Abstract

Unlike other public utilities, most water in the United States is supplied by publicly owned and operated waterworks. The predominance of the public sector in the supply of water was not always the case, however; private firms dominated US water supply throughout most of the 19th century. This article analyzes the puzzle of why water and sanitation systems were the only major utilities to become predominantly public by, first, reexamining historical accounts of the problems of contracting for water services in light of modern theories of economic organization and, then, evaluating hypotheses derived from those accounts using data on 373 waterworks serving US municipalities with populations over 10,000 in 1890. Among other results, municipal ownership is found to be related to the distribution of population and commerce within a city in ways that suggest that frictions between cities and private companies over system extensions and improvements played a significant role in the shift to municipal ownership.

Introduction

Most water in the United States, as in much of the rest of the world, is supplied through publicly owned and operated waterworks. Approximately 80% of US water supply firms were publicly operated in 1970 (Crain and Zardkoohi 1978; Feigenbaum and Teeples 1983); in 1995, 82% of water systems serving more than 500 customers and 86% of systems with over 3000 customers were publicly owned (US Environmental Protection Agency 1999, 2002). The predominance of the public sector in water supply was not always the case, however. Private water firms dominated US water supply for most of the 19th century. Of the 16 waterworks in operation in the United States in 1800, only one (in Winchester, Virginia) was publicly owned (Baker 1899: 14). As late as 1890, the majority (56%) of the 1878 waterworks then in existence remained private (ibid., 16). Even as public ownership rose, private water companies maintained a foothold in major urban markets into the 20th century. In 1917, “five important water companies that enjoy public franchises [were] still operating within the limits of greater New York. Indeed, about one fourth of the city's area, and 400,000 of its population, [remained] dependent directly upon the private companies for both domestic service and public fire protection” (Wilcox 1917: 550). The trend was clearly toward municipal ownership of water service, however, and by the turn of the century, the establishment of new municipally owned waterworks, combined with the conversion of private works to public control, had tilted the balance to public ownership nationally (Baker 1899: 26).

The shift to public ownership of waterworks was decidedly not part of a more general movement away from private provision of utility services. At the end of the century, when the majority of waterworks were municipally owned and operated, ownership of other utilities remained overwhelmingly private. In 1899, only 15% of electric light plants and less than 2% of gas works were publicly owned (US Bureau of Labor 1900: 12). The dominance of private supply of major public utility services continued into the 20th century: “In street railways … gas works … electric light and power plants, private ownership [was] preponderant … [and in] telephones private ownership [was] universal, or practically so” (Wilcox 1931: 5). Fewer than 5% of the more than 1000 gas companies operating in 1920 were publicly owned (Troesken 1997). After peaking at 8% in 1900, the share of electrical power generated by municipally owned utilities declined to 5% by 1932, compared to 94% for investor-owned utilities, with the federal and state governments producing the remaining 1% (Edison Electric Institute 1973: 24); in 1980, the shares were municipal, 3.8%; investor-owned, 78%; federal, state, and district governments, 15.5%; and cooperatives, 2.8% (Edison Electric Institute 1982: 20).1 Even public transit systems remained predominantly private until fairly recently: In 1948, only 36 of 1400 transit systems (fewer than 6%) were publicly owned (Jacobson and Tarr 1995: 10); the proportion of publicly owned urban transit systems was 32% in 1960 and did not exceed 50% until the 1970s (ibid.; see also Pashigian 1976: 1239–40). The only other utility service for which public ownership has been as prevalent as for water is sewer and drainage systems, where, from the beginning, “public ownership [was] so nearly universal as to remove this utility from the realm of controversy respecting public ownership” (Wilcox 1931: 4–5).

Why water and sanitation systems were so much more likely than other utilities to be publicly owned and operated has long been a puzzle. Noting the disparities in utility ownership, Delos Wilcox, probably the most prominent expert on public utilities of the late-19th and early-20th centuries, attributed the acceptance of municipal waterworks—even by “opponents of general municipal ownership of public utilities”—“partly to the nature of the utility and partly to the fact that the policy of municipal ownership of water works has become well established in this country” (1910: 398). But what in the nature of water supply led to such acceptance in the first place Wilcox was unable to say. Almost a century later, a resolution of the puzzle appeared no closer. Referring to “the aberrant case of waterworks” (1993: 314), George Priest concluded a detailed examination of 19th-century public utility governance by observing that “There is no clear explanation for why, of the variety of public utilities, water supply was the most likely to be publicly owned and managed” (317). The issue has continued to provoke scholarly debate (e.g., Melosi 2000; Troesken and Geddes 2003; Cutler and Miller 2006).

I approach the waterworks ownership puzzle by, first, reviewing late-19th- and early-20th-century accounts of the problems of public utility governance for insights into the origins of the preference for public ownership of water and sewer systems relative to other utilities. The historical literature suggests two theoretically plausible explanations for this divergence. The first relates to what Wilcox (1910) called the “administrative simplicity” of water and sanitation systems. In modern terms, because water and sanitation systems required less expertise and supervision to run than other public utilities—and, consequently, were less in need of high-powered incentives to achieve satisfactory performance—publicly owned water and sanitation systems were simply less disposed than other utilities to the operational and administrative inefficiencies typically associated with government ownership. The second explanation, manifest in descriptions of the major sources of friction between cities and private water companies, attributes the prevalence of municipal ownership to sources of contracting hazards peculiar to water and sewer systems, specifically (a) to constraints on pricing arrangements that necessitated contentious negotiations whenever system extensions or improvements were desired and (b) to the disruptiveness of street excavations, which provided a valuable, if costly, tactic in those negotiations.

To assess whether these considerations might have contributed to the conspicuous disparity in ownership between utility sectors, I investigate the extent to which administrative simplicity and contracting frictions can account for the observed variation in ownership among waterworks at the end of the 19th century. Because pumping and filtering increase the operational complexity of water (and sewer) systems, and thus the need for attention and expertise, Wilcox's administrative simplicity conjecture implies that municipal ownership should have been less likely for waterworks that pumped or filtered water. If, meanwhile, extensions and excavations were important sources of friction between cities and private water companies, municipal ownership should have been more prevalent (for reasons detailed below) in cities with densely populated centers and sparsely populated outlying areas than in communities with more uniformly distributed populations.

In contrast to prior efforts to explain the rise of municipal ownership of waterworks, which have relied either on historical accounts of a few, usually large, US cities or on broad aggregate trends, I confront the theory with data on the ownership and characteristics of 373 waterworks serving 346 cities with populations over 10,000 in 1890.2 The evidence shows municipal ownership of waterworks to be correlated with measures of service density and population distribution in ways that appear consistent with the sources of frictions between cities and water companies emphasized in the historical literature. The results also provide evidence, although weaker, supporting Wilcox's speculation that the relative administrative simplicity of water and sewer systems contributed to the viability of municipal provision compared to other utility services. By contrast, municipal ownership does not appear to correlate with the size either of cities or of waterworks and associated investments, contrary to what would be expected if either relationship-specific investments or capital market imperfections were the principal reason for the failure of private provision.

I begin the analysis with a brief history of water and sanitation system development in the United States, followed in Section 3 by a discussion of several prior explanations for the municipal ownership of waterworks and their limitations. In Section 4, I review 19th-century perspectives on the problem of public utility governance in light of modern theories of contracting and organization and develop the administrative simplicity and relational frictions hypotheses outlined above. The empirical analysis appears in Section 5, followed by concluding remarks in Section 6.

Urbanization and Investment in US Water and Sanitation Systems

Large-scale water systems have all the characteristics of public utilities. As one modern observer describes it, “The water industry is a classic case of natural monopoly… . Direct competition between firms in the provision of networks of mains and sewers within a given region would entail inefficient duplication of fixed assets… . The water supply business is capital-intensive, and average asset lives are very long” (Cowan 1993: 15–6).3 This was not always (and is still not everywhere) the case, however. Given water's indispensability for survival and its value in transportation and industrial and agricultural uses, the availability of reliable water supplies has naturally been an important consideration in settlers’ choice of location. Not surprisingly, therefore, as the North American continent was colonized, lands adjacent to water resources tended to be the most valuable and first settled. In small communities, individual or small-scale cooperative collection or extraction of water by means of wells or available surface sources such as rivers or lakes tended to be adequate: “Except in those portions of the country in which the rainfall is scanty, it is comparatively easy for individuals to supply themselves with water for necessary uses until population becomes congested” (Wilcox 1910: 399). Where they existed, cooperative water projects tended to be rather modest operations that raised few of the problems associated with public utilities: “Some of the earliest waterworks plants of the country consisted of a small and short pipeline from a spring on a hillside to a few houses below it. Those supplied with water … contributed equally to the cost of construction and the small expense necessary to maintenance… . The water supply in such cases is purely a co-operative affair, each paying his proper share of the expense involved, and no attempt being made to realize a profit for the benefit of any one” (Baker 1899: 37).4

Over time, however, population growth, industrial development, and technological change combined to increase demands on original water sources, reducing the practicality of individual and small-scale collective water supply. As claims on lands directly adjacent to rivers and lakes were gradually exhausted, newer settlers found it necessary to locate further from water sources, creating the need to transport water over longer distances. Exacerbating the problem, population growth and urban encroachment began to threaten local sources with depletion and pollution (Wilcox 1917: 558–9):

[T]he continuous encroachment of population on the watersheds … results in a gradual lessening of the amount of water available from these sources, while at the same time the increased demand for it tends to cause an overdraft upon the wells… . [T]he possibility of pollution as the growing city flows out to encompass these water-bearing lands tends to weaken the consumer's predilections for his local supply as compared with the supply from the mountains which the city now offers.

The construction of waterworks in the 19th century paralleled trends in population, industrialization, urbanization, and wealth. Between 1800 and 1900, the population of the United States rose from 5.3 to 76 million (US Bureau of the Census 1990). Whereas roughly 80% of the labor force was employed in agriculture in 1800 (Weiss 1992: 22), that proportion had fallen to 56% or less by 1860 (ibid.) and to 40% by 1900 (Lebergott 1964: 510). Correspondingly, and more importantly, the percent of the population living in areas classified as urban (population in excess of 2500) rose greatly, from 6% in 1800 to 40% 100 years later, while the number of places classified as urban rose from 33, with an average population below 10,000, in 1800, to 1743, with an average population just over 17,000, in 1900 (US Bureau of the Census 1990: 5). By 1905, 154 cities in the United States had populations of 30,000 or more (US Bureau of the Census 1907: 112–3); in 1800, there were only 2 (Gibson 1998). And by 1890, three cities—New York, Philadelphia, and Chicago—had populations in excess of 1 million. After increasing at a slow but steady rate for the first half of the century, with an average of two new works added per year from 1800 to 1860, the number of waterworks shot up dramatically after the Civil War (see column 1 of Table 1). Between 1880 and 1896, the number of waterworks increased from 598 to 3196; between 1890 and 1896 waterworks were being constructed at an average of 220 per year.

Table 1.

Public and Private Waterworks, 1800–1915

Year Total Public Private Percent of total
 
Public Private 
1800 16 15 6.2 93.8 
1805 23 21 8.7 91.3 
1810 26 21 19.2 80.8 
1815 26 21 19.2 80.8 
1820 30 25 16.7 83.3 
1825 32 27 15.6 84.4 
1830 44 35 20.5 79.5 
1835 54 15 39 27.8 72.2 
1840 64 23 41 35.9 64.1 
1845 70 27 43 38.6 61.4 
1850 83 33 50 39.8 60.2 
1855 106 48 58 45.3 54.7 
1860 136 57 79 41.9 58.1 
1865 162 68 94 42.0 58.0 
1870 243 116 127 47.7 52.3 
1875 422 227 195 53.8 46.2 
1880 598 293 305 49.0 51.0 
1885 1013 447 566 44.1 55.9 
1890 1878 806 1072 42.9 57.1 
1896 3196a 1690 1489 52.9 46.6 
1899 3326 1787 1539 53.7 46.3 
1915 4419b 3036 1346 68.7 30.5 
1924 9850 6900 2950 70.0 30.0 
Year Total Public Private Percent of total
 
Public Private 
1800 16 15 6.2 93.8 
1805 23 21 8.7 91.3 
1810 26 21 19.2 80.8 
1815 26 21 19.2 80.8 
1820 30 25 16.7 83.3 
1825 32 27 15.6 84.4 
1830 44 35 20.5 79.5 
1835 54 15 39 27.8 72.2 
1840 64 23 41 35.9 64.1 
1845 70 27 43 38.6 61.4 
1850 83 33 50 39.8 60.2 
1855 106 48 58 45.3 54.7 
1860 136 57 79 41.9 58.1 
1865 162 68 94 42.0 58.0 
1870 243 116 127 47.7 52.3 
1875 422 227 195 53.8 46.2 
1880 598 293 305 49.0 51.0 
1885 1013 447 566 44.1 55.9 
1890 1878 806 1072 42.9 57.1 
1896 3196a 1690 1489 52.9 46.6 
1899 3326 1787 1539 53.7 46.3 
1915 4419b 3036 1346 68.7 30.5 
1924 9850 6900 2950 70.0 30.0 

Source: Years 1800–1896: Baker (1899: 16); year 1899: US Bureau of Labor (1900: 12); year 1915: McGraw Waterworks Directory (1915: 615); year 1924: Waterman (1934) as cited in Melosi (2000: 120).

a

Ownership of 17 waterworks in 1896 was listed as “joint or unknown.”

b

Thirty-seven waterworks listed as “mixed ownership.”

Propelled in part by the proliferation of waterworks, the later years of the 19th century also saw the introduction and propagation of large-scale sewer systems. Increases in the availability of water increased the demand for water closets, which in turn greatly increased the quantity of sewage in need of disposal.5 In turn, the effective operation of sewer systems depended on the availability of ample quantities of water: “The wastes which, unless removed, tend to contaminate the local sources of water supply, cannot be removed except by means of a public system of water works. In other words, in order to remove the cause of the pollution of individual water supplies in cities, it is necessary to have a public supply; a sewage system is impossible without water works” (Wilcox 1910: 399; see also Cain 1977; Tarr 1996). The result was that, whereas no American city had sanitary sewers in 1857, sewer systems had become common in urban areas by the end of the century: two-thirds of urban residents in 1880 and over 80% in 1900 lived in areas served by sanitary sewer systems (see Table 2). Of major US cities, only New Orleans and Baltimore managed without sanitary sewer systems into the late 1800s (to the “disgust … of the average American citizen”; Wilcox 1910: 452).

Table 2.

US Population and Sewage Systems, 1880–1900

Year Total Urban With sewers With water treatment With sewage treatment 
1880 50,155,783 14,129,735 9,500,000 30,000 5000 
1890 62,947,714 22,106,265 16,100,000 310,000 100,000 
1900 75,994,575 30,159,921 24,500,000 1,860,000 1,000,000 
Year Total Urban With sewers With water treatment With sewage treatment 
1880 50,155,783 14,129,735 9,500,000 30,000 5000 
1890 62,947,714 22,106,265 16,100,000 310,000 100,000 
1900 75,994,575 30,159,921 24,500,000 1,860,000 1,000,000 

Source: Tarr (1996: 194).

The large investments—in distant impoundments, aqueducts, and distribution mains to exploit more distant water supplies or, alternatively, in facilities to preserve or purify nearby sources—that were needed to accommodate the fresh water demands of growing urban populations gave water supply its public utility characteristics. Those investments, though large, were not unique, however. In addition to water and sanitation systems, the century also saw the introduction of such modern public utility conveniences as electricity, telephones, and public transportation, plus a few, such as manufactured gas, that have since become obsolete. What differed was their governance. Like other utilities, early waterworks were mostly privately owned (see, again, Table 1). But, whereas the provision of other utility services remained mostly private, the proportion of waterworks run by private firms declined over the course of the century, from more than 90% in the first decade to under half by 1896 and to less than a third by 1915. For their part, sewer systems were, with rare exception, publicly owned from the beginning.6

Why Water?

At least a half-dozen tentative explanations for the disparate ownership of water and sanitation systems and the overall trend away from private ownership of waterworks have been offered or can be readily inferred from more general theories of economic organization. As Troesken and Geddes (2003) observe, however, most of these explanations exhibit theoretical or empirical shortcomings. One appealing hypothesis, for instance, is that the shift toward municipal ownership reflected changes in the political and legal environment toward the end of the 19th century. As Melosi describes it, “Several factors account for the political and economic climate that increasingly favored public [water] systems in the late 19th century: (1) the improved fiscal status of cities; (2) growing cooperation between large cities and state legislatures in the development or expansion of services; (3) skepticism of private companies to deliver services equitably and comprehensively; and (4) broadening of the regulatory power of cities, and especially states, with respect to public utilities” (2000: 120). In a similar vein, Cutler and Miller ascribe the late-century growth in municipal ownership—and, with it, the burst in waterworks construction—to innovations in local public finance: Given the “exorbitant” expense of constructing large-scale water systems, neither private water companies nor any but the largest cities could “afford to build systems to serve entire municipal populations” until the development of municipal bond markets toward the end of the century (2005: 173).7

Background political, legal, and economic institutions undoubtedly affect the relative efficiency of governance arrangements.8 But neither changes in the institutional environment nor jurisdictional differences are able to account for some of the most conspicuous variations in governance during this period. First, the trend toward municipal ownership of waterworks clearly predated the political and financial developments Melosi and Cutler and Miller cite. As illustrated in Figure 1, an extrapolation of municipal ownership rates between 1800 and 1896 predicts the fraction of waterworks municipally owned in 1920 (68%) and 1924 (70%) exactly.9 Similarly, differences among states in laws and policies affecting public utility ownership, whatever their effect, were not decisive: Every state except one (Nevada, whose nine waterworks were all private) had both public and private works as of 1896 (Baker 1899: 22–5).

Figure 1.

Percent of Waterworks Municipally Owned.

Figure 1.

Percent of Waterworks Municipally Owned.

In fact, private waterworks were constructed at a prodigious rate both before and after the turn of the century, contrary to the intimation of Culter and Miller. As seen in Table 1, the number of private waterworks increased by almost a thousand between 1885 and 1899 and by more than 1600 between 1915 and 1925.10 Moreover, that construction was not limited just to smaller cities. On the contrary, Figure 2 shows the number of public and private waterworks built each decade for 373 waterworks serving the largest cities (with populations over 10,000) in 1890.11 Of the 131 works built in these cities between 1880 and 1896, 97, or almost three quarters, were private. By comparison, less than half of all new waterworks constructed during the same period were private (see Table 1). The fact that waterworks construction in the largest cities at the end of the century was disproportionately private is hardly what one would expect if inadequate access to capital by private companies was the critical determinant of ownership form.

