Table 1 shows the magnitude of mortality decline in American cities during the period 1902-1929. Average annual urban death rates declined by 20% from 16.7 per thousand to 13.4 per thousand. The second column shows the dramatic fall in deaths from diarrhea, dysentery and typhoid (here called “waterborne diseases”). Deaths from these causes declined by 88% going from 8.9% of all urban deaths in 1902 to only 1.4% in 1929.

Tab. 1 -
Average Annual U.S. Urban Death Rates

Year	Total death rate (deaths per 1000)	Waterborne death rate (deaths per 10,000)	Waterborne deaths as a percentage of all deaths
1902	16.713	15.124	8.9%
1903	16.868	15.064	8.8
1904	16.686	15.377	9.1
1905	16.826	15.673	9.2
1906	17.136	16.455	9.5
1907	17.682	16.749	9.3
1908	16.370	15.171	9.1
1909	15.545	14.105	9.0
1910	16.657	14.232	8.5
1911	15.776	11.824	7.4
1912	15.234	10.072	6.6
1913	N.A.	N.A.	N.A.
1914	N.A.	N.A.	N.A.
1915	15.197	8.691	5.6
1916	15.503	9.131	5.7
1917	15.861	8.780	5.4
1918	20.691	8.279	3.9
1919	14.489	5.838	4.0
1920	14.721	5.751	3.9
1921	12.958	5.054	3.9
1922	13.272	4.080	3.0
1923	13.661	4.028	2.9
1924	13.128	3.502	2.6
1925	13.367	3.668	2.7
1926	13.774	3.240	2.3
1927	12.660	2.410	1.9
1928	13.715	2.213	1.6
1929	13.409	1.857	1.4
1929/1902	67.828	12.278
% change 1902-1929	19.769	87.722

It is well known that this dramatic decline in deaths from waterborne diseases resulted from improved sanitation in cities. Urban historians consider this period to be an era of reform, the first awakening of the environmental movement which led to an expansion in budgetary expenditures on sanitation works (Glaab and Brown, 1967; Mohl, 1985). We will examine the three elements of a modern sanitation system: water, sewers and refuse collection.

The development of modern urban water began with the delivery of water. Nearly all U.S. cities had invested in water delivery systems before 1900. In the period examined in this paper, cities were improving their water supply by building filtration plants and engaging in chemical purification with chlorine.

Sewers are an important part of the sanitation system because they remove human waste, treat it, and dispose of it where it will not contaminate the water supply. The period under study in this paper saw the building of many large, expensive sewer systems and sewage treatment facilities.

Less commonly studied than water and sewers is solid refuse collection and disposal which removes wastes from densely settled areas and disposes of it safely so that it does not contaminate the water supply or provide breeding grounds for disease spreading insects.

In order to examine the relationship between mortality decline and urban spending on sanitation, we collected mortality and expenditure data for the period 1899-1929 for all U.S. cities having a municipal water supply, populations over 100,000 in the 1920 Census, and nearly complete data on mortality experience and sanitation expenditures [1]. There are 48 cities in the sample.

The mortality data come from the Mortality Statistics of Cities which annually published death-by-cause statistics. We have constructed a waterborne death rate (WDR) series that includes deaths attributable to typhoid fever, diarrhea, and dysentery. This group of diseases will be referred to as “waterborne”, even though water is not the exclusive means of transmission. They were spread by impure food, as well as water, and by contact with feces and other filth. We expect, as did contemporaries, that these diseases were controlled by programs to deliver clean water, to remove and treat wastewater, and to collect refuse. While historical evidence on death-by-cause is notoriously problematic because of changing definitions of diseases and changes in diagnoses, those diseases studied in this paper were well identified in this period.

The data on municipal sanitation expenditures were published in various Census Bulletins to 1903 and in Financial Statistics of Cities beginning in 1905. We use data on annual operating costs and capital acquisition costs of water and sewage works. By 1907, virtually every American city had installed sewers, and most of the big cities in our sample were using filtration and chlorination to assure the safety of their water supplies (Tarr, McCurley, and Yosie, 1980). Not every series was reported every year. Few direct figures are available for 1904, and Financial Statistics of Cities was not published in 1913, 1914, or 1920 [2].

