^th century are associated with sizeable reductions in infant mortality (Anderson et al. forthcoming) and that diarrhea-causing bacterial infections are most common during the summer (Ramos-Alvarez and Sabin 1958; Fletcher et al. 2013).

By contrast, we find that filtering water is associated with a 15 percent reduction in diarrhea mortality among children under the age of two during the non-summer months. Between 1910 and 1930, 8 of the 26 cities in our sample adopted water filtration technology and non-summer diarrhea mortality fell by more than 70 percent. Had none of these cities adopted filtration technology, we estimate that the non-summer diarrhea mortality rate would have fallen by 50 to 60 percent.

We conclude by inviting further investigation in this area and acknowledging that we still do not know what caused the waning of summer diarrhea. Although improvements in medical care, the adoption of more hygienic practices, or better living conditions may have been responsible for its dissipation, it is also possible that the introduction of refrigeration technology played a role. During the period under study, important links in the refrigeration chain were forged (Rees 2013). As mechanical ice replaced ice cut from lakes and ponds and refrigerated transportation of perishables became the norm, city dwellers gained year-round access to fresh, unspoiled milk, meat, and produce.