Figure 2.

Number of Waterworks Built in Decade (ending date), by Original Ownership, Cities Larger than 10,000 in 1890 (n = 373).

Figure 2.

Number of Waterworks Built in Decade (ending date), by Original Ownership, Cities Larger than 10,000 in 1890 (n = 373).

Finally, a compelling resolution of the waterworks ownership puzzle must also explain why the attractiveness of municipal ownership was greater for water and sanitation systems than for other utilities and for some waterworks than others. All the factors identified by Melosi, for example, apply to other public utilities as well as water. Yet only water (and sanitation) systems experienced widespread municipalization. In principle, higher costs of waterworks compared to other public utilities could, as Cutler and Miller hypothesize, explain the discrepancy. But waterworks were not clearly far more expensive than other public utility systems. According to Bureau of Labor Statistics, total investment, including extensions and improvements, in the 3326 waterworks existing in 1899 was approximately $781 million, or an average investment of $235,000 per works; for gas works, the totals were $332 million invested in 965 works, or an average investment of $344,000 (US Bureau of Labor 1900: 12–4). Yet only 14 relatively small ($137,000 average investment) gas works were municipally owned (ibid.). As Troesken and Geddes (2003: 378) point out, if municipal ownership was a response to deficiencies in private relative to municipal capital markets, then public ownership should have increased with investment levels for all utilities, yet city size and public ownership of gas and electric utilities were negatively correlated.

One way in which water and sewer systems arguably differed from other public utilities was in their implications for public health and safety. As noted earlier, the growth of urban populations increased the need for centralized water and sewer systems to supply clean water and to remove wastes. In addition, waterworks were essential to assuring the availability of adequate water for fighting fires. But the existence of externalities does not by itself imply a need for public ownership:12 Cities in principle could have—and in practice did—contract for fresh and abundant water supplies from private suppliers.13 Consistent with this, Troesken (1999) found that private waterworks were actually more likely than public works to invest in water filtration systems and that the incidence of epidemics of waterborne diseases during the late 19th century was unrelated to ownership (see also Troesken and Geddes 2003: 376–7).14 Although it is true that a number of large fires were blamed on the failure of private water companies to supply adequate water or pressure for firefighting, public ownership was clearly no panacea. The great fires in New York (1845), Pittsburgh (1845), St. Louis (1849), Chicago (1871, 1874), Boston (1872), and Baltimore (1904) all occurred in cites with public waterworks. Even after the Great Chicago Fire of 1871, the Chicago Board of Aldermen refused to appropriate money for water system improvements requested by the Fire Department, contributing to Chicago's second great fire in 1874 (Rosen 1986: 3). The fact that cities blamed private companies for fire losses after the fact does not imply that private ownership necessarily increased the probability or size of fires, nor does it explain why the affected cities could not have contracted for greater fire protection, as they apparently had for water filtration.

Finally, differences in utility ownership might have arisen because of differences among utilities in the transaction costs of contracting and public ownership.15 As Troesken and Geddes (2003: 380) observe, standard transaction cost arguments could account for the relative preference for municipal ownership of waterworks if either (a) exceptionally large, idiosyncratic investments left private water suppliers particularly vulnerable to expropriation by local governments or (b) contracting for water supplies was peculiarly difficult or costly. Water and sanitation system assets were unquestionably location specific. But so were the major assets of other public utilities, which, like early waterworks, were willing to rely on long-term franchise contracts to protect their investments (see Priest 1993).16 By the same token, water and sanitation services were not conspicuously more complex or their cost and demand conditions notably more variable or uncertain than those of electricity, gas, or telephone services. Nor is it clear why contracting costs would have been higher for some communities than others or why contracting costs would have increased over time; if anything, one might expect contracting costs to have fallen as the legal system matured and the technology for measuring water quality improved.17

As evidence that contracting frictions did, nevertheless, contribute to waterworks municipalization, Troesken and Geddes (2003) offer an empirical analysis showing that private waterworks municipalized between 1897 and 1915 were significantly more likely than works that remained private to have experienced litigation with the municipality. Though their results are consistent with transaction cost reasoning, their analysis has several limitations. First, the case that legal frictions explained the relative preference for public ownership of waterworks is incomplete without evidence that contract disputes were more frequent or costly for waterworks than for other utilities. Litigation was in fact infrequent even for the waterworks in their study. Although 35% (258 of 726) of the waterworks they analyzed were municipalized, litigation over fire hydrants rentals, market entry, and franchise revocation—the three most common and empirically significant litigation categories in their analysis—each occurred in only about 3% of the population of works (2003: 389).18 And since no more than 16% of municipalized works experienced litigation in at least one of those three categories (ibid., 391), the vast majority—at least 217 of 258—of works that were municipalized were taken over without prior litigation in any of those categories. Finally, experience with litigation obviously cannot account for the much larger number of works that were publicly owned from the outset. What we really want to know is not whether frictions preceded municipalization but what conditions underlay and aggravated the frictions—actual or potential—that caused cities to prefer municipal ownership to franchise contracts.

Comparative Analysis

In this section, I examine 19th-century accounts, first, of the problems associated with governing public utilities generally and, second, of features peculiar to water and sewer systems that may have systematically altered the relative preference for municipal ownership and franchise contracting.

Nineteenth-Century Perspectives on Public Utility Governance

By the end of the 19th century, the ownership of public utilities had become a major public policy issue in the United States and the development and governance of these systems were of considerable interest to the public.19 Accordingly, the historical record contains numerous descriptions of and commentaries on the nature, problems, and organization of public utilities. Two writers in particular, Delos F. Wilcox and Moses N. Baker, provided detailed discussions and accounts of the history and experience of public utility governance in the United States at the turn of the century. Chief among the advantages they attributed to private ownership were (a) stronger incentives of private owners to operate utilities efficiently and (b) a reduced susceptibility to political influence and bureaucratic encumbrances. According to Baker, “The greatest argument for private ownership, and one that in a measure includes all others, is that under it every endeavor is made to conduct the works in the most prudent and economical manner, in order to insure the greatest possible return on the capital” (Baker 1899: 40). Similarly, Wilcox included at the top of a list of the advantages of private ownership, “(1) Greater freedom of initiative and adaptability to requirements of progress [and] (3) Incentive of profit to induce economy and efficiency in construction and operation” (1931: 6). Private utilities also benefited from their greater “[f]reedom from civil service red tape in the reward and promotion of efficient employees [and f]reedom from political interference and domination” (Wilcox 1931: 6).20 At a minimum, Baker maintained, “Private companies … certainly will not be accused of lowering rates unduly for political effect, as cities sometimes do” (1899: 42).

Nearly a century of experience with franchise contracting, however, had educated late 19th-century observers to the potential for conflict between the incentives of private owners and the interests of the communities they served and to the difficulty of resolving those conflicts contractually, particularly when circumstances warranted adjustments or extensions of the system (Wilcox 1931: 12):

Under a good system of supervision the original location and style of utility systems and works might be made to conform reasonably with the general city plan under a private system, but a great difficulty arises when it comes to subsequent readjustments that involve the expenditure of money without much direct benefit to the utilities.

Where rates paid to utilities were fixed by the contract, public officials chafed that private utilities were unwilling “to adopt the program of expansion to meet requirements based on community policy, independent of immediate or prospective profit” (Wilcox 1931: 6). Contracts that based rates on the utility's costs, on the other hand, often led private owners to overinvest, a bias typically associated, since Averch and Johnson (1962), with rate-of-return regulation (Wilcox 1931: 28):

[U]nder private ownership … the owners of a utility are urged by the most powerful economic motive imaginable to drive persistently, forcefully, adroitly, and effectively to swell the rate base by increasing the capital account. They have no motive to reduce the capitalization … but every motive to increase it, because increasing it adds to the base upon which their permissible profits are calculated.

To minimize conflicts, Baker advised utility contractors to exercise care in drafting franchise contracts (1899: 33):

[A]ll the conditions of a franchise should be plainly stated, so as to leave as little chance as possible for differences of interpretation. Ambiguity and indefiniteness are seeds from which spring annoying and expensive crops of litigation, neither city nor company being willing to acknowledge itself in the wrong.

At the same time, Baker acknowledged that such precision was unlikely to be achieved: “The obligations of the company are so various, and hinge on so many technicalities of law and engineering, that it is often no easy matter to determine them with exactness, much less to enforce them fully and impartially” (Baker 1899: 35). Cities as well as utilities could be expected to take advantage of whatever gaps remained in the contract. Perceptions by the public that utilities were earning excessive profits led to “demands for lower rates on every possible occasion” (Baker 1899: 33).21 Having failed to elicit concessions from the utilities, cities “will continue as heretofore threats or attempts to annul franchises on the ground of legal flaws” (Baker 1899: 33) or wield the powers of government to achieve much the same effect (Wilcox 1931: 12):

If under its franchise reservations or under its general police power, the city is in a position to act arbitrarily, there is a danger that it sometimes will do so …, result[ing] in uneconomical readjustments which never would be made if the city itself had to bear the cost, as it would under public ownership.

The “dependence [of private owners] upon public authorities for rights-of-way and franchise grants which terminate or may be revoked” constituted a major deterrent to private ownership (Wilcox 1931: 6).

Public ownership avoided many of these problems and could be expected to result in “less friction between consumers and purveyor” (Baker 1899: 46). But municipal control also forfeited the incentives of private owners to operate utilities efficiently and introduced the danger that “political rather than business methods may dominate” (Baker 1899: 37). There was, according to Baker, “no reason why a municipality cannot build and operate water-works as cheaply as a company”; if publicly owned utilities failed to do so, “it is due to political favoritism, corruption, incompetency, or negligence on the part of public officials, more largely than anything else” (Baker 1899: 47).22

Utility Attributes and Governance Costs

The preceding review suggests an understanding of the trade-offs inherent in the choice between franchise contracting and public ownership, at least among public utility experts of the period, that is remarkably congruent with modern, especially transaction cost, conceptions. The concerns they identify are, however, generic: The hazards of incomplete contracting and the inefficiencies of public production are obstacles to efficient procurement for all public utility services. The remainder of this section examines contemporaneous accounts of the distinctive features and problems of water and sanitation system operations and procurement for insights into the factors specific to those systems that might explain the discrepancy in how they were governed.

Administrative Simplicity.

Seeking to understand “the overwhelming preponderance of municipal ownership and operation of sewer systems” (1910: 452), Wilcox speculated that the preference for public ownership might be related to their “simplicity of administration” (453). According to Wilcox, sewer and water systems were distinctly uncomplicated: Whereas other major utilities either were “manufactured” (like electricity and, for much of the 19th century, gas) or used relatively sophisticated technology (telecommunications),23 sewer and water systems mainly provided access to existing natural resources, normally requiring nothing more than gravity to effect delivery once the infrastructure was in place. Sewerage systems were particularly easy to operate, involving—at least until the introduction of waste treatment facilities, which were uncommon for most of the 19th century—little more than allowing waste to flow through a system of pipes to be discharged downriver.24 As Wilcox described it, “Sewers are distinguished from other utilities in [the] respect … that once constructed they for the most part operate themselves. A sewerage system must be constructed on the gravity principle, and excepting in those cities whose topography is such that no natural outlet can be obtained, there is no pumping to be done.”25 Their operational simplicity, in turn, meant that sewer systems required comparatively little expertise and oversight: “In the absence of pumping or a purification plant, there is required, in what might be called a normal type of sewerage system, a very small administrative force for operating purposes. This simplicity of administration naturally is conducive to public ownership, and conversely is unattractive to private enterprise” (ibid., 453).

To a large degree, waterworks also tended to be self-operating: “Water is almost alone among public utilities in the fact that it is a natural rather than an artificial product… . If water can be brought from an elevation considerably higher than the highest point in the city which is to be supplied, the power of gravity may be depended upon for the distribution” (Wilcox 1910: 400–1). The exceptions, again, were where pumping or filtration was needed: Pumping required “expensive and powerful machinery” (ibid., 401) that had to be operated and maintained, while the need for filtration transformed water supply from delivery of a natural product into “a process of manufacture”: “Pure water must be ‘produced’ where it cannot be procured” (ibid., 400).

Wilcox was ultimately skeptical that administrative simplicity explained the disparity in attitudes toward public ownership, observing that “It is hard to believe that the almost entire absence of private operation of sewerage systems is the result merely … of simplicity of administration” (Wilcox 1910: 453). His conjecture nevertheless presents a theoretically sound and potentially testable explanation. As Williamson (1985: 131–56) has argued, although integrated firms or, in this case, publicly-owned utilities should be able to operate as efficiently as private ones simply by replicating what a private company would have done (cf. Baker in text at footnote 22 above)—including employing whatever experts or specialists are needed to operate the system—public administrators cannot simultaneously (a) retain the authority to intervene yet (b) provide the system's operators with high-powered incentives to run it at its potential. Having removed the incentive advantages of private ownership, public authorities will find that they must either monitor the effort of employee-operators more intensively or reconcile themselves to less efficient provision of utility services. How effective a public administrator will be at overseeing operators, and hence the efficiency of public ownership, will depend on, among other things, how complex the operations are and how knowledgeable the administrator is about the operations. In general, the more complex and unfamiliar the activities at issue, the less effective administration by a public authority is likely to be—the greater “the costs of organizing and the losses through mistakes,” in Coase's (1937: 395) words—and, correspondingly, the more important will be the judgment and incentives of experts.26

Applied to public utility governance, this administrative simplicity argument implies that public ownership should be more common for utilities whose operations are less complicated, require less technical expertise, and are less demanding of oversight and attention. To the extent that water and sanitation systems were self-operating or in other respects simpler to administer than other public utilities, the incentive and specialization disadvantages of public relative to private ownership may have been low enough to tilt the balance in favor of public ownership. By the same logic, the need for pumping and filtration would have made private ownership more attractive.

Relational Frictions.

Although conflicts and “friction” (Baker 1899: 46) between cities and private companies—especially “when it comes to subsequent readjustments that involve the expenditure of money” (Wilcox 1931: 12)—were common to all utilities, the literature suggests two features of water and sanitation systems—their pricing and, at least for water, the disruptiveness of repairs—that may have aggravated frictions when such adjustments were needed.

Utility Pricing and System Extensions.

Turn-of-the-century public utilities employed a range of sophisticated pricing arrangements. Electricity, for example, was subject to “very complex” metered rates: “The rates are almost always graduated according to purpose of use, and quantity used” and sometimes “on the time or constancy of use,” to which “service or minimum charges” might also be added (Wilcox 1931: 69), whereas for manufactured gas “the meter system may be considered universally in use” (ibid., 68). Several factors combined to limit the level and structure of prices for water and sewer services, however. First, unlike electricity, gas, and telephone service, all of which had substantial variable costs, use of water and sanitation services in most 19th-century communities was effectively nonrivalrous: Water and sanitation systems typically operated with substantial excess capacity, incurred no (except where pumping or filtering was required) production cost, and, because water availability was a central factor in settlers’ location decisions, often had negligible resource costs.27 Second, the positive externalities of centralized water and sewers, both individually and in relation to each other (see Section 2 above), created a public health interest that dictated against pricing that would significantly deter connection or use.28 Finally, the size of connection or access fees that utilities were able to charge for water and sewer service was also limited by the difficulty of excluding access and preventing resale of these services: Whereas electricity and gas are difficult, even dangerous, to transport and store, water could be carried and stored in cisterns or other containers or diverted with pipes or hoses with little cost or risk.29

A consequence of these constraints was that the fees that private water companies were able to charge consumers directly were typically inadequate to cover their costs.30 While serving as deputy commissioner of water supply for New York City in 1917, Wilcox estimated that revenues collected through water fees were less than half of what would be needed to generate a fair return on the city's investment (1917: 569).31 To make up this shortfall, communities found it necessary to subsidize waterworks from general revenues. In the case of franchised waterworks, transfers to private companies were most often effected through hydrant rates, which were set to assure that hydrant rentals “always covered the company's interest obligations incurred on the cost of construction” and “guaranteed the survival of a private company in its first years of operation” (Anderson 1984: 219). Consistent with the view that hydrant “rental rates probably reflect[ed] differences in the total construction cost of the systems” (ibid., 220) rather than the marginal costs of hydrant services, hydrant rental rates varied widely between cities. Of 587 private works that reported hydrant rates separately (as of 1888), hydrant rentals ranged from free (43 works) to $160 per hydrant (one works), with an average of $45 per hydrant, a modal charge of $50 per hydrant (89 works), and 45 works charging $100 or more per hydrant (Baker 1889: lxxxv; see also Anderson 1984: 219–20). Works that did not receive separate hydrant rentals typically received transfers from cities in other forms: “Nearly all the private works receive an indirect rental of some stated sum including water supply of city buildings, etc.” (Baker 1889: lxxxv).

The dependence of waterworks on what amounted to lump-sum transfers—given that hydrant numbers as well as rates were fixed by contract (Anderson 1984: 219)—to recover their capital costs had two potential implications for the efficiency of franchise contracting. First, inframarginal (service-invariant) payments forfeit the incentives for unilateral adjustments that service-based charges provide: A private utility whose receipts vary automatically with service levels has an incentive to undertake system improvements, extensions, and repairs that increase revenues more than costs without the need to reopen the agreement. By contrast, where service adjustments are not priced or are underpriced, every proposal for improvement or extension of the system becomes an occasion for potentially contentious and costly negotiation.32

Second, inframarginal payments are relatively attractive targets for rent seeking. Because service-contingent payments affect incentives both to consume and to supply services, attempts to appropriate rents that move such payments away from their efficient level reduce the surplus available to appropriate and, therefore, tend to be self-limiting.33 Changes in inframarginal payments, by contrast, are purely distributional: Because they do not distort substantive incentives to consume or produce, appropriations effected through modifications of inframarginal transfers leave quasi-rents (largely) intact. Thus, whereas a city that forced a significant reduction in service rates from a gas, electricity, or telephone company would likely suffer near-term service degradations (given their nontrivial marginal costs of production) as well as underinvestment over the longer term, a city that reneged on promised hydrant rentals would likely experience only the latter.