Financial data were used to construct two kinds of variables: capital variables and current operating costs variables. Expenditures on capital were aggregated over all years up to the year of observation and then divided by the population in the year of observation, thereby producing an estimate of the per capita values of the works. The per capita value of sewage facilities (SEWKALL) and refuse collection and disposal facilities (REFKALL) were both constructed in this manner [4]. The accumulated value of capital in waterworks (WATKALL) includes the value of the waterworks at the beginning of the period. This value was reported in the Census bulletins, and, for most cities, this is the value in 1899 [5]. The variable WATKALL cap-tures improvements in water quality that resulted from filtration, the construction of filter beds and plants; SEWKALL captures investments in sewage treatment technologies.

Information on annual operating expenditures for water, sewers, and refuse were used to create the variables WATERAV3, SEWERAV3, AND REFUSEAV3 which are the average operating expenditures per capita for the year under observation and the previous two years [6]. The variable WATERAV3 captures improvements in water quality resulting from disinfection, expenditures for chlorination, in addition to other annual costs.

Five variables were used for control purposes. Since the regressions are estimated on a pooled cross-section, time-series basis, three variables are used to control for variation across time. These are time, YEAR, a city's land area, LANDAREX, and its assessed valuation per capita, ASSDPCX. Land area provides a measure of geographical size and change within the study period, while the assessed valuation measures the city's ability to pay. Since this was also an era of annexation and consolidation, the inclusion of these variables controls for this type of growth.

Two dummy variables were employed. The first, WAR, includes the period of the First World War and its aftermath, which included a vigorous inflation and a virulent outbreak of influenza. The other, LATE20, controls for three years in the late 1920s when many cities overestimated their populations. The results are presented in Table 2, and variables are defined in Table 3.

Tab. 2 -
Waterborne Disease Death Rate Regression Results

Variable	Coefficient	t-ratio	Mean
WATKALL	‑0.000276	‑0.164	31.484
WATERAV3	0.028089	1.141	1.4751
SEWKALL	‑0.017236**	‑7.495	12.313
SEWERAV3	‑0.075172	‑1.078	0.28493
REFKALL	‑0.084792**	‑4.836	0.29151
REFUSEAV3	‑0.058417**	‑2.004	0.89742
YEAR	‑0.068934**	‑17.690	1915.3
ASSDPC	0.004785	1.593	10.931
LANDAREA	0.000191	0.858	263.83
WAR	0.224520**	6.394	0.12446
LATE 20	‑0.095607**	‑2.505	‑2.505

R² 0.866 N 1157 Dependent Variable is log of the Waterborne Disease Death Rate ** statistically significant at the 95% confidence level * statistically significant at the 90% confidence level

Tab. 3 -
Definitions of Independent Variables Included in Regressions

WATKALL	Sum of all capital expenditures on waterworks prior to the year under observation plus the value of municipal waterworks in 1899 (or in the year acquired) in per capita terms
WATERAV3	Average operating expenditures on waterworks and water treatment over the two preceding years and the year under observation in per capita terms
SEWKALL	Sum of all capital expenditures on sewage facilities up to the year under observation in per capita terms
SEWERAV3	Average operating expenditures on the sewer system over the two previous years and the year under observation in per capita terms
REFKALL	Sum of all capital expenditures on refuse collection and disposal up to the year under observation in per capita terms
REFUSEAV3	Average operating expenditures on refuse collection and disposal over the two preceding years and the year under observation in per capita terms
YEAR	A trend variable, the year under observation
ASSDPC	Assessed valuation in hundreds of dollars per person
LANDAREA	Square miles in hundreds of square miles
WAR	A dummy variable equal to 1, if year = 1917-20
LATE20	A dummy variable equal to 1, if year = 1925-27

The regression results presented in Table 2, show that expenditures on sewers and refuse collection had a large effect on reducing the death rates from waterborne diseases. The coefficients on the variables for expenditures on water delivery and treatment do not show a large impact in this period reflecting the fact that the large, initial effect of water-related capital expenditures on mortality from waterborne diseases took place in the latter portion of the xix^th century before our study begins.