Conflicts between cities and water companies over hydrant rates were, in fact, common and intense. Disagreements over hydrant rentals “were the major source of friction between cities and private companies” (Anderson 1984: 220) and were the most frequent subject of litigation between cities and private water utilities reported by Troesken and Geddes (2003: 391). Jacobson's (1989) detailed description of San Francisco's experience is illustrative. To overcome the water company's resistance to new investment (used as a “bargaining tool,” according to Jacobson 1989: 18), the San Francisco Board of Supervisors agreed to payments of $2.50 per hydrant per month in 1882, increased to $5.00 in 1895, “in return for the company making investments in system extension and pipe enlargement for fire protection” (ibid., 18). Beginning in 1898, however, following investments by the company that achieved an increase in per-capita consumption of more than a third between 1880 and 1890 despite population growth of almost 30%, the city undertook a series of rate reductions—characterized as a “breach of trust” by the company—“cut[ting] hydrant payments … from the previous level of $5.00 per hydrant per month to a rate amounting [to] $1.75 per hydrant per month …, despite previous implicit agreements with Spring Valley to maintain existing charges in return for water company investments in system improvements” (ibid., 20).

Although no franchise contract was completely free from such hazards, the disparity between service-based fees and the costs of system extensions and improvements for water and sewer systems meant that extensions and enhancements were more likely for these than for other utilities to require monetary transfers from the city, payments that were both costly to negotiate and especially vulnerable to reneging once the promised investments were complete. The added frictions associated with these transfers may have been enough to make municipal ownership of water and sewer systems the more appealing option.34

Disruption and Contract Evasion.

If a party to a water franchise agreement did become dissatisfied with the terms of the agreement, the nature of water systems provided ample opportunities and a set of tactics with which to evade performance or force renegotiation. Over the intermediate to long term, companies could threaten to, or actually, withhold investments in system extensions and improvements, risking “insufficient protection to health and property,” as the San Francisco case illustrates.35 More immediately, water companies were able to cause significant disruptions to traffic and commerce, particularly in urban settings, by altering the diligence with which they conducted repairs and installations. The potential for disruption to residents and businesses was clearly a major concern to 19th-century observers. Baker identified “the right to dig up, occupy, and use streets constructed and maintained at public expense, thus interfering more or less with the use of the public highway for street traffic” as the “fundamental” element of water franchise contracts (1899: 31), while Wilcox observed, “Perhaps the most serious way … in which franchise holding companies tend to disturb public comfort is in the constant tearing up of the streets for the construction or repair of underground fixtures. It often seems astonishing that business can continue to be done in spite of these long-drawn-out and frequently-recurring interferences with the ordinary uses of the city highways” (1910: 122).36 An article in the New York Evening Post from the 1890s reveals the extent of “the damage caused by the frequent tearing up of streets for the making of repairs and new construction”:37

In Manhattan, for example, utility companies dug more than 59,000 transverse excavations for utility connections and repairs in the island's 391.5 miles of paved streets in one year, 1891, alone. Together with the many longitudinal excavations made, this averaged out to one excavation for every thirty-five feet of street. As city dwellers, businessmen, and public officials from the across the country often complained, this constant digging was making the urban pavement an unsightly and unnavigable ‘thing of shreds and patches’.

Particularly in congested areas, the ability of private water or sewer companies to disrupt daily life and commerce through unnecessary or dilatory excavations, and of cities to exaggerate claims of such, would have provided each a potentially effective, if costly, bargaining tactic in rate disputes. Combined with the greater occasion for such disputes described earlier, this added source of friction stood to make franchise contracting less conducive to private water and sewer provision than for other utilities.

Variations in Waterworks Ownership

Although public ownership of sewer systems was nearly universal from the outset, a great deal of variation in the ownership of waterworks persisted into the late 19th century and beyond. In this section, I use data on the characteristics of waterworks and the cities they served to assess the extent to which the administrative simplicity and relational frictions concerns identified above contributed to the choice of waterworks ownership in the 1890s.

Hypotheses

Table 3 summarizes some testable implications of the preceding discussion. Wilcox's administrative simplicity hypothesis, depicted in the first row, predicts that municipal ownership should have been less prevalent—denoted by the minus sign in parentheses in the second column—for waterworks that pumped or filtered water, activities that placed greater demands on the attention and expertise of operators.

Table 3.

Hypotheses and Variables

Hypothesis Attributes Variables 
1. Administrative simplicity Filtering, pumping (−) Filters, pumping (pump capacity) 
2. Relational frictions   
    a. Disruptions Service (net) density (+) Taps per mile of mains 
    b. Extensions Population distribution uniformity (−) Population (gross) density, mains per square mile 
3. Capital market imperfections/relationship-specific investment Investment size (+) Infrastructure, filters, pump capacity, reservoir capacity, mains, hydrants 
Hypothesis Attributes Variables 
1. Administrative simplicity Filtering, pumping (−) Filters, pumping (pump capacity) 
2. Relational frictions   
    a. Disruptions Service (net) density (+) Taps per mile of mains 
    b. Extensions Population distribution uniformity (−) Population (gross) density, mains per square mile 
3. Capital market imperfections/relationship-specific investment Investment size (+) Infrastructure, filters, pump capacity, reservoir capacity, mains, hydrants 

The second row lists the two factors identified in 19th-century accounts as particularly aggravating relational frictions in water franchise contracts: (a) the disruptiveness of system improvements and repairs and (b) the need to negotiate lump-sum transfers to secure system extensions. The first of these—the potential to disrupt commerce and traffic and the cost of such disruptions—would have been greater in congested areas than in more sparsely populated locations, indicated by the plus sign beside service density in the second column. The distribution of population within a city is also likely to affect the second factor: the incidence and contentiousness of conflicts over system extensions. Except where a city's population was uniformly distributed throughout its territory, the high up-front costs of installing water mains favored serving the most densely populated areas of a city first and extending service to less dense, outlying areas only as the population grew large enough to justify new investment.38 Occasions for system extensions were therefore more likely to arise in cities with unevenly distributed populations. Moreover, because the profitability of an extension depended on the density of customers along newly installed mains, the size of any transfer needed to induce a private company to make system extensions would have been inversely related to population density in the unserved area: the lower the density the larger the necessary transfer. Combined, these considerations suggest that relational frictions were likely to have been most severe, and therefore municipal ownership most prevalent, in cities with densely populated city centers (high “net density”) and sparsely populated outlying areas and, conversely, that private ownership would have been more common in cites with more uniformly distributed populations.39

Finally, the third row of Table 3 portrays the capital market imperfections and relationship-specific investment hypotheses, both of which predict that municipal ownership should be positively related to the size of investments in the system.

Variations between Works

Data and Variables.

My principal sources for data on waterworks and city characteristics are The Manual of American Water-Works, 1897 (Baker 1897) and Social Statistics of Cities, 1890 (US Bureau of the Census 1895). Baker (1897) contains entries for more than 3000 waterworks in operation in the United States in 1896 and reports information on ownership and a variety of waterworks characteristics (described below). The Social Statistics of Cities (Table 66: 68–77) also includes information, although more limited than Baker's, on waterworks serving US cities with populations of at least 10,000 in 1890. For the purpose of this study, I limit the analysis to waterworks with entries in Baker (1897) that served cities with 1890 populations exceeding 10,000. Three hundred and eight-one waterworks serving 346 cities fit this criterion, of which 373 entries had sufficient information to be included in the analysis. I then supplemented these data with information on city and waterworks characteristics from the Social Statistics of Cities and other sources.40

Tables 4 and 5 contain variable definitions and descriptive statistics for city and waterworks characteristics, respectively. As revealed in the tables, the data include several variables that bear on the hypotheses summarized in Table 3. Among the waterworks characteristics of interest, variables indicating whether a waterworks filtered or pumped water (rows 9 and 10 of Table 5) relate directly to the administrative simplicity hypothesis. The data also contain several measures of the size of waterworks investments (pumping capacity, reservoir capacity, miles of water mains, and number of hydrants) and of the existence of other large investments (infrastructure and filters) that can be used to evaluate the capital market imperfection and relationship-specific investment hypotheses.41

Table 4.

Variable Definitions and Descriptive Statistics, City Characteristics

  Mean SD Minimum Maximum Observed 
1. 1890 population City population in 1890 49,778 132,421 10,002 1,515,301 346 
2. 1900 population City population in 1900 67,201 184,470 7188 2,050,600 333 
3. Population change, 1890–1900 Pop1900 − Pop1890 16,069 52,217 −37,897 598,725 333 
4. Average annual population growth, 1890–1900 ln(Pop1900/Pop1890)/10 0.024 0.020 −0.064 0.104 333 
5. City area City area in square miles, 1890 11.5 16.3 0.5 160.6 315 
6. Population density Gross population density in 1890 = Pop1890/city area 5341 4656 199 37,562 315 
7. Dwellings, 1890 Number of dwellings in city in 1890 7651 15,869 1467 187,052 334 
8. Dwellings per square mile Dwellings/city area 908 680 35 4061 315 
9. Population per dwelling Pop1890/dwellings 5.81 1.31 4.31 18.52 334 
10. January average low temperature Thirty-year average minimum temperature 19.5 9.0 56 346 
11. Multiple works =1 if city served by more than one waterworks 0.08    346 
12. System number Number of waterworks serving city 1.10 0.45 346 
  Mean SD Minimum Maximum Observed 
1. 1890 population City population in 1890 49,778 132,421 10,002 1,515,301 346 
2. 1900 population City population in 1900 67,201 184,470 7188 2,050,600 333 
3. Population change, 1890–1900 Pop1900 − Pop1890 16,069 52,217 −37,897 598,725 333 
4. Average annual population growth, 1890–1900 ln(Pop1900/Pop1890)/10 0.024 0.020 −0.064 0.104 333 
5. City area City area in square miles, 1890 11.5 16.3 0.5 160.6 315 
6. Population density Gross population density in 1890 = Pop1890/city area 5341 4656 199 37,562 315 
7. Dwellings, 1890 Number of dwellings in city in 1890 7651 15,869 1467 187,052 334 
8. Dwellings per square mile Dwellings/city area 908 680 35 4061 315 
9. Population per dwelling Pop1890/dwellings 5.81 1.31 4.31 18.52 334 
10. January average low temperature Thirty-year average minimum temperature 19.5 9.0 56 346 
11. Multiple works =1 if city served by more than one waterworks 0.08    346 
12. System number Number of waterworks serving city 1.10 0.45 346 

Source: Variables 1, 2, 5, 7: US Bureau of the Census (1895, 1901); variable 10: US Department of Commerce (1991); variables 11–12: Baker (1897).

Table 5.

Variable Definitions and Descriptive Statistics, Waterworks Characteristics

  Mean SD Minimum Maximum Observed 
1. Ownership, 1896 =1 if works was publically owned in 1896 0.53    373 
2. Ownership, 1890 =1 if works was publically owned in 1890 0.50    359 
3. Ownership, original =1 if works was publically owned when first constructed 0.40    373 
4. Year of change in ownership Year in which waterworks ownership changed, if any 1874 20 1810 1896 60 
5. Year current works built Year in which the waterworks currently serving the city was built 1873 17 1796 1896 373 
6. Year first supplied Year in which city first served by waterworks 1868 23 1872 1896 373 
7. Source: surface =1 if city obtains water from lake, river, spring, or stream 0.82    373 
8. Source: wells =1 if city obtains water from wells 0.27    373 
9. Filters =1 if waterworks filters water 0.20    373 
10. Pumping =1 if waterworks pumps water 0.90    373 
11. Pumping capacity Pumping capacity (in millions of gallons) 12.90 31.23 380.79 364 
12. Infrastructure =1 if works has aqueduct, conduit, tunnel, tank, or standpipe 0.42    373 
13. Reservoir capacity Capacity of reservoir owned by waterworks (in millions of gallons) 535 4545 77,150 373 
14. Mains, 1890 Miles of water mains owned by the waterworks, 1890 58.0 102.4 930 281 
15. Mains, 1896 Miles of water mains owned by the waterworks, 1896 74.1 138.0 1615 349 
16. Hydrants 1890 Number of hydrants served by waterworks, 1890 456.7 965.7 23 8500 279 
17. Hydrants 1896 Number of hydrants served by waterworks, 1896 560.6 1321.9 16,466 359 
18. Taps, 1890 Number of taps (connections) served by the waterworks, 1890 5815 17,199 12 170,911 268 
19. Taps, 1896 Number of taps (connections) served by the waterworks, 1896 7315 20,244 95 233,792 318 
20. Meters 1890 Number of meters installed by waterworks, 1890 441.0 1634.7 19,600 282 
21. Meters 1896 Number of meters installed by waterworks, 1896 732.5 2391.1 32,329 367 
22. Percent metered, 1890 Meters90/Taps90 16.4 88 268 
23. Percent metered, 1896 Meters96/Taps96 14 22.3 100 318 
24. Taps per mile of mains, 1890 Taps90/Mains90 72.6 45.5 0.8 312.5 268 
25. Taps per mile of mains, 1896 Taps96/Mains96 75.6 34.3 17.3 228.6 303 
26. Mains per square mile, 1890 Mains90/city area 6.4 5.1 0.2 42.2 271 
27. Sewer system =1 if area served by waterworks had broad sewer coverage 0.74    373 
  Mean SD Minimum Maximum Observed 
1. Ownership, 1896 =1 if works was publically owned in 1896 0.53    373 
2. Ownership, 1890 =1 if works was publically owned in 1890 0.50    359 
3. Ownership, original =1 if works was publically owned when first constructed 0.40    373 
4. Year of change in ownership Year in which waterworks ownership changed, if any 1874 20 1810 1896 60 
5. Year current works built Year in which the waterworks currently serving the city was built 1873 17 1796 1896 373 
6. Year first supplied Year in which city first served by waterworks 1868 23 1872 1896 373 
7. Source: surface =1 if city obtains water from lake, river, spring, or stream 0.82    373 
8. Source: wells =1 if city obtains water from wells 0.27    373 
9. Filters =1 if waterworks filters water 0.20    373 
10. Pumping =1 if waterworks pumps water 0.90    373 
11. Pumping capacity Pumping capacity (in millions of gallons) 12.90 31.23 380.79 364 
12. Infrastructure =1 if works has aqueduct, conduit, tunnel, tank, or standpipe 0.42    373 
13. Reservoir capacity Capacity of reservoir owned by waterworks (in millions of gallons) 535 4545 77,150 373 
14. Mains, 1890 Miles of water mains owned by the waterworks, 1890 58.0 102.4 930 281 
15. Mains, 1896 Miles of water mains owned by the waterworks, 1896 74.1 138.0 1615 349 
16. Hydrants 1890 Number of hydrants served by waterworks, 1890 456.7 965.7 23 8500 279 
17. Hydrants 1896 Number of hydrants served by waterworks, 1896 560.6 1321.9 16,466 359 
18. Taps, 1890 Number of taps (connections) served by the waterworks, 1890 5815 17,199 12 170,911 268 
19. Taps, 1896 Number of taps (connections) served by the waterworks, 1896 7315 20,244 95 233,792 318 
20. Meters 1890 Number of meters installed by waterworks, 1890 441.0 1634.7 19,600 282 
21. Meters 1896 Number of meters installed by waterworks, 1896 732.5 2391.1 32,329 367 
22. Percent metered, 1890 Meters90/Taps90 16.4 88 268 
23. Percent metered, 1896 Meters96/Taps96 14 22.3 100 318 
24. Taps per mile of mains, 1890 Taps90/Mains90 72.6 45.5 0.8 312.5 268 
25. Taps per mile of mains, 1896 Taps96/Mains96 75.6 34.3 17.3 228.6 303 
26. Mains per square mile, 1890 Mains90/city area 6.4 5.1 0.2 42.2 271 
27. Sewer system =1 if area served by waterworks had broad sewer coverage 0.74    373 

Source: Variables 1, 3–13, 15, 17, 19, 23, 27: Baker (1897); variables 2, 14, 16, 18, 20, 22, 24: US Bureau of the Census (1895).

The discussion of relational frictions accompanying Table 3 suggested that frictions were likely to be greatest in cities with a combination of congested city centers and sparse peripheral areas: The disruptiveness of repairs would be greater in more densely populated areas, while the disputes over extensions would be more likely to arise in cities with low density outlying areas. Historical population statistics, however, are generally not fine enough to identify the distribution of population within a city. Moreover, cities with similar average or “gross” densities, measured as the ratio of population to total land area, can have very different population distributions. For example, although the population density of Newark, New Jersey, in 1890 was slightly greater than that of Philadelphia—16 versus 13 persons per acre—Philadelphia's most densely populated ward (covering 122 acres) had more than two and a half times the number of residents per acre—163 versus 63—as Newark's densest (covering 142 acres) (US Bureau of the Census 1895: 12).42

In the absence of data on intracity population distributions, I exploit the following relationships to test the relational frictions hypothesis. First, disruptiveness should be positively related to service density, measured as the number of connections (taps) per mile of water mains: The higher the number of connections per mile of mains, the lower the average distance between consumers and, by the reasoning given earlier, the greater the disruptive potential of excavations. Second, to the extent that connections are roughly proportional to population (within service areas), a higher population per square mile (gross density), holding service (net) density (taps per mile of mains) constant, implies a larger service area relative to total area: Average distance between consumers (taps per mile of mains) will remain constant as population density increases only if consumers are spread out more uniformly across a given area. Hence, higher taps per mile of mains and, holding taps per mile constant, lower population per square mile should favor municipal ownership. Alternatively, because higher population density implies higher service to total area, municipal ownership should be associated with lower miles of mains per square mile of area, again holding taps per mile of mains constant.43

The first row of Table 5 shows that 53% (199) of the 373 works in the data were publicly owned as of 1896, essentially the same as the proportion of all waterworks publicly owned as of that year (see Table 1). Given that the waterworks in this study serve the largest cities, this provides the first indication that city size was not the primary, or at least only, determinant of public ownership. Only 40% of these waterworks were municipally owned at the time they were constructed, however (Table 5, row 3): 60 works changed ownership between the time they were built and 1896, 55 from private to public, and 5 from public to private, the latter including one works in New Orleans that switched from private to public in 1868 and back to private in 1878.44

Table 6 shows differences in mean city characteristics for cities served by private, public, and multiple works. Cities served by public works were larger on average both in population and in land area (although differences in the median population are considerably smaller), and had slightly higher average population density and population per dwelling, than cities served by private works. The largest differences, however, are between cities served by single and multiple works, the latter being substantially larger and denser than the single-works cities. Multiple-works cities are also located in significantly warmer climates on average than single-works cities.

Table 6.

Mean City Characteristics by 1896 Ownership

Ownership in 1896 Private Public Multiple worksa 
Total observations 139 181 26 
1. 1890 population 23,926 (median: 14,270) 55,471 (median: 21,652) 148,358 (median: 37,741) 
2. 1900 population 31,128 (median: 18,437) 73,535 (median: 28,021) 213,070 (median: 51,721) 
3. Population change, 1890–1900 6723 16,838 59,998 
4. Average annual population growth, 1890–1900 0.023 0.023 0.030 
5. City area 8.8 12.5 17.8 
6. Population density 4824 5171 9288 
7. Dwellings 4329 8382 19,689 
8. Dwellings per square mile 876.9 869.8 1352.9 
9. Population per dwelling 5.45 5.96 6.53 
10. January average low temperature 19.9 18.5 23.9 
11. Number of waterworks 2.35 
Ownership in 1896 Private Public Multiple worksa 
Total observations 139 181 26 
1. 1890 population 23,926 (median: 14,270) 55,471 (median: 21,652) 148,358 (median: 37,741) 
2. 1900 population 31,128 (median: 18,437) 73,535 (median: 28,021) 213,070 (median: 51,721) 
3. Population change, 1890–1900 6723 16,838 59,998 
4. Average annual population growth, 1890–1900 0.023 0.023 0.030 
5. City area 8.8 12.5 17.8 
6. Population density 4824 5171 9288 
7. Dwellings 4329 8382 19,689 
8. Dwellings per square mile 876.9 869.8 1352.9 
9. Population per dwelling 5.45 5.96 6.53 
10. January average low temperature 19.9 18.5 23.9 
11. Number of waterworks 2.35 
a

Twenty-one cities had two works, four had three works, and one (Brooklyn) had seven works.

Finally, Table 7 reports mean waterworks characteristics for waterworks that were always public or private and for works that converted from private to public and from public to private. Somewhat surprisingly given the general trend toward public ownership, waterworks that began and remained privately owned were built (a statistically significant) four-and-a-third years later on average than always-public works.45 The table also indicates that always-private waterworks were slightly more likely to obtain water from wells and less likely to obtain water from surface sources such as lakes and rivers. Always-private works were also less likely than the other ownership categories to serve locations with sewer systems. Consistent with the administrative simplicity hypothesis, and with Troesken (1999), private waterworks were more likely than public waterworks to operate filter systems. Private works, however, were no more likely to pump water, and had substantially less pumping capacity, than public works, contrary to what would be expected if pumping water made operating water systems more complex.

Table 7.

Mean Waterworks Characteristics by 1896 Ownership

 Private to public Always public Always private Public to privatea 
Total observations 55 144 169 
1. Year current works built 1860 1873 1877 1866 
2. Year first supplied 1846 1871 1875 1864 
3. Year of change in ownership 1873 — — 1884 
4. Source: surface (proportion) 0.89 0.84 0.77 0.80 
5. Source: wells (proportion) 0.25 0.24 0.30 0.20 
6. Sewers 0.89 0.81 0.63 1.00 
7. Filters 0.05 0.16 0.27 0.20 
8. Pumping 0.91 0.89 0.91 0.80 
9. Pumping capacity (millions of gallons) 26.88 13.30 7.89 16.70 
10. Infrastructure 0.49 0.38 0.44 0.60 
11. Reservoir capacity (millions of gallons) 1260 140 641 383 
12. Mains, 1890 120.4 52.9 35.5 38.3 
13. Mains, 1896 179.6 71.4 45.3 66.6 
14. Taps, 1890 14,424 5536 2180 4204b 
15. Taps, 1896 19,982 6908 2998 2878 
16. Meters 1890 1004 333 312 58 
17. Meters 1896 1664 761 409 363 
18. Hydrants 1890 1028 416 237 451 
19. Hydrants 1896 1380 566 283 646 
20. Percent metered, 1890 
21. Percent metered, 1896 13 15 13 16 
22. Taps per mile of mains, 1890 87.4 79.4 57.5 89.2b 
23. Taps per mile of mains, 1896 92.0 79.9 60.4 45.0 
24. Mains per square mile, 1890 7.4 5.8 6.8 2.5 
 Private to public Always public Always private Public to privatea 
Total observations 55 144 169 
1. Year current works built 1860 1873 1877 1866 
2. Year first supplied 1846 1871 1875 1864 
3. Year of change in ownership 1873 — — 1884 
4. Source: surface (proportion) 0.89 0.84 0.77 0.80 
5. Source: wells (proportion) 0.25 0.24 0.30 0.20 
6. Sewers 0.89 0.81 0.63 1.00 
7. Filters 0.05 0.16 0.27 0.20 
8. Pumping 0.91 0.89 0.91 0.80 
9. Pumping capacity (millions of gallons) 26.88 13.30 7.89 16.70 
10. Infrastructure 0.49 0.38 0.44 0.60 
11. Reservoir capacity (millions of gallons) 1260 140 641 383 
12. Mains, 1890 120.4 52.9 35.5 38.3 
13. Mains, 1896 179.6 71.4 45.3 66.6 
14. Taps, 1890 14,424 5536 2180 4204b 
15. Taps, 1896 19,982 6908 2998 2878 
16. Meters 1890 1004 333 312 58 
17. Meters 1896 1664 761 409 363 
18. Hydrants 1890 1028 416 237 451 
19. Hydrants 1896 1380 566 283 646 
20. Percent metered, 1890 
21. Percent metered, 1896 13 15 13 16 
22. Taps per mile of mains, 1890 87.4 79.4 57.5 89.2b 
23. Taps per mile of mains, 1896 92.0 79.9 60.4 45.0 
24. Mains per square mile, 1890 7.4 5.8 6.8 2.5 
a

Includes one works in New Orleans, which switched from private to public in 1868 and back to private in 1878.

b

The high values for public-to-private taps and taps per mile of mains for 1890 reflect a reported value of taps for New Orleans in 1890 of 10,311, compared to only 4800 taps reported for 1896.

The next six characteristics—infrastructure, reservoir capacity, miles of mains, and the number of taps, meters, and hydrants—measure the size of investments. By all measures, works that changed from private to public ownership were larger than other works. Always private works were generally smaller as measured by mains, taps, meters, and hydrants but had substantially greater reservoir capacity and were slightly more likely to have made large infrastructure investments than works that had always been public. Finally, the last three rows report the number of taps (connections) per mile of water mains (for 1890 and 1896) and miles of mains per square mile of city area (for 1890). Public works exhibited greater service density (taps per mile) on average than private works but showed no clear pattern for mains per square mile.

Probit Estimates.

Table 8 reports results of various specifications of probit regressions relating municipal ownership to city and waterworks characteristics.46 Observations are waterworks, and the dependent variable takes the value of one if the waterworks was municipally owned. The first column shows results including the variables associated in Table 4 with administrative simplicity (filters and pump capacity) and investment size (infrastructure, reservoir capacity, mains, and hydrants, as well as filtering and pump capacity, which also represent investments) plus gross population density (population per square mile) and the number of systems serving the city in which the waterworks is located.47 Consistent with the administrative simplicity hypothesis, the coefficient on filters is negative and significant, implying that waterworks that operate filters were 20 percentage points more likely to be privately owned (marginal probability at the means of other variables). The coefficient on pump capacity, although negative, as predicted by the administrative simplicity hypothesis, is insignificant. Of the six variables measuring investment size, only the coefficients on filters, reservoir capacity, and hydrants are significant, and two of the three—on filters and reservoir capacity—are negative, contrary to the capital market imperfections and relationship-specific investment hypotheses. Only the number of hydrants is positively associated with municipal ownership: The addition of 100 hydrants increases the probability of municipal ownership by approximately 4 percentage points (at the means of all variables).

Table 8.

Probit Estimations of Waterworks Ownership (Dependent Variable = 1 if Municipally Owned)

 (1) Ownership 1896 (2) Ownership 1896 (3) Ownership 1896 (4) Ownership 1890 (5) Ownership 1890 (6) Ownership 1890 (7) Ownership 1890 (8) Ownership 1890 
Filters −0.513 (2.54)* −0.456 (2.67)** −0.363 (1.57) −0.408 (1.92) −0.390 (1.86) −0.353 (1.32) −0.350 (1.31) −0.376 (1.39) 
Pump capacity (millions) −0.015 (1.11) −0.018 (1.32) −0.082 (1.35) −0.003 (0.34) −0.0002 (0.03) −0.010 (1.25) −0.010 (1.28) −0.009 (0.96) 
Population per square mile (‘000s) −0.017 (0.65)  −0.082 (2.88)** −0.056 (2.37)*  −0.105 (3.24)**  −0.104 (3.18)** 
Mains per square mile     −0.039 (2.43)*  −0.068 (3.00)**  
Taps per mile  0.012 (2.74)** 0.018 (2.95)** 0.010 (3.19)** 0.009 (2.82)** 0.011 (2.75)** 0.008 (2.22)* 0.010 (2.70)** 
Infrastructure −0.043 (0.27) −0.131 (0.96) 0.0043 (0.03) −0.040 (0.28) −0.089 (0.63) 0.088 (0.54) 0.033 (0.21) 0.101 (0.61) 
Reservoir capacity (millions) −0.0002 (2.58)** −0.0002 (2.28)* −0.0002 (2.25)* −0.00005 (1.83) −0.0001 (1.50) −0.0002 (2.27)* −0.0002 (2.13)* −0.0002 (2.28)* 
Mains −0.00003 (0.08) 0.0004 (0.20) 0.0008 (0.29) 0.0001 (0.29) 0.003 (1.09) 0.002 (0.57) 0.005 (1.34) 0.001 (0.17) 
Hydrants 0.0012 (2.54)* 0.0010 (2.39)* 0.0010 (2.41)* 0.0003 (0.89) 0.0001 (0.22) 0.00005 (0.11) −0.00004 (0.08) 0.0003 (0.49) 
System number −0.291 (0.99)  −0.317 (1.16) −0.244 (1.73) −0.334 (2.41)* −0.478 (2.09)* −0.510 (2.15)* −0.520 (1.91) 
January average low temperature      −0.015 (1.40) −0.017 (1.64) −0.015 (1.34) 
1890 Population (‘000s)      0.007 (1.99)* 0.004 (1.15) 0.003 (0.44) 
City area      −0.019 (1.98)* −0.016 (1.46) −0.018 (1.80) 
Average annual growth rate, 1890–1900      2.215 (0.49) 2.810 (0.66)  
Population change, 1890–1900 (‘000s)        0.012 (1.83) 
Sewers      0.735 (3.18)** 0.723 (3.11)** 0.754 (3.20)** 
Wells      −0.103 (0.52) −0.050 (0.25) −0.130 (0.64) 
Age      0.047 (2.47)* 0.049 (2.52)* 0.046 (2.37)* 
Age squared      −0.001 (2.20)* −0.001 (2.16)* −0.001 (2.04)* 
Constant 0.430 (1.07) −0.689 (2.54) −0.395 (0.95) −0.036 (0.14) 0.130 (0.46) −0.295 (0.64) −0.199 (0.41) −0.196 (0.43) 
Observations 288 290 257 252 252 248 248 248 
Number of states 41 40 39 40 40 40 40 40 
Wald χ2 (d.f.) 19.81 (8) 56.54 (7) 22.84 (9) 39.04 (9) 31.76 (9) 99.44 (17) 109.53 (17) 128.56 (17) 
Log likelihood −177.2351 −168.3391 −136.7307 −154.9921 −155.69607 −137.2648 −138.9587 −136.3225 
Pseudo-R2 0.09 0.14 0.14 0.10 0.09 0.19 0.18 0.20 
 (1) Ownership 1896 (2) Ownership 1896 (3) Ownership 1896 (4) Ownership 1890 (5) Ownership 1890 (6) Ownership 1890 (7) Ownership 1890 (8) Ownership 1890 
Filters −0.513 (2.54)* −0.456 (2.67)** −0.363 (1.57) −0.408 (1.92) −0.390 (1.86) −0.353 (1.32) −0.350 (1.31) −0.376 (1.39) 
Pump capacity (millions) −0.015 (1.11) −0.018 (1.32) −0.082 (1.35) −0.003 (0.34) −0.0002 (0.03) −0.010 (1.25) −0.010 (1.28) −0.009 (0.96) 
Population per square mile (‘000s) −0.017 (0.65)  −0.082 (2.88)** −0.056 (2.37)*  −0.105 (3.24)**  −0.104 (3.18)** 
Mains per square mile     −0.039 (2.43)*  −0.068 (3.00)**  
Taps per mile  0.012 (2.74)** 0.018 (2.95)** 0.010 (3.19)** 0.009 (2.82)** 0.011 (2.75)** 0.008 (2.22)* 0.010 (2.70)** 
Infrastructure −0.043 (0.27) −0.131 (0.96) 0.0043 (0.03) −0.040 (0.28) −0.089 (0.63) 0.088 (0.54) 0.033 (0.21) 0.101 (0.61) 
Reservoir capacity (millions) −0.0002 (2.58)** −0.0002 (2.28)* −0.0002 (2.25)* −0.00005 (1.83) −0.0001 (1.50) −0.0002 (2.27)* −0.0002 (2.13)* −0.0002 (2.28)* 
Mains −0.00003 (0.08) 0.0004 (0.20) 0.0008 (0.29) 0.0001 (0.29) 0.003 (1.09) 0.002 (0.57) 0.005 (1.34) 0.001 (0.17) 
Hydrants 0.0012 (2.54)* 0.0010 (2.39)* 0.0010 (2.41)* 0.0003 (0.89) 0.0001 (0.22) 0.00005 (0.11) −0.00004 (0.08) 0.0003 (0.49) 
System number −0.291 (0.99)  −0.317 (1.16) −0.244 (1.73) −0.334 (2.41)* −0.478 (2.09)* −0.510 (2.15)* −0.520 (1.91) 
January average low temperature      −0.015 (1.40) −0.017 (1.64) −0.015 (1.34) 
1890 Population (‘000s)      0.007 (1.99)* 0.004 (1.15) 0.003 (0.44) 
City area      −0.019 (1.98)* −0.016 (1.46) −0.018 (1.80) 
Average annual growth rate, 1890–1900      2.215 (0.49) 2.810 (0.66)  
Population change, 1890–1900 (‘000s)        0.012 (1.83) 
Sewers      0.735 (3.18)** 0.723 (3.11)** 0.754 (3.20)** 
Wells      −0.103 (0.52) −0.050 (0.25) −0.130 (0.64) 
Age      0.047 (2.47)* 0.049 (2.52)* 0.046 (2.37)* 
Age squared      −0.001 (2.20)* −0.001 (2.16)* −0.001 (2.04)* 
Constant 0.430 (1.07) −0.689 (2.54) −0.395 (0.95) −0.036 (0.14) 0.130 (0.46) −0.295 (0.64) −0.199 (0.41) −0.196 (0.43) 
Observations 288 290 257 252 252 248 248 248 
Number of states 41 40 39 40 40 40 40 40 
Wald χ2 (d.f.) 19.81 (8) 56.54 (7) 22.84 (9) 39.04 (9) 31.76 (9) 99.44 (17) 109.53 (17) 128.56 (17) 
Log likelihood −177.2351 −168.3391 −136.7307 −154.9921 −155.69607 −137.2648 −138.9587 −136.3225 
Pseudo-R2 0.09 0.14 0.14 0.10 0.09 0.19 0.18 0.20 

Robust z statistics (absolute value) clustered on state in parentheses.

*Significant at 5%.

**Significant at 1%.

The lack of a significant effect of (gross) population density when included by itself (i.e., without controlling for service [net] density) is not surprising given the ambiguous relationship between population density and intracity population distribution: Higher population per square mile can result from greater density either in an already densely populated part of a city, which would, by hypothesis, increase relational frictions, or in less populated areas, which would tend to reduce frictions (by lessening conflicts over extensions). The results in column (2) of Table 8 replace gross density with net (service) density, measured as taps (connections) per mile of mains. The coefficient on taps per mile of mains is significantly positive, with an estimated marginal effect of 0.005: 10 additional taps per mile increases the probability of municipal ownership by 5 percentage points. Other results change only marginally compared to the specification in column (1).48

Column (3) reports results including both gross and net density. The coefficients on both taps per mile of mains and population per square mile are significant and their signs consistent with the relational frictions hypothesis. At the means, increasing taps per mile by 10 raises the probability of municipal ownership by 6 percentage points, while an increase of 1000 in population per square mile—holding taps per mile constant—reduces the probability of municipal ownership by 3 percentage points. Other results remain similar except that the coefficient on filters is no longer significant.

With the exception of population per square mile, all the variables used in the estimations reported in columns (1) through (3), including the dependent variable (ownership), are 1896 values (from Baker 1897). Because data on city characteristics are from the 1890 Census, however, using 1890 values (where available) for specifications relying heavily on city characteristics seems more appropriate. For purposes of comparing the effects of using 1890 values, I reestimated the specification in column (3) substituting 1890 values (from the Social Statistics of Cities, 1890) for ownership, taps per mile, mains, and hydrants. As seen in column (4), the coefficients on taps per mile and population per square mile retain their signs and significance. The coefficients on filters and reservoir capacity (both of which are 1896 values) drop just below the 0.05 level in significance, while the coefficient on hydrants, previously significant, becomes insignificant.

The specification in column (5) repeats that of column (4) but substitutes miles of mains per square mile, as a measure of the breadth of water service coverage, for population per square mile. (See footnote 43 and accompanying text.) The results are again similar. In particular, taps per mile of mains remains positive and significant, while more mains per square mile is negative and significant, indicating that waterworks covering more of a city's area are (holding service density constant) more likely to be privately owned. (Marginal effect: 0.015.)

The last three columns of Table 8 report results augmenting the basic model with a set of variables controlling for climate, population, city area, population growth (rate and absolute change), the existence of sewers, whether the waterworks obtained water from wells (as opposed to surface sources), and when the works was constructed (age and age squared). Of these added controls, only the coefficients on sewers and age are significant in all three specifications: Waterworks in cities with sewer systems are significantly more likely—by 29 percentage points at the means—to be municipally owned. The quadratic in age implies that, other factors held constant, the oldest and newest waterworks were least likely to be municipal, with works built around 1860 having the highest probability of municipal ownership. The only other controls showing significant effects are population and city area, but only in one of the three specifications (column (6)).