In the early xx^th century, when we observe these cities, the largest impact of sanitation expenditures on waterborne death rates are associated with improvements in sewer services which greatly reduce the contamination of water supplies. This was especially true in cities where these water and sewer works were highly interdependent, for example a city located on a lake or a river where the manner in which the sewage was disposed might pollute the water supply. The addition of water filtration works during this period did reduce waterborne deaths, but the overall effect is small and not statistically significant. Filtration was especially effective in reducing cholera outbreaks and their associated spikes in waterborne deaths.

In a forthcoming paper we are able to examine the role of improvements in water supply in more detail (Cain and Rotella, 2001). That paper uses a larger sample of cities, albeit many for which we have less complete information, and divides cities into groups according to the type of water resource on which they are located—salt water, lakes, large rivers, smaller rivers (Cain, 1977). Cities located on salt water typically have not drawn their water supply close at hand, but rely on sources hundreds of miles removed from the city. On the other hand, they discharge their wastes into the adjacent salt water. Cities located on fresh water lakes have historically used the lake for both water supplies and waste disposal. These cities must separate the water intake and sewer outfall to avoid befouling their drinking water with their wastes. This interdependency creates difficult and costly problems. Cities located on major rivers have drawn their water upstream from the city and disposed their wastes down-stream, taking care that the potential sewage backwash not reach the water intake. Cities located on smaller rivers have had to look elsewhere for water; they utilize distant lakes and rivers or rely on well water. These cities still dispose of their wastes in the river but may have to build sewers to a down-stream point where the river can receive a large volume of wastewater. The debates about filtration that are discussed in the next section part of this paper mostly took place in river cities.

In the larger sample of cities we found that expenditures on water-related capital (WATKALL) have a large (and statistically significant) effect on reducing mortality in salt water cities. This effect does not show up in the smaller sample in this paper because the number of salt water cities is small.

The Sewer capital variable has a large negative effect on the waterborne disease death rate. This is the period in which sewage treatment is developed and the initial works constructed. The annual expenditures on sewers (SEWERAV3) also has a large negative effect.

The solid waste results are intriguing. Both variables (REFKALL and REFUSEAV3) have a large negative effect, albeit the absolute expenditures were small relative to the others. The logical tie between refuse expenditures and their impact on the death rate is not well developed. Nevertheless, given the close connection between waterborne and food borne diseases, the regular removal and disposal of food wastes in such a way as to keep them from the watercourse was an important step.

The variables which control for variation across time have the expected signs and magnitudes. The strong downward trend in mortality is captured by YEAR. These years saw tremendous increases in medical knowledge and education, as well as important changes in food preparation with canning, dehydration, and refrigeration producing large changes in the way the typical household confronted meal planning and preparation. The two variables controlling for changes in city size, ASSDPC and LANDAREA, are included because many of these cities grew by annexation and consolidation, although the great age of annexation was ending just as the study period begins (Cain, 1983). Since annexed areas could be either healthy or unhealthy, either well endowed with sanitation capital or not, the small, statistically insignificant coefficients should not be surprising.

WAR is a dummy variable for the years 1917-1920 during which three major events may have had a positive effect on waterborne death rates. The first is the effect that wartime controls and postwar inflation might have on expenditure levels, conceivably postponing some to postwar years where additional sums were paid due to inflation. The second is rapid population growth in some urban areas, much of it due to the migration of agricultural workers from the South to urban industrial jobs in the North, and to growth of cities with large military installations. The crowding this created, both during and after the war, put pressure on existing sanitation systems and may have led to increases in waterborne death rates. In fact, there is a marked slowing in the rate of decrease in these diseases, followed by an acceleration in the early 1920s. The third effect is due to the influenza epidemic of 1918-1919 which may have increased the disease environment so that more people were susceptible to the effects of waterborne contamination. The sign and magnitude of the coefficient on WAR confirms the negative impact of wartime pressures.

The second dummy variable, LATE20, is for the period 1925-1927. During this period, many cities consistently overestimated their populations given the levels reported in the 1930 Census. Thus, death rates would tend to be underestimated.

The results in Table 2 can be used to estimate the number of lives saved by an increase in expenditures on sanitation. Average municipal per capita expenditures on all six sanitation expenditures was $46.75. Over all the cities in the pooled sample, a one percent increase in all of the six categories would have let to a nearly 3% decline in the annual death rate. Put another way, an increase in annual expenditures on $4.66 would have averted 18 deaths in the average size city. We argue that this is a large payoff—a payoff that interested policy makers and the public at large.