Of the primary variables, the coefficients on taps per mile and on population per square mile (in columns (6) and (8)) and mains per square mile (in column (7)) remain significant after addition of these controls, while reservoir capacity again becomes significant but, as in columns (1)–(3), is negative, contrary to the capital markets and specific investment hypotheses. Finally, although significant only at the 0.10 level, absolute population change (column (8)) is positively related to municipal ownership. To the extent cities with higher growth are more likely to require new investments or improvements in existing structures, this result is also consistent with the relational frictions hypothesis. The rate of population growth (columns (6) and (7)) shows no significant effect on ownership, however.

Reduced-Form and IV Estimates.

The preceding identifies correlations between ownership and city and waterworks characteristics but leaves open the possibility that those correlations are the result rather than the cause of ownership form. To account for the possible endogeneity of waterworks characteristics, I estimated reduced-form and instrumental variable versions of the model. For the reduced-form specification, I excluded continuous waterworks characteristics—pump and reservoir capacity, miles of mains, number of hydrants, and taps per mile of mains—but retained categorical variables for wells, infrastructure, filters, and pumping (in place of pump capacity) on the grounds that water source and the need for discrete investments on structures like tunnels and standpipes and for filtering and pumping (as distinct from pumping capacity) are likely to be determined by natural attributes such as a city's location and topography and the quality of its water resources. In place of taps per mile as a measure of net density, I substituted population per dwelling on the assumption that relatively congested cities are more likely to have crowded living quarters.49 Probit results from the reduced-form estimation are reported in Figure 3. The coefficients on three variables of interest—filters, population per dwelling, and population per square mile—are all significant and have signs that correspond to previous results. Of the other variables, population, sewers, and age (and its square) are also significant, as they had been in some specifications in Table 8.

Figure 3.

Reduced-Form Probit Estimates of Waterworks Ownership (Dependent variable = 1 if municipally owned in 1890; absolute value of t statistics in parentheses; *significant at 5%; **significant at 1%).

Figure 3.

Reduced-Form Probit Estimates of Waterworks Ownership (Dependent variable = 1 if municipally owned in 1890; absolute value of t statistics in parentheses; *significant at 5%; **significant at 1%).

Table 9 reports results from several specifications using instrumental variables for service density (taps per mile), mains per square mile, and two measures of investment size: miles of mains and number of hydrants.50 Panel A contains the “second-stage” probit coefficients on the probability of municipal ownership, and Panel B reports the corresponding estimates from the “first-stage” regressions of the instrumented variables.51 The identifying instruments in the first-stage regressions are the first 10 variables in Panel B for specification (1) (which omits population per square mile from the second-stage probit equation) and the first 9 variables for specifications (2)–(7). Overall, the first-stage regressions predict taps per mile, mains per square mile, miles of mains, and the number of hydrants fairly well by cross-section standards (as measured by their R2s). Although few of the coefficients on the identifying instruments in the taps-per-mile regressions (the first three columns of Panel B) are significant at conventional levels, the variables are highly significant as a group, as indicated by the corresponding χ2 and F tests that their coefficients are jointly zero.52 The Wald test of exogeneity, reported at the bottom of Panel A, indicates that the null hypothesis of exogeneity of the instrumented variables is rejected only for the specification in columns (2) and (marginally) (3).53

Table 9.

Instrumental Variable Estimates

Panel A. Second-Stage Probit (Dependent Variable = 1 if Municipally Owned in 1890) 
 (1) MLE (2) MLE (3) Two step (4) Two step (5) Two step (6) Two step (7) Two step 
Filters −0.432 (2.02)* −0.251 (1.15) −0.324 (1.40) −0.346 (1.51) −0.229 (1.42) −0.362 (1.53) −0.340 (1.33) 
Pumping −0.044 (0.15) −0.097 (0.33) −0.076 (0.23) −0.170 (0.51) −0.076 (0.24) −0.080 (0.25) −0.180 (0.51) 
Population per square mile (‘000s)  −0.094 (3.68)** −0.099 (2.91)**  −0.097 (2.51)* −0.093 (2.59)** −0.122 (2.60)** 
∧Taps per mile 0.007 (2.00)* 0.020 (4.86)** 0.019 (3.52)** 0.013 (2.72)** 0.017 (2.29)* 0.016 (2.37)* 0.022 (2.44)* 
∧Mains per square mile    −0.062 (1.87)    
∧Mains     0.001 (0.33)  −0.011 (1.04) 
∧Hydrants      0.0002 (0.78) 0.001 (1.16) 
January average low temperature −0.022 (2.42)* −0.018 (1.97)* −0.018 (1.84) −0.017 (1.82) −0.019 (1.88) −0.019 (1.90) −0.010 (0.76) 
Sewers 0.597 (2.79)** 0.465 (2.08)* 0.553 (2.41)* 0.579 (2.54)* 0.549 (2.40)* 0.540 (2.32)* 0.559 (2.24)* 
Constant −0.201 (0.52) −0.666 (1.76) −0.598 (1.39) −0.255 (0.63) −0.524 (1.10) −0.467 (1.02) −0.669 (1.27) 
Observations 258 258 258 258 258 256 256 
χ2 (d.f.) 26.39 (5) 55.99 (6) 30.78 (6) 26.21 (6) 31.40 (7) 32.04 (7) 29.83 (8) 
Log likelihood −1411.856 −1406.518      
Wald test of exogeneity (d.f.) 0.02 (1) (p = 0.88) 5.59 (1) (p = 0.02) 3.45 (1) (p = 0.06) 1.89 (2) (p = 0.39) 2.83 (2) (p = 0.24) 3.83 (2) (p = 0.15) 4.97 (3) (p = 0.17) 
MLE, maximum likelihood estimation.   ^indicates instrumented variable. 
Panel A. Second-Stage Probit (Dependent Variable = 1 if Municipally Owned in 1890) 
 (1) MLE (2) MLE (3) Two step (4) Two step (5) Two step (6) Two step (7) Two step 
Filters −0.432 (2.02)* −0.251 (1.15) −0.324 (1.40) −0.346 (1.51) −0.229 (1.42) −0.362 (1.53) −0.340 (1.33) 
Pumping −0.044 (0.15) −0.097 (0.33) −0.076 (0.23) −0.170 (0.51) −0.076 (0.24) −0.080 (0.25) −0.180 (0.51) 
Population per square mile (‘000s)  −0.094 (3.68)** −0.099 (2.91)**  −0.097 (2.51)* −0.093 (2.59)** −0.122 (2.60)** 
∧Taps per mile 0.007 (2.00)* 0.020 (4.86)** 0.019 (3.52)** 0.013 (2.72)** 0.017 (2.29)* 0.016 (2.37)* 0.022 (2.44)* 
∧Mains per square mile    −0.062 (1.87)    
∧Mains     0.001 (0.33)  −0.011 (1.04) 
∧Hydrants      0.0002 (0.78) 0.001 (1.16) 
January average low temperature −0.022 (2.42)* −0.018 (1.97)* −0.018 (1.84) −0.017 (1.82) −0.019 (1.88) −0.019 (1.90) −0.010 (0.76) 
Sewers 0.597 (2.79)** 0.465 (2.08)* 0.553 (2.41)* 0.579 (2.54)* 0.549 (2.40)* 0.540 (2.32)* 0.559 (2.24)* 
Constant −0.201 (0.52) −0.666 (1.76) −0.598 (1.39) −0.255 (0.63) −0.524 (1.10) −0.467 (1.02) −0.669 (1.27) 
Observations 258 258 258 258 258 256 256 
χ2 (d.f.) 26.39 (5) 55.99 (6) 30.78 (6) 26.21 (6) 31.40 (7) 32.04 (7) 29.83 (8) 
Log likelihood −1411.856 −1406.518      
Wald test of exogeneity (d.f.) 0.02 (1) (p = 0.88) 5.59 (1) (p = 0.02) 3.45 (1) (p = 0.06) 1.89 (2) (p = 0.39) 2.83 (2) (p = 0.24) 3.83 (2) (p = 0.15) 4.97 (3) (p = 0.17) 
MLE, maximum likelihood estimation.   ^indicates instrumented variable. 
Panel B. First-Stages Estimates of the Instruments 
 (1) Taps per mile (2) Taps per mile (3–5) Taps per mile (4) Mains per square mile (5) Mains (6–7) Hydrants 
1. 1890 population (‘000s) −0.034 (1.16) −0.009 (0.30) −0.033 (1.12) 0.001 (0.38) 0.540 (15.69)** 6.011 (23.48)** 
2. City area 0.510 (2.59)** 0.341 (1.74) 0.504 (2.54)* −0.029 (1.51) 1.218 (5.22)** 3.009 (1.73) 
3. Dwellings per square mile 0.007 (0.57) 0.013 (0.71) 0.007 (0.59) 0.007 (5.72)** 0.057 (3.88)** 0.242 (2.22)* 
4. Population per dwelling −0.064 (0.02) 2.033 (0.71) 0.428 (0.01) 0.448 (1.54) 2.148 (0.60) 47.705 (1.78) 
5. Percent of state population urban 0.148 (1.36) 0.227 (2.27)* 0.153 (1.43) 0.028 (2.70)** 0.519 (4.11)** 2.914 (3.11)** 
6. Wells −0.946 (0.19) −0.476 (0.10) −0.930 (0.18) 0.173 (0.36) −4.980 (0.83) −5.762 (0.13) 
7. System number 10.115 (2.22)* 8.443 (1.89) 10.097 (2.15)* −0.134 (0.30) −1.289 (−0.23) −83.500 (2.00)** 
8. Age 0.883 (2.55)* 0.954 (2.95)** 0.893 (2.54)* 0.173 (0.36) 0.928 (2.25)* −9.764 (3.16)** 
9. Age squared −0.001 (0.22) −0.003 (0.68) −0.001 (0.24) −0.001 (0.04) −0.021 (4.06)** −0.200 (5.18)** 
10. Population per square mile (‘000s) 3.408 (1.55) 1.947 (0.95) 3.314 (1.52) −0.184 (0.88) −7.780 (3.04)** −38.027 (1.99)* 
11. Filters −8.815 (1.65) −8.908 (1.67) −8.811 (1.60) 0.160 (0.31) 4.956 (0.77) 69.779 (1.45) 
12. Pumping 8.770 (1.19) 10.291 (1.40) 8.880 (1.17) −1.189 (1.64) 6.572 (0.74) 86.328 (1.30) 
13. January average low temperature −0.115 (0.51) −0.048 (0.21) −0.111 (0.48) 0.028 (1.27) 0.586 (2.17)* −2.183 (1.08) 
14. Sewers 5.259 (1.03) 5.356 (1.04) 5.266 (1.00) 0.216 (0.43) 10.600 (1.71) 85.637 (1.83) 
Constant 3.194 (0.15) −10.792 (0.55) 2.287 (0.11) −1.846 (0.92) −51.450 (2.09)* −303.348 (1.64) 
Observations 258 258 258 258 258 256 
R2   0.49 0.65 0.87 0.93 
Fa   16.40 (14, 243) 32.16 (14, 243) 121.42 (14, 243) 214.03 (14, 241) 
χ2 test for variables 1–10 = 0a 214.59 (10)      
χ2 test for variables 1–9 = 0a  81.12 (9)     
F test for variables 1–9 = 0a   8.84 (9, 243) 9.91 (9, 243) 154.07 (9, 243) 258.32 (9, 241) 
Panel B. First-Stages Estimates of the Instruments 
 (1) Taps per mile (2) Taps per mile (3–5) Taps per mile (4) Mains per square mile (5) Mains (6–7) Hydrants 
1. 1890 population (‘000s) −0.034 (1.16) −0.009 (0.30) −0.033 (1.12) 0.001 (0.38) 0.540 (15.69)** 6.011 (23.48)** 
2. City area 0.510 (2.59)** 0.341 (1.74) 0.504 (2.54)* −0.029 (1.51) 1.218 (5.22)** 3.009 (1.73) 
3. Dwellings per square mile 0.007 (0.57) 0.013 (0.71) 0.007 (0.59) 0.007 (5.72)** 0.057 (3.88)** 0.242 (2.22)* 
4. Population per dwelling −0.064 (0.02) 2.033 (0.71) 0.428 (0.01) 0.448 (1.54) 2.148 (0.60) 47.705 (1.78) 
5. Percent of state population urban 0.148 (1.36) 0.227 (2.27)* 0.153 (1.43) 0.028 (2.70)** 0.519 (4.11)** 2.914 (3.11)** 
6. Wells −0.946 (0.19) −0.476 (0.10) −0.930 (0.18) 0.173 (0.36) −4.980 (0.83) −5.762 (0.13) 
7. System number 10.115 (2.22)* 8.443 (1.89) 10.097 (2.15)* −0.134 (0.30) −1.289 (−0.23) −83.500 (2.00)** 
8. Age 0.883 (2.55)* 0.954 (2.95)** 0.893 (2.54)* 0.173 (0.36) 0.928 (2.25)* −9.764 (3.16)** 
9. Age squared −0.001 (0.22) −0.003 (0.68) −0.001 (0.24) −0.001 (0.04) −0.021 (4.06)** −0.200 (5.18)** 
10. Population per square mile (‘000s) 3.408 (1.55) 1.947 (0.95) 3.314 (1.52) −0.184 (0.88) −7.780 (3.04)** −38.027 (1.99)* 
11. Filters −8.815 (1.65) −8.908 (1.67) −8.811 (1.60) 0.160 (0.31) 4.956 (0.77) 69.779 (1.45) 
12. Pumping 8.770 (1.19) 10.291 (1.40) 8.880 (1.17) −1.189 (1.64) 6.572 (0.74) 86.328 (1.30) 
13. January average low temperature −0.115 (0.51) −0.048 (0.21) −0.111 (0.48) 0.028 (1.27) 0.586 (2.17)* −2.183 (1.08) 
14. Sewers 5.259 (1.03) 5.356 (1.04) 5.266 (1.00) 0.216 (0.43) 10.600 (1.71) 85.637 (1.83) 
Constant 3.194 (0.15) −10.792 (0.55) 2.287 (0.11) −1.846 (0.92) −51.450 (2.09)* −303.348 (1.64) 
Observations 258 258 258 258 258 256 
R2   0.49 0.65 0.87 0.93 
Fa   16.40 (14, 243) 32.16 (14, 243) 121.42 (14, 243) 214.03 (14, 241) 
χ2 test for variables 1–10 = 0a 214.59 (10)      
χ2 test for variables 1–9 = 0a  81.12 (9)     
F test for variables 1–9 = 0a   8.84 (9, 243) 9.91 (9, 243) 154.07 (9, 243) 258.32 (9, 241) 

Absolute value of z statistics in parentheses.

a

Degrees of freedom in parentheses.

*Significant at 5%.

**Significant at 1%.

Taps per mile is instrumented and included in all specifications. (Instrumented variables are denoted by a ^ in Panel A of Table 9.) The estimates reported in columns (1) and (2) were obtained using conditional maximum likelihood and differ only by the addition of population per square mile in column (2). The coefficient on taps per mile is positive and significant in both specifications, and population per square mile is negative and significant, mirroring previous results. The coefficient on filters is negative in both specifications but significant only in column (1), and the existence of sewers continues to be positively related to municipal ownership. The coefficient on January average low temperature, which was previously only marginally significant in some specifications, is significantly negative in both columns.54

The estimates reported in columns (4)–(7) include instrumented values of mains per square mile, miles of mains, and number of hydrants, either alone or in combination, in addition to taps per mile. It is common, as is the case here, for nonconcavity of the likelihood function to cause convergence problems using conditional maximum likelihood estimation when more than one explanatory variable is instrumented (StataCorp 2007: 20). The specifications in columns (4)–(7) therefore use an alternative two-step method, which can be used for hypothesis testing.55 Column (3) contains results using this two-step estimator to reestimate the specification of column (2) for comparison. As can be seen, the results are quite similar. The specification in column (4) repeats that in column (3) but substituting mains per square mile, instrumented for potential endogeneity, for population per square mile. The significance of the coefficient on mains per square mile falls just below the 0.05 level but, otherwise, the results are consistent with the preceding specification.

The remaining three columns augment the specification in column (3) with instrumented values of mains and hydrants, first individually and then together, as measures of waterworks investment. None of the coefficients on mains and hydrants differ significantly from zero in any of the specifications, while results for taps per mile and population density remain consistent with previous estimations. The variable sewers also retains its significance, while January average low temperature remains marginally significant (at the 0.10 level) except in specification (7).

Overall, the IV results support the main findings of the ordinary probit and reduced-form results. The estimated effect of net density (taps per mile) is positive and significant in all seven specifications, and the effect of gross density (population per square mile) is negative and significant in all five specifications in which it appears, while the coefficient on (instrumented) mains per square mile falls just short of significance at the 0.05 level. The coefficients on filters and pumping consistently have negative signs, but only the variable filters is ever significant in any of the IV estimations. None of the estimates provide any evidence of a positive effect of investment levels on municipal ownership.

Summary and Interpretation.

The most robust relationships in the preceding analysis are the correlations between ownership and measures of net and gross density. Waterworks are significantly more likely, both statistically and economically, to be municipally owned the higher the number of connections per mile of mains and the lower population (or, alternatively, miles of mains) per square mile. Figure 4 illustrates the importance of these density measures on municipal ownership based on the results in column (2) of Panel A of Table 9. The dark central curve shows the estimated probability of municipal ownership as a function of taps per mile of mains at the means of the other variables, and the lighter curves above and below show the estimated probabilities corresponding to the 5th and 95th percentile values of population per square mile (650 and 13,250), respectively. The marginal effect of taps per mile (at the means of other variables) is 0.008: 10 additional connections per mile of mains increase the probability municipal ownership by 8 percentage points. The corresponding marginal effect for population (in thousands) per square mile is −0.037.56 At the mean taps per mile of 73 (indicated by the vertical line in Figure 4), the probabilities of municipal ownership at the 5th and 95th percentile populations per square mile (indicated by the top and bottom curves) differ by 45 percentage points: 0.74 versus 0.29.

Figure 4.

Estimated Probability of Municipal Ownership.

Figure 4.

Estimated Probability of Municipal Ownership.

These results are consistent with the historical accounts suggesting that conflicts over extensions, especially into relatively sparsely populated outlying areas, and disruptions to commerce from street excavations, which would have increased with service density, were two major sources of friction between cities and private water companies. Alternative explanations for the results are, of course, possible. One alternative (suggested by a referee) is that areas with high service densities were also more likely to require investments in additional capacity (such as larger mains) to accommodate increased demands for consumption and firefighting. In the same way that conflicts arose over extensions of service to outlying areas, the need for improvements in existing service areas may also have been a source of conflict, especially if the improvements did not result directly in increased revenue for the company. Although taps per mile would then proxy for conflicts over system improvements rather than the disruptiveness of excavations, this interpretation is still consistent with relational frictions as a motive for municipal ownership.57

The results also provide some, albeit weak, support for Wilcox’ administrative simplicity hypothesis. The coefficients on filters and pumping (or pump capacity) are consistently negative, but only those on filters are ever significant at conventional levels.58 Waterworks with filters were between 10 and 20 percentage points less likely to be municipally owned in 1890, depending on specification.

By contrast, the results offer little support for the imperfect capital markets or relationship-specific investment hypotheses. With the exception of significant positive coefficients on the number of hydrants in 3 of 10 specifications (see Tables 8 and 9), none of the measures of system size and investments show a significant positive relation with municipal ownership, while two measures—filters and reservoir capacity—show significantly negative correlations with municipal ownership (at the 0.05 level) for a combined 10 of 24 coefficients in Figure 3 and Tables 8 and 9.59 Of particular note, the coefficients on the infrastructure dummy and miles of water mains—probably the best measures of investment size, a priori—never approach significant levels. The size and location specificity of waterworks investments, like those of other public utilities, no doubt account for why the provision of water service was governed by either long-term contracts or public ownership, but there is little evidence to suggest that the size of those investments was the determining factor in the choice between the two governance forms.

Of the remaining variables, two that show relatively consistent correlations with municipal ownership across specifications are sewers and temperature.60 None of the hypotheses examined provide a prediction regarding the relation between the presence of sewers and the ownership of waterworks. A possible explanation, given that sewer systems were almost always municipally owned, is that the higher probability of municipal ownership of waterworks in cities with sewer systems reflected scope economies in water and sewer system administration: The administrative burden of operating city-owned waterworks may have been lower for cities that already administered city-owned sewer systems.61

The negative correlation between temperature and municipal ownership could be consistent with either the investment or relational frictions hypotheses inasmuch as the potential for freezing in colder climates stands to increase both initial investment costs and the need for repairs and thereby the potential for disruptions. As Wilcox observed, “water is a thing that is subject to the action of frost, so that special precautions must be taken to prevent interruption of the supply in winter. For this reason, the construction of mains and house connections is expensive, because in a cold climate they must be laid deep in the ground where they cannot be reached by frost in the coldest weather. Otherwise it will be necessary … to incur great expense and inconvenience in frequently thawing out and repairing frozen pipes” (1910: 401).62 As in the case of hydrants, however, it is unclear why this source of investment expense, but not other more direct measures of investment levels, would prompt municipal ownership. Finally, the inverse relationship between temperature and municipal ownership is the opposite what would be expected if public health concerns motivated municipal ownership given that the frequency and severity of both typhoid and cholera epidemics were greater in warmer climates.63

Trends and the Role of the Institutional Environment

The preceding provides a snapshot at a point in time of the characteristics and ownership of waterworks built over the course of a century. Although panel data would be superior, in its absence a cross-sectional analysis can be justified inasmuch as cities faced an ongoing choice of whether to retain or change ownership given their circumstances at the time. Nevertheless, a satisfactory resolution of the waterworks ownership puzzle should conform with the long-term trend away from private and toward public ownership of waterworks over the course of the 19th century.

At an aggregate level, that trend is consistent with several hypotheses. To the extent, for example, that waterworks construction was becoming more expensive, as Cutler and Miller contend, the shift to municipal ownership is consistent with both relationship-specific investment and capital market imperfections hypotheses. Relational frictions could also explain the trend, however, if the growing urbanization of the American population increased either the density of areas served by waterworks or the need for system extensions.64 And even though the increase in water filtration over this period (see Table 2) works against the administrative simplicity hypothesis, the overall trend might still be reconciled with that explanation to the extent that greater familiarity with and routinization of waterworks operations over time reduced the incentive disadvantages of public ownership.

Finally, the possibility remains that changes over time in the institutional environment were responsible for the general trend away from private water supply. A thorough analysis of the role of institutions is beyond the scope of this study. Nevertheless, to gauge the extent to which state-level institutional factors might have influenced waterworks ownership or otherwise affect the preceding results, I reestimated the determinants of ownership including indicators for restrictions on state and local borrowing and state fixed effects. The borrowing restriction variables are derived from Table 11.8 of Wallis and Weingast, which identifies whether states “have a restriction that limits the procedures by which states can issue debts, typically a referendum; … have absolute dollar limits on debt[, or have] some type of restriction or regulation on the issue of debt by local governments,” along with the year in which each restriction was adopted (2008: 352). Variables were set equal to one if a waterworks was constructed in a state and year in which a corresponding restriction applied and zero otherwise.65

The estimations reported in column (1) of Table 10 use the same specification as in column (8) of Table 8 except for the inclusion of the three borrowing restriction indicators. (The control variables and equation constant are included but not reported in all three columns.) The results indicate that waterworks constructed in states with restrictions or limits on state borrowing were less likely to be municipally owned in 1890 (by approximately 25 percentage points), and those in states with restrictions on local borrowing were more likely to be municipally owned (by 17 percentage points), although the coefficient on local restrictions is significant only at the 0.10 level.66 Finally, columns (2) and (3) of Table 10 report conditional logit results including state fixed effects with and without the borrowing restriction indicators. Perhaps not surprisingly, borrowing restrictions no longer have significant effects when state fixed effects are included.67

Table 10.

Estimation of Waterworks Ownership with State Effects (Dependent Variable = 1 if Municipally Owned in 1890)

 (1) Probit (2) Conditional logit (3) Conditional logit 
Filters −0.393 (1.67) −0.512 (1.07) −0.530 (1.09) 
Pump capacity (millions) −0.010 (0.92) −0.025 (1.05) −0.030 (1.22) 
Population per square mile (‘000s) −0.092 (2.70)** −0.201 (2.74)** −0.195 (2.63)** 
Taps per mile 0.011 (3.22)** 0.022 (2.97)** 0.022 (2.86)** 
Infrastructure 0.089 (0.43) −0.130 (0.30) −0.103 (0.23) 
Reservoir capacity (millions) −0.0002 (1.77) −0.001 (1.61) −0.001 (0.74) 
Mains 0.001 (0.34) −0.005 (0.49) −0.005 (0.39) 
Hydrants 0.0003 (0.48) 0.001 (0.84) 0.002 (1.09) 
State borrowing restriction −0.761 (2.55)*  −2.301 (1.13) 
State borrowing limit −0. 751 (2.00)*  1.372 (0.59) 
Local borrowing restriction 0.478 (1.83)  0.151 (0.14) 
Controlsa Yes Yes Yes 
State fixed effects No Yes Yes 
Observations 248 222 222 
Number of states 40 26 26 
χ2 (d.f.) 73.77 (20) 49.09 (17) 50.52 (20) 
Log likelihood −132.6759 −73.1505 −72.2868 
Pseudo-R2 0.22 0.25 0.26 
 (1) Probit (2) Conditional logit (3) Conditional logit 
Filters −0.393 (1.67) −0.512 (1.07) −0.530 (1.09) 
Pump capacity (millions) −0.010 (0.92) −0.025 (1.05) −0.030 (1.22) 
Population per square mile (‘000s) −0.092 (2.70)** −0.201 (2.74)** −0.195 (2.63)** 
Taps per mile 0.011 (3.22)** 0.022 (2.97)** 0.022 (2.86)** 
Infrastructure 0.089 (0.43) −0.130 (0.30) −0.103 (0.23) 
Reservoir capacity (millions) −0.0002 (1.77) −0.001 (1.61) −0.001 (0.74) 
Mains 0.001 (0.34) −0.005 (0.49) −0.005 (0.39) 
Hydrants 0.0003 (0.48) 0.001 (0.84) 0.002 (1.09) 
State borrowing restriction −0.761 (2.55)*  −2.301 (1.13) 
State borrowing limit −0. 751 (2.00)*  1.372 (0.59) 
Local borrowing restriction 0.478 (1.83)  0.151 (0.14) 
Controlsa Yes Yes Yes 
State fixed effects No Yes Yes 
Observations 248 222 222 
Number of states 40 26 26 
χ2 (d.f.) 73.77 (20) 49.09 (17) 50.52 (20) 
Log likelihood −132.6759 −73.1505 −72.2868 
Pseudo-R2 0.22 0.25 0.26 

Absolute value of z statistics in parentheses.

a

Equation constant and the following control variables included but not reported: system number, January average low temperature, 1890 population, city area, 1890–1900 population change, sewers, wells, age, and age squared.

*Significant at 5%.

**Significant at 1%.

The results in column (1) suggest that institutional factors, although coarsely measured, may have contributed to choice of governance form for waterworks, supporting further work on the subject.68 At the same time, the results show the relationships between ownership and city and waterworks characteristics identified in the preceding analysis to be robust to state-level institutional differences.

Conclusions

Water and sewer systems are unique among major US public utilities in being predominantly publicly owned. Yet for most of the 19th century, most US cities obtained water, like other public utility services, from private companies under franchise contracts. Moreover, substantial numbers of private waterworks continued to be built and operated into the 20th century even as municipal ownership became ascendant: Between 1896 and 1924, the number of private waterworks doubled, from 1489 to 2950. The puzzle then is not just why public ownership of waterworks came to dominate but why, despite the overall trend, large numbers of new and existing waterworks remained private.

I approach the puzzle, first, by reviewing late-19th- and early-20th-century accounts of the problems of public utility governance in light of modern theories of economic organization. In addition to revealing a remarkably sophisticated understanding of the trade-offs inherent in the choice of governance arrangements, those accounts suggest two theoretically plausible explanations for the relative prevalence for municipal ownership of water and sewer systems: (a) the administrative simplicity of their operations compared to other utilities, which may have reduced the well-known inefficiencies of public supply, and (b) a combination of (1) constraints on pricing arrangements that necessitated costly negotiation whenever system extensions or improvements were desired and (2) the disruptiveness of street excavations, which provided a valuable if costly tactic in those negotiations.

Analysis of the ownership of waterworks serving cities with populations over 10,000 in 1890 provides some support for both explanations. Waterworks that filtered or pumped water—operations that transform water supply from delivery of a “natural product” into “a process of manufacture” (Wilcox 1910: 400)—showed a modest tendency to be privately owned. Although that tendency was statistically weak, it nevertheless seems plausible that the lower technical demands of operating water and sanitation systems relative to producing electricity, gas, and telephone service made the loss of high-powered incentives and other drawbacks to government ownership of these systems less of a deterrent to public supply.

In contrast, the results relating ownership to service and population densities require no such qualifications. Waterworks were consistently more likely to be municipally owned the higher the number of connections per mile of water mains and the lower either population per square mile or miles of mains per square mile of city area. High service densities (many connections per mile of mains) would tend to increase the disruptiveness of excavations, while service extensions to areas with low population density (hence, fewer potential customers) would require larger transfers—in effect, subsidies—the size and continued payment of which were a common source of friction between cities and private companies. The conceptual link between the hypotheses and the empirical measures is tenuous enough, however, that a conclusion that conflicts over extensions and excavations were central determinants of waterworks ownership would not be justified without the substantial corroboration of the historical accounts. At a minimum, the strength of the findings demand that alternative hypotheses account for these correlations.

Finally, the evidence offers no support for the frequently made claim that municipal ownership of waterworks was a function of city or system size. None of the measures of investment size are consistently associated with municipal ownership, and those showing the most significant effects—filters and reservoir capacity—are negatively related to municipal ownership. This does not mean that the location-specific nature of these investments is unimportant. As in other contexts, transactors will be reluctant to make investments without some form of protection, usually long-term contracts or vertical integration. But which of those alternatives will be preferred often turns, as Williamson has emphasized, on factors affecting the relative costs of adapting to changing circumstances under those alternatives, as appears to be the case here.

Understanding the forces underlying the ownership of waterworks is of more than just historical interest. The World Bank has recently been actively studying privatization of waterworks as a way of addressing severe water problems in developing countries (see Shirley 2002).69 The dominance of public ownership of water and sanitation services in the United States should, at a minimum, give policymakers pause: If, despite an institutional environment conducive to private ownership, American water and sanitation systems are overwhelmingly publicly owned and operated, is it reasonable to expect privatization to yield long-term gains in developing countries where the environment for private enterprise may be much less hospitable? Understanding the determinants of variations in waterworks ownership both over time and between communities may help inform whether privatization of water systems in developing countries makes sense (see Crocker and Masten 2002, for further discussion).

At a broader level, the findings in this article have implications for our understanding of contracting and firm boundaries. The first is the importance of ex post governance costs, that is, the costs of adapting to new circumstances and of efforts to evade or force renegotiation of contract terms. A second is the role of prices and price adjustment in reducing those costs: In addition to their traditional roles in aligning incentives and distributing surpluses, prices also affect the likelihood that parties will engage in opportunistic efforts to redistribute contractual surpluses during contract execution. Finally, although the literature has tended to emphasize factors affecting the costs of market transactions, systematic variations in the costs of internal organization may also affect firm boundaries, as the administrative simplicity hypothesis regarding waterworks ownership illustrates.

Funding

University of Michigan Ross School of Business.

I would like to thank Louis Cain, Guy Holburn, Pablo Spiller, Catherine Waddams, seminar participants at the University of California at Berkeley, University of California at Los Angeles, the University of Michigan, the University of Western Ontario, Washington University-St. Louis, and Western Michigan University, and three anonymous referees for helpful comments. I would also like to thank Blake Gower, Celia Kujala, and Meagan Callan Masten for excellent research assistance. This article builds on an earlier study, Crocker and Masten (2002), which was commissioned by The World Bank.

References

Anas
Alex
Arnott
Richard
Small
Kenneth A
“Urban Spatial Structure,”
Journal of Economic Literature
 , 
1998
, vol. 
36
 (pg. 
1426
-
64
)
Anderson
Letty
“Hard Choices: Supplying Water to New England Towns,”
Journal of Interdisciplinary History
 , 
1984
, vol. 
15
 (pg. 
211
-
34
)
Averch
H
Johnson
LL
“Behavior of the Firm under Regulatory Constraint,”
American Economic Review
 , 
1962
, vol. 
52
 (pg. 
1052
-
69
)
Baker
George
Gibbons
Robert
Murphy
Kevin J
“s Happens: Relational Adaptation in Contracts, Firms, and Other Governance Structures.”
 , 
2009
 
Working Paper, Massachusetts Institute of Technology (July)
Moses N
Baker
1889. The Manual of American Water-Works, 1888
 , 
1889
New York
The Engineering New Publishing Co
———. ed
The Manual of American Water-Works, 1897
 , 
1897
New York
The Engineering New Publishing Co
Bemis
EW
———
“Water-Works,”
Municipal Monopolies
 , 
1899
New York
Thomas Y. Crowell & Company
Cain
Louis
“An Economic History of Urban Location and Sanitation,”
Research in Economic History
 , 
1977
, vol. 
2
 (pg. 
337
-
89
)
Coase
Ronald
“The Nature of the Firm,”
Economica
 , 
1937
, vol. 
4
 (pg. 
386
-
405
)
———
“The Problem of Social Cost,”
Journal of Law and Economics
 , 
1960
, vol. 
3
 (pg. 
1
-
44
)
Fuchs
VR
———
“Industrial Organization: A Proposal for Research,”
Policy Issues and Research Opportunities in Industrial Organization
 , 
1972
New York
National Bureau of Economic Research
———
“The Nature of the Firm: Origin, Meaning, Influence,”
Journal of Law, Economics, and Organization
 , 
1988
, vol. 
4
 (pg. 
3
-
47
)
Cowan
S
“Regulation of Several Market Failures: The Water Industry in England and Wales,”
Oxford Review of Economic Policy
 , 
1993
, vol. 
9
 (pg. 
14
-
22
)
Crain
WM
Zardkoohi
A
“A Test of the Property Theory of the Firm: Water Utilities in the United States,”
Journal of Law and Economics
 , 
1978
, vol. 
21
 (pg. 
395
-
408
)
Crocker
Keith J
Masten
Scott E
Pretia Ex Machina? Prices and Process in Long-Term Contracts,”
Journal of Law and Economics
 , 
1991
, vol. 
34
 (pg. 
69
-
99
)
———
“Regulation and Administered Contracts Revisited: Lessons from Transaction-Cost Economics for Public Utility Regulation,”
Journal of Regulatory Economics
 , 
1996
, vol. 
9
 (pg. 
5
-
39
)
Shirley
M
———
“Prospects for Private Water Provision in Developing Countries: Lessons from 19th-Century America,”
Thirsting for Efficiency: The Economics and Politics of Urban Water System Reform
 , 
2002
Oxford
Elsevier Science Ltd
Cutler
David
Miller
Grant
Glaeser
E
Goldin
C
“Water, Water Everywhere: Municipal Finance and Water Supply in American Cities,”
Corruption and Reform: Lessons from America’s Economic History
 , 
2006
Chicago, IL
University of Chicago Press
Demsetz
H
“Why Regulate Utilities?,”
Journal of Law and Economics
 , 
1968
, vol. 
11
 (pg. 
55
-
66
)
Edison Electric Institute
Historical Statistics of the Electric Utility Industry Through 1970
 , 
1973
New York
Edison Electric Institute
———
Statistical Yearbook of the Electric Utility Industry, 1981
 , 
1982
Washington, DC
Edison Electric Institute
Feigenbaum
S
Teeples
R
“Public versus Private Water Delivery: A Hedonic Cost Approach,”
Review of Economics and Statistics
 , 
1983
, vol. 
65
 (pg. 
672
-
78
)
Gibson
Campbell
“Population of the 100 Largest Cities and Other Urban Places in the United States: 1790 to 1990.”
 , 
1998
 