The dominant view of urban government budgetary decision making is the incrementalist approach which implies that, in general, municipal expenditure patterns change slowly and do not respond to specific conditions or to public pressures (Wildavsky, 1984). As MacDonald and Ward argue, “the local public sector was routinely both insulated from the crises generated by socioeconomic development and autonomous from the demands of specific groups in the urban environment” (McDonald and Ward, 1984, 31). Exceptions to the incremental approach to budgeting come only when there is a particularly dramatic crisis. Perhaps an extraordinary rise in mortality would precipitate a crisis allowing for an extraordinary increase in funding for sanitation projects.

Sanitation projects in one city had strong demonstration effects, and there were many of cooperatively funded demonstration projects. These lead to interrelationships among cities that make it difficult to estimate regression models [7]. Consequently, we have examined the question of whether expenditures can be associated with mortality crises through a straightforward counting approach in which we count the number of mortality shocks in a city that were closely followed by a notable expenditure increase in that same city. We defined a mortality shock as a year in which the actual waterborne death rate was more than one standard error above its trend in that city over the period 1899-1929. The waterborne death rate shocks in our 48 cities were grouped into 98 episodes (some lasting more than one year). They are heavily concentrated in the years 1904-1910, with a few during World War I, and only 3 in the 1920s. These episodes are listed in the Appendix Table.

APPENDIX TABLE

City	Waterborne death rate shock (years)	Response	No response
Albany	00	X
	15	X
Atlanta	06-07	X
Baltimore	01	X	X
	07
	17-18		X
Boston	01	X
	05	X
	08-09		X
Buffalo	06-08	X
	18	X
Cambridge	00-01	X
	08,		X
	20	X
Camden	99	X
	16-18		X
Chicago	08-12	X
	16		X
Cincinnati	04,06	X
Cleveland	03		X
	10		X
	17		X
Columbus	04-05	X
	08	X
Dayton	07, 09-10	X
	16		X
Denver	02		X
	06-08	X
Detroit	00	X
	06		X
	10		X
	16-18	X
Fall River	07-10		X
Grand Rapids	04-06	X
	10	X
Hartford	00-01	X
	17	X
Jersey City	06-08, 10	X
Kansas City	01, 03, 05	X
20	X
Louisville	04, 06-07	X
	26	X
Lowell	05		X
	08, 10	X
Milwaukee	06, 08-10, 12		X
Minneapolis	06	X
	10	X
Nashville	00	X
	04, 06-07	X
Newark	07		X
New Bedford	10	X
18	X
New York	01	X
	05-08	X
Philadelphia	06-07	X
	10		X
Pittsburgh	02, 04	X
	06, 07	X
Portland	05-09	X
Providence	01, 03	X
	06	X
Reading	01	X	X
	06-08	X
	18		X
Richmond	01	X
	07-08		X
Rochester	06, 08-12	X
	16	X
St. Louis	01, 03		X
	11	X
St. Paul	01		X
	06	X
	09-10		X
Salt Lake City	05, 07	X
Seattle	04-06	X
Spokane	03-06	X
	10	X
Springfield	99	X
Syracuse	07-10	X
	18		X
Toledo	00	X
	05-06	X
	09-10	X
	16		X
Trenton	07-08	X
	10-11		X
Washington	01-02	X
	06	X
Wilmington	07	X
	16-18	X
Worcester	01	X
	06		X
	10-12		X
Yonkers	07-11	X
Youngstown	03	X
	10	X
	17-18	X

We defined a “response” to be a situation in which an analogously defined shock in the expenditure series (i.e., an expenditure more than one standard error above trend) occurred within three years of a shock in the mortality series. In 29 mortality shock episodes, there was no response from the affected city. In all other cases, an expenditure substantially above trend occurred within three years. We count responses in 69 of the 98 mortality episodes. So response was common, but not universal.