Population Division Working Paper No. 27, U.S. Bureau of the Census
Goldberg
Victor P
“Regulation and Administered Contracts,”
Bell Journal of Economics
 , 
1976
, vol. 
7
 (pg. 
426
-
48
)
———
“Price Adjustment in Long-Term Contracts,” 1985
Wisconsin Law Review
 , 
1985
(pg. 
527
-
43
)
Goldberg
Victor P
Erickson
John R
“Quantity and Price Adjustment in Long-Term Contracts: A Case Study in Petroleum Coke,”
Journal of Law and Economics
 , 
1987
, vol. 
31
 (pg. 
369
-
98
)
Hart
Oliver
“Hold-Up, Asset Ownership, and Reference Points,”
Quarterly Journal of Economics
 , 
2009
, vol. 
124
 (pg. 
267
-
300
)
Jacobson
Charles D
“Same Game, Different Players: Problems in Urban Public Utility Regulation, 1850–1987,”
Urban Studies
 , 
1989
, vol. 
26
 (pg. 
13
-
31
)
Jacobson
Charles D
Tarr
Joel A
“Ownership and Financing of Infrastructure: Historical Perspectives.”
1995
 
Policy Research Working Paper No. 1466, The World Bank
John
Richard R
“Telephomania: The Contested Origins of the Urban Telephone Operating Company in the United States, 1879–1894.”
 , 
2005
 
Great Cities Institute Working Paper No. GCP-05-02, University of Illinois at Chicago
Kim
Sukkoo
“Urban Development in the United States, 1690–1990,”
Southern Economic Journal
 , 
2000
, vol. 
66
 (pg. 
855
-
80
)
———
“The Reconstruction of the American Urban Landscape in the Twentieth Century.”
2002
 
Working Paper No. 8857, National Bureau of Economic Research. http://www.nber.org/papers/w8857
———
“Changes in the Nature of Urban Spatial Structure in the United States, 1890–2000,”
Journal of Regional Science
 , 
2007
, vol. 
47
 (pg. 
273
-
87
)
Klein
Benjamin
Werin
Lars
Wijkander
Hans
“Contracts and Incentives: the Role of Contract Terms in Assuring Performance,”
Contract Economics
 , 
1992
Cambridge, MA
Basil Blackwell
———
“Why Hold-Ups Occur: The Self-Enforcing Range of Contractual Relationships,”
Economics Inquiry
 , 
1996
, vol. 
34
 (pg. 
444
-
63
)
Lebergott
Stanley
Manpower in Economic Growth: The American Record Since 1800
 , 
1964
New York
McGraw-Hill
Levy
Brian
Spiller
Pablo T
“The Institutional Foundations of Regulatory Commitment: A Comparative Analysis of Telecommunications Regulation,”
Journal of Law, Economics & Organization
 , 
1994
, vol. 
10
 (pg. 
201
-
46
)
Masten
Scott E
“Equity, Opportunism, and the Design of Contractual Relations,”
Journal of Institutional and Theoretical Economics
 , 
1988
, vol. 
144
 (pg. 
180
-
95
)
———
“Long-Term Contracts and Short-Term Commitment: Price Determination for Heterogeneous Freight Transactions,”
11 American Law and Economics Review
 , 
2009
 
79–111
Masten
Scott E
Meehan
James W
Jr.
Snyder
Edward A
“The Costs of Organization,”
Journal of Law, Economics, and Organization
 , 
1991
, vol. 
7
 (pg. 
1
-
25
)
McGraw Waterworks Directory
1915
New York
McGraw Publishing Company, Inc
Melosi
Martin
The Sanitary City: Urban Infrastruture in America from Colonial Times to the Present
 , 
2000
Baltimore, MD
Johns Hopkins University Press
Menes
Rebecca
“The Effect of Patronage Politics on City Government in American Cities, 1900–1910.”
1999
 
Working Paper No. 6975, National Bureau of Economic Research
———
 
2006. “Limiting the Reach of the Grabbing Hand: Graft and Growth in American Cities, 1880 to 1930,” in E. Glaeser and C. Goldin, eds., Corruption and Reform: Lessons from America’s Economic History. Chicago, IL: University of Chicago Press
National Research Council Committee on Privatization of Water Services in the United States
“Privatization of Water Services in the United States: An Assessment of Issues and Experience.”
2002
 
American Academy of Sciences (http://www.nap.edu/catalog/10135.html)
Oyer
Paul
“Why Do Firms Use Incentives That Have No Incentive Effects?,”
Journal of Finance
 , 
2004
, vol. 
59
 (pg. 
1619
-
50
)
Pascual
Mercedes
Bouma
Menno J
Dobson
Andrew P
“Cholera and Climate: Revisiting the Quantitative Evidence,”
Microbes and Infection
 , 
2002
, vol. 
4
 (pg. 
237
-
45
)
Pashigian
B Peter
“Consequences and Causes of Public Ownership of Urban Transit Facilities,”
Journal of Political Economy
 , 
1976
, vol. 
84
 (pg. 
1239
-
59
)
Priest
George L
“The Origins of Utility Regulation and the ‘Theories of Regulation’ Debate,”
Journal of Law and Economics
 , 
1993
, vol. 
36
 (pg. 
289
-
323
)
Rivers
Douglas
Vuong
Quang H
“Limited Information Estimators and Exogeneity Tests for Simultaneous Probit Models,”
Journal of Econometrics
 , 
1988
, vol. 
39
 (pg. 
347
-
66
)
Rosen
Christine Meisner
The Limits of Power: Great Fires and the Process of City Growth in American
 , 
1986
Cambridge
Cambridge University Press
Shammas
Carole
“The Space Problem in Early United States Cities,”
William and Mary Quarterly
 , 
2000
, vol. 
57
 (pg. 
505
-
42
)
Shirley
Mary
Thirsting for Efficiency: The Economics and Politics of Urban Water System Reform
 , 
2002
 
Oxford: Elsevier Science Ltd
Solley
Wayne B
Pierce
Robert R
Perlman
Howard A
“Estimated Use of Water in the United States in 1995,”
U.S. Geological Circular 1200
 , 
1998
Washington, DC
US Government Printing Office
Spiller
Pablo T
“An Institutional Theory of Public Contracts: Regulatory Implications.”
2008
 
Working Paper No. 14152, National Bureau of Economic Research. http://www.nber.org/papers/w14152
Spiller
Pablo T
Savedoff
William D
Savedoff
WD
Spiller
PT
“Government Opportunism and the Provision of Water,”
Spilled Water: Institutional Commitment in the Provision of Water Services
 , 
1999
Washington, DC
Inter-American Development Bank
StataCorp
Statistical Software: Release 10, Reference I-P
 , 
2007
College Station, TX
Stata Corporation
Tarr
Joel A
The Search for the Ultimate Sink: Urban Pollution in Historical Perspective
 , 
1996
Akron, OH
The University of Akron Press
Troesken
Werner
“The Sources of Public Ownership: Historical Evidence From the Gas Industry,”
Journal of Law, Economics & Organization
 , 
1997
, vol. 
13
 (pg. 
1
-
25
)
———
“Typhoid Rates and the Public Acquisition of Waterworks, 1880–1925,”
Journal of Economic History
 , 
1999
, vol. 
59
 (pg. 
927
-
48
)
Troesken
Werner
Geddes
Rick
“Municipalizing American Waterworks, 1897–1915,”
Journal of Law, Economics & Organization
 , 
2003
, vol. 
19
 (pg. 
373
-
400
)
US Bureau of Labor
Water, Gas, and Electric-Light Plants Under Private and Municipal Ownership, Fourteenth Annual Report of the Commissioner of Labor, 1899
 , 
1900
Washington, DC
US Government Printing Office
US Bureau of the Census
Report on the Social Statistics of Cities in the United States at the Eleventh Census: 1890
 , 
1895
Washington, DC
US Government Printing Office
———
Twelfth Census of the United States—1900, Census Reports Volume I—Population Part I
 , 
1901
Washington, DC
US Government Printing Office
———
Statistics of Cities Having a Population over 30,000: 1905
 , 
1907
Washington, DC
US Government Printing Office
———
1990 Population and Housing Unit Counts: United States (CPH-2)
 , 
1990
Washington, DC
US Government Printing Office
US Department of Commerce
CLIM81: Climatography of the U.S., No. 81: Monthly Station Normals
 , 
1991
Asheville, NC
National Climate Data Center
US Energy Information Administration
“The Changing Structure of the Electric Power Industry 2000: An Update.”
2000
 
US Environmental Protection Agency
“National Characteristics of Drinking Water Systems Serving Populations Under 10,000.”
Office of Water
1999
Washington, DC
 
(EPA 816-R-99–010)
———
“Community Water System Survey.”
Office of Water
 , 
2002
Washington, DC
 
(EPA 815-R-02–005A)
Veysey
Laurence R
The Emergence of the American University
 , 
1965
Chicago, IL
The University of Chicago Press
Wallis
John J
Weingast
Barry R
Garrett
Elizabeth
Graddy
Elizabeth A
Jackson
Howell E
“Dysfunctional or Optimal Institutions? State Debt Limitations, the Structure of State and Local Governments, and the Finance of American Infrastructure,”
Fiscal Challenges: An Interdisciplinary Approach to Budget Policy
 , 
2008
New York
Cambridge University Press
Waterman
Earle Lytton
Elements of Water Supply Engineering
 , 
1934
New York
Wiley and Sons
Weiss
Thomas
Gallman
RE
Wallis
JJ
“U.S. Labor Force Estimates and Economic Growth, 1800–1860,”
American Economic Growth and Standards of Living before the Civil War
 , 
1992
Chicago, IL
The University of Chicago Press
Wilcox
Delos F
Municipal Franchises: A Description of the Terms and Conditions upon which Private Corporations Enjoy Special Privileges in the Streets of American Cities
 , 
1910
, vol. 
Vol. 1
 
Rochester, NY
The Gervaise Press
———
“The Regulation of Private Water Companies in New York City,”
31 Journal of the New England Water Works Association
 , 
1917
 
550–74
———
The Administration of Municipally Owned Utilities
 , 
1931
New York
Municipal Administration Service
Williamson
Oliver E
“The Vertical Integration of Production: Market Failure Considerations,”
American Economic Review
 , 
1971
, vol. 
61
 (pg. 
112
-
23
)
———
“Franchise Bidding for Natural Monopolies—In General and with Respect to CATV,”
Bell Journal of Economics
 , 
1976
, vol. 
7
 (pg. 
233
-
62
)
———
“Transaction-Cost Economics: The Governance of Contractual Relations,”
Journal of Law and Economics
 , 
1979
, vol. 
22
 (pg. 
233
-
62
)
———
“Credible Commitments: Using Hostages to Support Exchange,”
73 American Economic Review
 , 
1983
(pg. 
519
-
40
)
———
The Economic Institutions of Capitalism
 , 
1985
New York
The Free Press
———
The Mechanisms of Governance
 , 
1996
New York
Oxford University Press
Wooldridge
Jeffrey M
Econometric Analysis of Cross Section and Panel Data
 , 
2002
Cambridge, MA
MIT Press
Bank
World
World Bank Development Report 1994: Infrastructure for Development
 , 
1994
New York
Oxford University Press
1
Although investor-owned utilities continue to account for the bulk of electricity production and sales, publicly owned electric utilities actually outnumber investor-owned companies: The 239 investor-owned utilities operating in 1998 accounted for 75% of electricity sales compared to only about 15% for the 2009 nonfederal publicly (i.e., municipal and state) owned utilities (US Energy Information Administration 2000: 17–8, 26, 28). (The remainder was provided by cooperatives, 8.6%, and federal utilities, 1.5%.) The discrepancy between market share and numbers results from the fact that publicly owned electric utilities generally serve a single, small (usually rural) community, whereas private electric companies usually serve large numbers of communities within broad geographic regions. In contrast to electric utilities (and gas works; see Troesken 1997; Troesken and Geddes 2003: 378), privately owned waterworks primarily serve single communities.
2
Troesken and Geddes (2003) also use a large data set in their study of waterworks municipalization. Their analysis, however, only examines conversions of originally private works to municipal ownership, not why cities chose municipal ownership from the outset or why, as shown here, significant numbers of private works continued to be built despite the overall trend toward municipal ownership.
3
Cowan adds that “the existence of a network is not in itself a barrier to competition. Firms could compete to supply water … if they shared access to the pipe networks. The prospect of such competition in service provision is, however, small … .because of the high cost of pumping long distances” (1993: 16).
4
Individual provision, usually from wells, and cooperative organization, mainly for small-scale water systems such as for homeowners associations, remain a small but significant source of water. Approximately 16% of the US population obtains water for domestic use through such “self-supply” (Solley et al. 1998).
5
For the first three-quarters of the century, city dwellers continued to dispose of their wastes in cesspools and privy vaults or, illegally, in streets and vacant lots or in storm sewers, whose use for sewage disposal was discouraged and sometimes forbidden by city ordinance (Tarr 1996: 8–9). The increasing volume of waste quickly overwhelmed these methods of disposal, however. Overflowing cesspools and improperly disposed of waste created public nuisances; cellars “flooded with stagnant and offensive fluids” at such a rate as to make cleaning “nearly futile.” Central water supplies, once contaminated, distributed “fecal pollution over immense areas” (ibid., 10). Faced with the progressive inadequacy of traditional waste disposal methods, public officials in some states sought to stem the tide of filth and disease by barring installation of water closets in towns without sewers.
6
The Municipal Year Book of 1902 listed 47 cities and towns, among those with populations of at least 3000, having sewer systems operated by private companies under franchise contracts, compared to 1045 communities with publicly owned and operated systems (Wilcox 1910: 451).
7
Cf. Melosi (2000: 123): “By and large, rapidly increasing demand for water rose beyond the capacity of most private companies to meet it.” Cutler and Miller conflate the sharp rise in waterworks construction that occurred at the end of the century with the more gradual shift in waterworks ownership that began much earlier and then seek to explain both as a single phenomenon. The fact, discussed below, that large numbers of private waterworks continued to be built through 1896, and then again between 1915 and 1924, shows that waterworks construction and waterworks governance are separable issues.
8
For an excellent discussion of the relation between the institutional environment and public utility governance generally, see Levy and Spiller (1994).
9
A simple regression of the percent of waterworks publicly owned (from Table 1) against a time trend for the years 1800 to 1896 yields predictions for 1920 and 1924 of 67.88% and 69.78%, respectively.
10
Because conversions in ownership affect the number of works of each ownership type at a point in time, and because conversions from private to public were more numerous, more private works and fewer public works were being constructed than implied by the changes in the totals, especially during the later decades covered by Table 1. According to Baker (1899: 24), 205 of the 1690 public works in 1896 were previously private, whereas 20 of the 1489 private works were originally public. Troesken and Geddes (2003) report that an additional 258 private works were converted from private to public between 1896 and 1915, but see footnote 18.
11
The data on which Figure 2 is based are described in more detail in Section 5.
12
The logical link between externalities and ownership was, of course, severed by Coase (1960). An analogous critique applies to the natural monopoly rationale for public utility regulation and, by extension, for public ownership (see Demsetz 1968).
13
Although some waterworks supplied water for consumption or firefighting alone, the vast majority, including all 1489 of the private works in 1896 shown in Table 1, “furnish[ed] both domestic supply and fire protection” (Baker 1897: H). In addition to the 3196 “complete” waterworks in Table 1, Baker identified another 207 works providing domestic supply only and 108 that supplied water for fire protection only (ibid., G, H).
14
The trend toward municipal ownership of waterworks also appears to predate understanding by public authorities and the public as a whole of the role of water as a conduit for disease; although the number of public works grew throughout the century, the relationship between waste and disease did not become generally accepted until the 1890s (Tarr 1996: 123).
15
Many of the central arguments and concepts associated with modern transaction cost theory can, in fact, be traced to the debate over public utility governance. See, in particular, Williamson (1976) and Goldberg (1976). For a review of the development of the theory and the related empirical literature as it applies to public utility governance, see Crocker and Masten (1996).
16
Spiller and Savedoff (1999) provide a detailed and insightful analysis of appropriation problems in the supply of water in the context of modern-day Latin America. Among their predictions is that, holding the institutional environment constant, governmental incentives to expropriate quasi-rents associated with public utility investments will be larger “when there are fewer private operators in the infrastructure sector; when the sectors do not, in general, require massive investment programs; and when technological change is not an important factor in the sector” (ibid., 8). An implication of their analysis is that the desire to maintain incentives for future private investment will deter cities that anticipate a need for large future investments from expropriating quasi-rents associated with previous investments and, conversely, that appropriation hazards will increase as the need for future investment falls. By this argument, more durable infrastructure for water supply could favor public ownership by decreasing the frequency of reinvestment and, thereby, increasing appropriation hazards. The argument also suggests that public ownership should be more common, other things the same, for cities with lower anticipated growth and with more complete water systems.
17
Wilcox describes the process of water quality measurement and the equivalence of that process for public and private waterworks (1917: 560–1):

The [water] department maintains a laboratory at which its own water supply is subjected to continual examination as to its chemical and bacteriological contents. In like manner, samples are taken monthly from the different pumping stations of the several private companies and subjected to careful scrutiny. Whenever such examinations show indications of possible pollution, the companies are notified and special investigations are made to find out what the trouble is. These laboratory tests are also used for the purpose of determining the quality of the water with relation to such matters as its iron content, its hardness, or the presence of obnoxious gases tending to impair its potability. In this way the department performs for the private water companies and for their consumers a fundamental service by the insistence upon high standards in relation to the purity and potability of the water supply drawn from private sources, the same as it does as to the sources which the municipality itself has developed [emphasis added].

18
Although Troesken and Geddes more than once describe the municipalization rate as 45% (2003: 389, 391), their data indicate that the number of works municipalized between 1897 and 1915 was 258 of 726 works or 35% (ibid., 389, 390). The research also contains other discrepancies. For example, the authors claim repeatedly that their data contain “the universe of all private water companies operating in America in 1897” (ibid., 398, also 385, 389), yet their sample contains fewer than half (726 versus 1489) of the private waterworks identified in their source, Baker (1987). In addition, both their statement that “roughly 40% of all municipalization proceedings ended in litigation” (ibid., 381) and their conclusion that litigation in any category increased the probability that a works would be municipalized by 20–40 percentage points (ibid., 392) are hard to reconcile with their descriptive statistics, which indicate that fewer than 18% of works eventually municipalized had experienced any type of prior litigation (ibid., 391).
19
Stanford University's controversial firing of economist Edward A. Ross in 1900, for example, resulted, in part, from Ross’ unpopular position (pro) on municipal ownership of public utilities (see, e.g., Veysey 1965: 401–2).
20
Other advantages on Wilcox’ list were territorial flexibility, freedom of association with other utilities, and freedom from debt limits and flexibility in financial arrangements (Wilcox 1931: 6).
21
According to Wilcox, public demands for rate reductions were often pure rent redistributions: “often the demand for lower rates has arisen from the public recognition of high or excessive dividends as much as from any claim that the rates charged in themselves are unreasonable and excessive” (Wilcox 1931: 28).
22
Corruption was a particular concern in a period known for political bosses and party machines. Notorious political machines dominated municipal governments in New York City, San Francisco, Denver, Cincinnati, and Philadelphia, among other cities (see Menes 1999, 2006). Whether private or public ownership was more conducive to the exercise of corruption was and remains an open question, however. Several major cities, including New York, San Francisco, St. Louis, and Grand Rapids, obtained bribery convictions in connection with the awarding of utility franchises (see Wilcox 1910: chapter V). The ability to check corruption in the operation of utilities (as opposed to the awarding of franchises) may also have been greater under public than private ownership: “the difficulty of proving fraud or corruption” could make it harder to dislodge a private company that had secured a franchise than to turn out a “thieving city council or board of public works” (Baker 1899: 48). But as Baker observed, “The man who stands ready to secure personal advantages or profit to the detriment of those whom he has been chosen to serve will not be at all particular whether he does this through public or private ownership of municipal monopolies” (ibid., 36). For a discussion of modern-day problems of government opportunism in water supply in Latin America, see Spiller and Savedoff (1999).
23
For an overview of the technological and operational problems of 19th-century telephony, see John (2005).
24
This is not to say that the design of sewer systems did not involve important engineering and technological issues. On the contrary, the issue of the most effective and economical sewer design was a matter of significant debate among civil engineers in the late 1800s (see Tarr 1996). The complexities of sewer systems, however, involved their design rather than their operation.
25
Sewerage pumping and purification were, in fact, quite exceptional. As late as 1905, of the 154 cities having populations greater than 30,000, only 9 pumped all, and another 9 some, of their sewage and only 16 operated purification plants (Bureau of Census Special Report 1905, as reported by Wilcox 1910: 452–3).
26
The notion that differences in costs of internal administration play a role in integration decisions can be found in “The Nature of the Firm” (Coase 1937). Since then, Coase (1972, 1988) has consistently emphasized differences in the costs of organizing activities within firms as the essential determinant of the boundaries of firms. Strictly speaking, the argument that operational complexity and unfamiliarity favor contracting requires that costs of contracting do not also increase—or, at least, do not increase as rapidly as the inefficiencies of public ownership—with the complexity and unfamiliarity of the production process. It is worth noting, in this regard, the distinction between operational and product complexity. Operational complexity is likely to exacerbate the difficulty of monitoring production processes, whereas product complexity is likely mainly to complicate contracting and be relatively less important for supervision. See Masten et al. (1991) for further discussion and for evidence that dissimilarity (unfamiliarity) increases internal organization costs relative to contracting costs.
27
In the case of waterworks, capacity was driven mainly by the peak-load requirements of water for firefighting, leaving capacity for water delivery far in excess of a community's consumption needs. According to Baker (1899: 39),

the cost of the public service depends far less upon the total consumption in a year than upon the comparatively vast quantities which may be demanded on special brief occasions, principally at fires. Thus the use of water for all public purposes may be not over five per cent of the total consumption for the year, but the whole system must be so proportioned as to be capable of delivering an enormous increase of volume during a severe conflagration.

28
Note that it is the effect of public health externalities on pricing arrangements, rather than the externalities per se, that impairs private provision in the analysis below. Wilcox attributed “the absence of private operation of sewerage systems,” in part, to “the unwillingness of citizens generally to purchase a service of this nature of their own free will.” (1910: 453). The consequences of deterring use of sanitary sewers in large cities are best left uncontemplated.
29
The low marginal cost of water and sanitation services combined with the considerable “capital investment” and “operating expenses in meter testing and maintenance and frequent meter reading” required for metering generally made metered pricing of these services uneconomical, except where doing so would result in “savings in pumping and in water waste inspection” (Wilcox 1931: 67). Accordingly, metering of water and sanitation, unlike electricity and gas, was uncommon. As of 1888, a substantial majority (1106 of 1668) of waterworks did not meter water use at all (Baker 1889: lxxxiii). In those communities that did use meters, metered connections, limited mostly to commercial and industrial customers, were a small percentage of all connections: Only 95 of the 562 works that used meters in 1888 had 100 or more meters in use (ibid.). Even in larger cities (with populations over 10,000), less than 14% of water connections were metered as of 1896 (see Table 5). Nonexcludability (resale) was also a problem for telephone service and was a factor in the eventual move from flat rate to metered pricing of phone calls (see John 2005).
30
As noted earlier, the prices a private company could charge were generally determined by its franchise contract, which, in later years, sometimes established municipal versions of public utility rate commissions with the power to approve rate requests (see Priest 1993: 321; Melosi 2000: 463, n. 31).
31
The difficulty of recovering capital costs was “all the more important in the case of a utility such as sewerage, where usually under public ownership no rates at all are charged for the service rendered” (Wilcox 1931: 26). More recently, a 1994 World Bank Development Report calculated that prices cover only about 30% of costs in the water sector in developing countries compared with over 50% in the power sector, 80% in the gas sector, and over 150% in the telecommunications sector (World Bank 1994: 47). Subsidies for water systems in these countries were estimated at $18 billion, $13 billion of which was attributed to “underpricing” and $5 billion to illegal connections (ibid., 121).
32
The adaptation advantage of usage-based pricing described here is similar to standard incentive arguments but recognizes that (a) unrealized gains from trade due to misaligned incentives present opportunities for reopening negotiations and (b) avoidance of renegotiation costs is an important objective of contract design.
33
Conversely, proposals to adjust inefficiently high or low service-based prices have a positive sum nature and are thus conducive to mutually advantageous modifications. For further discussion, see Williamson (1979: 251) and Crocker and Masten (1991: 79, 89). Contracting research emphasizing the role of price in reducing the incidence of post-agreement conflict includes Williamson (1983), Goldberg (1985), Goldberg and Erickson (1987), Masten (1988), and Klein (1992, 1996). Recent theoretical models in this vein include Baker et al. (2009) and Hart (2009). For empirical evidence on the relation between pricing and post-agreement conflict, see Crocker and Masten (1991), Oyer (2004), and Masten (2009).
34
Note that the prediction here differs from that of Spiller and Savedoff (1999) (see footnote 16): According to the argument of Spiller and Savedoff, the need for future investment attenuates appropriation hazards and thereby the need for public ownership, whereas the argument here is that the need for uncontracted for extensions and improvements (unprogrammed adaptations, in Williamson's terminology [1971: 113; 1996: 17]) increases frictions between cities and private utilities and makes municipal ownership more likely.
35
The quotation is from an 1881 letter from the president of Spring Valley Water Works to the San Francisco Board of Supervisors, warning of a reduction in its expenditures in response to a threatened rate cut, as quoted by Jacobson (1989: 19).
36
Although telephone and electrical lines were sometimes buried, delivery of electricity and phone service primarily entailed installation of aboveground poles and wiring. Gas lines, like water and sewer lines, were installed below ground, however. Sewer franchises were rare, but the few instances where sewerage services were privately supplied illustrate the extent to which cities were concerned with the potential for disruption. Among the provisions of a 1906 contract between the city of Long Branch, New Jersey (population 8872), and the Long Branch Sewer Company were requirements that the utility not “open any street or alley to a greater extent than 500 feet at any one time” and return the street to its original condition and that no excavation be left open “for any unreasonable length of time” (Wilcox 1910: 455–6). Similarly, the contract governing sewer services for Austin, Texas (population 22,258), required the company to “‘exercise due care and diligence in unnecessary obstruction to public travel’ on the streets, or as to ‘any injury or unnecessary interference with any pipes either of gas or water which may be lawfully located beneath the surface thereof when such sewers are laid.’ The company was also to take ‘every reasonable precaution against accident and danger to persons or property in the execution of the rights and privileges’ granted, and was to ‘cause all excavations to be properly lighted and guarded at night’, and was to restore the streets ‘to their former condition as near as may be without unnecessary delay’” (ibid., 457–8).
37
“Mains and Subways,” New York Evening Post, reprinted in Engineering Record, 41 (June 23, 1900: 601), as quoted by Rosen (1986: 37–8).
38
Compare Blake (1956: 77; as cited in Cain 1977: 139): “[T]he actual and prospective profits of the companies were rarely great enough to induce the directors to build systems adequate to provide all needs. The companies laid their pipes through the districts that promised the largest returns and left the poorer and more remote districts without a supply.” See also Anderson (1984: 220, citing Baker 1898): “Typically, private companies concentrated on the most profitable areas of a city, ignoring its outlying districts or poorer sections.”
39
Net density measures an “activity divided by the area devoted to that specific activity” (Kim 2002: 1), whereas gross density measures the activity relative to total area. Kim observes that “Net density … is preferred but is practically impossible to calculate” (2002: 1). Compare Anas et al. (1998: 1438). As discussed below, a feature of the present study is the availability of an appealing measure of net density, namely, the number of connections per mile of water mains.
40
As reported in Tables 4 and 6, 26 (8%) of the 346 cities in the data were served by more than one waterworks. Of these, 21 had 2 works, 4 had 3 works, and 1 (Brooklyn) had 7 works.
41
The number of hydrants, although not particularly expensive in themselves, may capture the cost of the extra capacity needed to accommodate the exceptional volumes and pressure necessary for firefighting (see footnote 27). The number of fire hydrants might also reflect the severity of fire hazards and, inasmuch as fire hydrant rentals were a primary mechanism through which cities subsidized private water suppliers, could correlate with the size of transfers that were reportedly a frequent source of friction between cities and private water companies.
42
Kim (2007) calculates density gradients, which measure the rate at which population density declines as one moves away from the city center, for 87 metropolitan areas from 1890 to 2000. In addition to covering a smaller set of cities than analyzed here, Kim's analysis uses 1950 metropolitan area definitions, which represent much larger areas than are of concern here. See Kim (2007: 281). The limitations of gross population density can also be seen in the sometimes dramatic changes in measured population density that have occurred over time with changes in city boundaries. The population density of New York City, for example, dropped from 29,933 to just 9574 persons per square mile between 1880 and 1900 following annexations increasing its area from 40.3 to 359 square miles (Shammas 2000: 509). Similarly, Baltimore's density fell from 26,585 to 14,710 per square mile, despite population growth of more than 100,000, as a result of land acquisitions between 1880 and 1890, while Boston's density in 1890 was 4842 persons per square mile lower than its 1870 density of 20,781 owing to land acquisitions that more than doubled its size from 15 to 37.8 square miles (ibid.). Such changes in land area were common during this period. According to Kim (2007: 278, Table 2), the land area of 119 of the largest US cities (with populations over 25,000) increased by 44% between 1890 and 1910.
43
To illustrate the relationships just described, imagine a city whose territory consists of a line segment of length A. Suppose the city has R residents located distance d from each other, where d·R < A. The length of water mains needed to provide each resident with water service is then M = d·R, and connections per length of mains will be R/M = d. Holding R/M constant and increasing population density, R/A, implies that M/A, the ratio of mains to city “area,” must increase. In other words, increasing population density while holding constant connections per length of mains implies both (a) the distribution of residents over more of the city's area and (b) a higher proportion of the area served by mains. Note that, to the extent that public waterworks served low-density areas that private works would not, taps per mile would tend to be lower for public works than for private works, which works against the relational frictions hypothesis.
44
This works was once again taken over by the city in 1907 (Wilcox 1910: 436–7).
45
The earlier average construction dates for works that changed ownership could simply reflect that the older a system is the more likely it is to have changed ownership even if such changes occurred randomly. A corollary to this is that attributes of waterworks in the private-to-public and public-to-private categories will reflect to some degree that the largest cities tended to be the earliest to construct water systems and therefore were most likely to have experienced a change in ownership.
46
Z statistics on coefficients are calculated using robust standard errors clustered on the state in which the waterworks is located. Potential endogeneity of the explanatory variables is discussed and instrumental variable estimates reported in Section 5.2.3.
47
System number is included in specifications containing city characteristics to control for the fact that waterworks operating in cities with multiple works serve only a portion of the city area and population. Because multiple public works within a given city is impossible, every city with multiple works has at least one private works, which likely accounts for the negative coefficient on this variable. Excluding system number or limiting observations to cities with only one waterworks leaves results essentially unchanged.
48
System number is excluded because only waterworks characteristics are included in this specification. When included, system number is significantly negative and other results are not materially affected.
49
At some level, of course, even a city's population may be endogenous to the existence and quality of a waterworks. See, for example, Melosi (2000: 119): “The push for municipal ownership … had as much to do with the desire to influence the growth of cities as to settle disputes with private companies over specific deficiencies.”
50
All estimations reported in Table 9 use the ivprobit command in Stata 10 (StataCorp 2007).
51
I have omitted from Panel B the results from the first-stage regression for mains corresponding to specification (7) of Panel A, which differs from the regression in the fourth column of Panel B by only two observations and yields essentially identical results.
52
The demographic instruments are highly correlated with each other. Their coefficients in the taps-per-mile regressions are often highly significant when the variables are included individually or in subsets.
53
The Wald test of exogeneity tests the null hypothesis that the error terms of the structural and instrumental variables equations are uncorrelated. See Wooldridge (2002: 472–7). When more than one variable is instrumented, it tests the exogeneity of all instrumented variables jointly. The failure to reject the null of exogeneity in specifications (4)–(7) likely reflects a lack of endogeneity of mains, hydrants, and mains per square mile: Controlling for population and city area, none of the three differ significantly for municipal versus private works.
54
The coefficient on January average low temperature was significant at the 0.10 level in the reduced-form regression and in specification (7) of Table 8. In addition to being clearly exogenous, climate is plausibly related, as discussed below, to relational frictions and investment levels. Results with and without this variable are virtually identical.
55
The two-step estimator yields consistent estimates of the parameters of the second-stage probit equation only up to scale. For discussions, see Rivers and Vuong (1988) and Wooldridge (2002: 472–7).
56
The estimated marginal effects for miles of mains per square mile range from −0.015 to −0.027 (from specifications (5) and (7) of Table 8, respectively).
57
This interpretation is also consistent with the earlier account of the San Francisco experience in which increased hydrants rentals were exchanged for “investments in system extensions and pipe enlargement for fire protection” (Jacobson 1989: 18; emphasis added). These alternatives are, of course, not mutually exclusive, and both could have played a role. Nor do they not exhaust the possible explanations. It is conceivable, for example, that local politics varied with the demographic variables in the analysis in a way that yields the observed ownership patterns. I have seen nothing in the historical record, however, to suggest such a relation.
58
Although never significant in the reported results, the effect of pump capacity occasionally achieved significance (at the 0.10 level) in some unreported regressions.
59
As noted above, the number of fire hydrants could also proxy for the severity of fire hazards.
60
Although the coefficient on January average low temperature is significant at the 0.05 level only twice, it is negative and significant at the 0.10 level in 7 of 11 specifications in Figure 3 and Tables 8 and 9.
61
For an analysis of the relation between waterworks and sewer systems more generally, see Cain (1977).
62
Note that freezing potential is one factor distinguishing water lines from gas pipes.
63
On the relation between climate and typhoid rates, see Troesken (1999: 928, 942). On climate and cholera, see Pascual et al. (2002). The connection between climate and disease had already been made by the late 1800s: “Records afford evidence of an undoubted relation between the meteorology of a place and its liability to cholera activity” (H.W. Bellew 1884, as quoted in Pascual et al. 2002: 237). As note above (footnote 14), the connection between disease and water, as opposed to climate, was less widely appreciated.
64
On urbanization and its effects on population density over 19th and early 20th centuries, see Kim (2000, 2002, 2007).
65
Of the 248 observations in the specification reported in column (1) of Table 10, 58% were subject to state borrowing restrictions, 17% to state borrowing limits, and 53% to local borrowing restrictions; 23% were subject to no borrowing restrictions. As Wallis and Weingast make clear, such categorizations mask much variation in the details of the restrictions both between states and within states over time. Nevertheless, they provide some indication of the borrowing environment in states at the time works were constructed.
66
The coefficients on state borrowing limits and restrictions are not significantly different from each other, and a single variable indicating either limits or restrictions is significant at the 0.05 level. In addition, all three coefficients are significant at the 0.05 level when original ownership, instead of 1890 ownership, is the dependent variable.
67
Note that the use of fixed effects eliminates 14 states (26 observations) in which all works in the sample are either municipal or private. The loss of significance of the borrowing restrictions variables in column (3) is not due to the reduction in observations; both state restriction coefficients are significant at the 0.05 level and the local restriction coefficient at the 0.10 level when the specification in column (3) is reestimated with the same observations but without the fixed effects.
68
Again, the role of the institutional environment in governance choices is beyond the scope of this study, and interpretations of the estimated effects of state-level restrictions are at best tentative without further investigation. Wallis and Weingast present evidence, for example, that state borrowing restrictions shifted borrowing for infrastructure from the state to local level (2008: 347–55). Such increased borrowing might have raised the cost of capital to cities, thereby “crowding out” municipal borrowing for waterworks in states with such restrictions. Wallis and Weingast also hypothesize that state-imposed restrictions on local borrowing may have served to enhance the credibility of city obligations and allowed cities to borrow on better terms: “[S]tringent [local] debt restrictions, particularly debt restrictions that prohibit special purpose government from gaining access to general government revenues aid the promotion of infrastructure investment and raising money in the capital market” (ibid., 359). This would be consistent with the otherwise paradoxical finding that municipal ownership of waterworks was more likely in states with local borrowing restrictions. It should also be noted that, in analyzing the potential role of state-level institutions, the endogeneity of state laws needs also to be taken into account, as Wallis and Weingast make clear.
69
There has also been growing interest in “outsourcing” water system management (though typically not ownership) in the United States and other developed countries. See, for example, National Research Council Committee on Privatization of Water Services in the United States (2002). Wilcox was skeptical of the prospects for hybrid arrangements: “Certain important difficulties arise where ownership and operation represent diverse interests. The maintenance, replacement, improvement and extension of urban utility facilities is almost inextricably bound up with operation. The care of property is part of operation. So when a municipal utility is leased to a private operating company, great prudence must be shown in drawing up the lease, or a great deal of intimate supervision undertaken, to insure the upkeep of the plant and integrity of the investment” (1931: 35). For a comparative analysis of public and private contracts with an emphasis on water systems that is largely compatible with the analysis here, see Spiller (2008).