In most cases, the response took the form of water or sewage purification. At the beginning of the twentieth century, water and sewage purification were quite rare with 11 cities in our sample having some form of water purification and only 4 having sewage purification. The most common form of water treatment was sedimentation. Filtration was reported as serving the entire population of the city in only one city, Atlanta. There was clearly a great deal of room for a response to occur. General Statistics of Cities: 1909 indicated the number of sample cities with sewage purification systems had increased to 8. General Statistics of Cities: 1915 indicated the number of sample cities with water purification systems had increased to 33; by then none relied exclusively on sedimentation.

We turned to the Engineering News to see if we could find reports of the mortality shocks and/or the municipal expenditure responses. The Engineering News is a weekly journal of “civil, mechanical, mining and electrical engineering”. It contains articles, letters, and editorials on matters pertinent to each of these areas. A typhoid outbreak might result in a 2000-word article on causes and potential cures. The construction of a new water purification plant might be the subject of a 7500-word essay complete with a half-dozen drawings and one or two pictures.

We found discussions in the Engineering News of the waterborne disease shock in 14 of the 69 cases where there was a response (20%), and 6 of the 29 cases where there was none (21%). In each case, the disease was typhoid fever, and the role of pollution was often noted. We found a discussion of the response in 13 of the 14 cases where we found an article about the disease shock (93%), and in 28 of the 56 cases where we were unable to find mention of the shock (50%).

In some of the 6 cases where we found mention of the disease shock, but in which there was no response, we were able to identify the reasons for the lack of response. An example is the failure of Albany to respond to an increase in the typhoid death rate in 1900. Albany's new water filtration plant, then the largest in the country, opened in September 1899 (Engineering News, 14 Jan. 1900, 31; 9 Aug. 1900, 88). The number of deaths from typhoid averaged 85 per annum for the period 1890-98. The data for the first 8 months of 1899 are similar, but the total number of deaths for the last four months dropped from 24 to 7. This suggests the rate would be about 25 per annum after the opening of the filtration plant. Albany experienced 39 typhoid-related deaths through the first 11 months of the plant's operation, a rate that was unexpectedly high. No response was forthcoming for two reasons. First, the State Board of Health reported that typhoid was unusually prevalent in the state that autumn, and, second, physicians reported 14 of the 39 cases were contracted elsewhere (Engineering News, 27 June, 1901, 463).

As time passed, it became clear the filtration plant at Albany effectively reduced the incidence of waterborne diseases. The importance of demonstration effects was duly noted by the Engineering News (21 Nov. 1907, 556): “It has been obvious for a number of years past that the time was rapidly approaching when American cities would adopt filtration for all their surface water supplies, as generally as have for some decades the cities of Great Britain and Germany… Following the lead of such smaller cities as Lawrence, Mass., and Albany, N.Y., there have recently been constructed or are now being built model filtration plants for Providence, Philadelphia, Washington, Pittsburg, Cincinnati, and Louisville [8]…”

It is not at all surprising to see one city after another joining the water purification ranks. The wonder is that the delay was so long and that progress, once well begun, was not more rapid, for the typhoid statistics of the country and the increasing pollution of the streams from which the greater proportion of the volume of municipal water supplies was drawn both pointed to the emphatic need for purer water. By the turn of the century, experts and much of the public realized that polluted water increased the incidence of diarrheal disease, and the waterborne death rate, in general.

In 1904, typhoid was widespread, and the editors of Engineering News solicited accounts from those cities that were experiencing epidemics or where typhoid was endemic. Serious outbreaks occurred in Columbus, Ohio, and Pittsburgh, Pennsylvania. These episodes make clear what our counting exercise is attempting to capture, especially the long-lags and demonstration effects. The Columbus outbreak came as little surprise to the editors of Engineering News (11 Feb. 1904, 129) who described the city as being “notorious for a dangerous water supply, a high typhoid fever rate, a defective system or a lack of any real system of sewage disposal and for dilly-dallying with the needed reforms after recognizing them and securing plans to put them into effect”.

Columbus drew water at two pumping stations. The main station was located at the confluence of the Scioto and Olentangy Rivers and contained an extensive filtering chamber that significantly reduced the flow of water. When a greater flow was demanded, the city took water directly from the Scioto River, a “dangerous” source.

Columbus was well aware its supply capabilities had not kept pace with demand. Beginning in 1893, the city commissioned expert engineers to make recommendations as to how best to increase the supply. A consensus emerged within the next ten years that the Scioto should be the source and that the supply should pass through a purification plant. In addition, there were predictions of dire consequences should the city continue to use untreated river water. However, it was reported that “with surprising apathy the mass of the people have refused to produce the necessary funds for obtaining and purifying a proper supply and have condemned and actually blocked […] any steps to so procure a supply. It is not only the mass of the people […] but also and especially politicians?”(Engineering News, 11 Feb. 1904, 135)

One evening paper reported that many people believed the city was exaggerating the number of cases of typhoid reported in order to garner support for “unneeded” public improvements.

The success of filtration works elsewhere, however, led the city fathers of Columbus to undertake “an era of public improvements rarely equaled by an American city” [9]. Water purification was approved at the polls in November 1904; sewage purification, the previous year. (Engineering News, 21 Sept. 1905, 313)

The Engineering News (24 Feb. 1904, 176) reported the epidemic in Pittsburgh and Allegheny in even stronger terms. Typhoid was endemic in western Pennsylvania. Residents of both cities (their consolidation was still a few years off) were supplied with untreated river water from the Allegheny and Monongahela Rivers [10]. “These two cities have so long borne an evil reputation among sanitarians through their persistent neglect to provide a pure water supply. […] Pittsburg and Allegheny are centers of typhoid infection and as such they are a constant menace to the health of other parts of the country. […] It is time for the press of the whole country to unite in a protest against the typhoid situation in these two cities. Self-destruction, bad as it is, might be overlooked, but wholesale murder must be put down.”

As in Columbus, the remedy for the “evil” had been proposed for many years, but “political and mercenary interests” had prevented the expenditure on purification. The 5 years before the typhoid epidemic in 1904 saw many political battles, missteps, and corruption around proposed bond issues and bids for a new water purification works. The project was repeatedly postponed. Things changed after the epidemic. In December 1904, the city passed 10 ordinances to facilitate construction of the filtration works and a new bond issue was authorized. By 1907, the filtration plant was nearing completion and effective demand was building for sewage treatment as well.

Not all cities responded to mortality shocks by investing in more sanitation as did Columbus and Pittsburgh. Some cities had recently undertaken improvements and did not have the will to spend more. And there was still argument about the effectiveness of modern sanitation in preventing disease. In the early 1900, although engineers and public health professionals were convinced of the value of modern sanitation methods, politicians and the general public still seemed to have questions about the wisdom of undertaking such enormous expenditures. Uncertainty about the effectiveness of water filtration and treatment was fed by a few instances of disease outbreaks in cities which had recently built expensive plants—notably Philadelphia and Washington.

Average municipal per capita expenditures on all six sanitation expenditures considered in the regression analysis was $46.75. Over all the cities in the pooled sample, a one percent increase in each of the six categories would have saved 18 lives annually in the average-sized city. Cities could and did buy themselves lower death rates.

The benefits of investing in sanitation were high, but the costs were high too. So our second question focuses on the decision to undertake these costs by asking: is there any evidence that mortality shocks from waterborne diseases contributed to expenditures on sanitation? According to the Hollingsworths, “excepting moments of crisis, budget decision making operates to produce a process of incrementalism in public expenditures” (Hollingsworth and Hollingsworth, 1979, 157-58). We look for evidence that a local mortality shock caused cities to make a major investment in sanitation and overcome the incremental nature of city budgeting. Our counting exercise tells us that many individual cities did respond, directly and indirectly, to mortality shocks. However, a local mortality shock is only one of many possible events that could serve as a defining “moment of crisis”. For example, a mortality shock in another city could supply such a moment. What we have learned from the Engineering News suggests that demonstration effects were a particularly important factor. Therefore, we give a qualified yes to the question; urban mortality shocks did lead to increases in municipal sanitation expenditures. But, direct response to a mortality shock is just part of the explanation of city-level expenditure patterns. A full explanation will include demonstration effects from other cities, technological imperatives, and a wide range of political factors such as reform movements and machine politics. We have gone only part of the way to an explanation of how cities decided to undertake the massive public expenditures that resulted in the virtual disappearance of death from typhoid, dysentery and diarrhea in U.S. cities.