The first major application of punched cards—processing information from the 1890 United States census—was hailed on the front page of Scientific American and praised in other contemporary science and technology publications as a great advance over earlier, manual methods of processing and as a manifestation of American efficiency and technical ingenuity.1 This view that the invention of punched cards was a testament to America’s being in the forefront of technological inventions was also expressed in later, technical books on census processing and the punched-card industry.2 However, these later authors were operating with hindsight, as they had experienced the subsequent success of punched cards, in various applications, using substantially improved equipment.
By contrast, social historian Margo J. Anderson has shown that the introduction of punched cards to process census returns grew out of public demand for more and better census data, combined with Congress’s hesitation to establish a permanent census office. The latter decision resulted in inadequate and sporadic funding, and a frequent turnover in administrative management.3 This situation triggered the development of the first punched-card system to process population census returns.
This chapter extends Anderson’s insights by examining how returns in the United States were processed in successive Census Offices. These offices provided the institutional setting for the invention and development of the original punched-card system to process the census in 1890.
The United States decennial census was one of the compromises resulting from the Constitution of 1787. The census was to create a basis for apportioning seats among the states in the federal House of Representatives and for sharing the burden of paying direct taxes to the federal government. The fathers of the Constitution did not invent the concept of a census, but the United States was the first country to introduce regular population counts.4
The front page of Scientific American on 30 August 1890, depicting the technical success of Herman Hollerith’s punched-card processing of the U.S. census. (Scientific American 30 August 1890: 127)
The Constitution prescribed that all people in the Union be counted every ten years, and the first census took place in 1790. The Constitution required an enumeration of the number of free persons and the number of slaves.5 From the first census, however, more than the constitutionally minimum information was collected, and the objective of the censuses became to obtain accurate data on the military and industrial strength of the country. After the Civil War (1861–1865), the Constitution’s racial distinction was amended so that all people had to be counted on an equal footing.6 In practice, however, the racial distinction was upheld through elaborate statistical reports in the census publications.
During the nation’s first century, the task of processing the censuses grew in scale and particularly in scope. The increasing population yielded a larger scale effort, as Americans grew from 3.9 million people in 1790 to 76 million in 1900 (a factor of 19). This expansion was surpassed by the simultaneous extension of the scope of the endeavor, created by the burgeoning ambitions of politicians, bureaucrats, and statisticians. The outcome was a dramatic increase in the size of the censuses’ statistical reports, which grew from fifty-six pages in 1790 to 26,408 pages in 1890 (a factor of 472).7
This growth in scale and scope was a product of the absence of a permanent census organization. A new office was established for each census, only to be closed down when it had completed its assignments or ran out of funding. All these factors caused great variations from one census to the next, which can be substantiated through the lack of continuity in the position as head of the successive Census Offices and the substantial variation census publications. During the first twelve censuses, from 1790 to 1900, only twice did same person head two census operations.8
In the same period, U.S. census publications varied much more than in comparable countries with permanent census organizations. However, as the United States had no permanent census office, many statisticians were recruited for each census, even though their main affiliation remained elsewhere. Therefore a large part of the statistics community in the United States came to work on the censuses, which promoted a network of statisticians.
In the nineteenth century, no leader of the census operations, who considered himself a statistician, had any academic training in statistics. Even in the 1880s, few American statisticians had formal instruction before they entered the field. In 1887, statistics was only taught at three universities in the United States: Johns Hopkins (in Baltimore), Massachusetts Institute of Technology (in Cambridge, Massachusetts), and Columbia College (now Columbia University in New York). Moreover, no federal or state statistical institution offered systematic instruction.9 Professional statisticians only joined the census office in the 1930s.10 Therefore, the collection and processing of information for earlier censuses lacked both the organized transfer of experience from one census to the next and trained personnel.
From the 1840s onward, interest groups tried to establish a permanent census office. The most vigorous argument against a permanent office was that it would require eternal federal appropriations. However, permanent statistical departments existed in many other governmental bodies and in several states in the late nineteenth century, and every nation state in Europe had a permanent organization responsible for census taking. Within the U.S. federal government, the Department of the Treasury compiled foreign trade statistics as early as 1789, and in 1866 a permanent office was established for this purpose. From 1862, the Department of Agriculture compiled and tabulated agricultural statistics. From 1884, the Bureau of Labor in the Department of the Interior kept labor statistics. From 1891, the Federal Bureau of Immigration in the treasury compiled immigration statistics.11
All these federal institutions published statistics regularly—either yearly or more frequently. Further, several states of the United States established their own statistics offices, such as Massachusetts (1869), Pennsylvania (1872), Connecticut (1873), and Ohio (1877).12 In addition, the American Statistical Association had been founded in 1839 as a professional body. Its activities included publications, a library, and seminars.13 European states had established central statistical offices responsible for census operations much earlier, for example, France (1799), Prussia (1810), England (1837), and Austria-Hungary (1863).14
Processing Census Schedules
Two basic changes in producing U.S. census statistics took place in 1850: the individual became the analytic unit and processing returns became centralized at Washington, D.C. From 1790 to 1840, the household had been the key unit and the individual districts had tabulated the results. The 1850 changes can be traced to criticism about inaccuracies in the 1840 census.15 Introducing the individual as the census unit allowed the generation of more detailed statistics. Centralized processing could provide more uniform and, therefore, more exact outcomes than did processing in the hundreds of enumeration districts. These two changes reflected a growing general interest in detailed and reliable statistics, and the same changes took place in Great Britain, France, and Germany between 1840 and 1900.
In the United States, these changes in the way the census was handled caused the number of units to be processed to rise by a factor of 7.5 compared with the previous count. In 1840, data on 3.1 million families had been tabulated, whereas information on 23 million individuals was processed in 1850.16 This made processing a critical problem that became conspicuous as a result of the simultaneous centralization of processing in Washington. Several alternatives existed that could solve this problem: reducing the task to the minimal constitutional requirement (a head-count, including slaves), employing additional clerks, or relying on technical aids.
Reducing the scope was never tried; on the contrary, the census continued to grow. From 1850 to 1890, the number of inhabitants rose by almost threefold, from 23 million to 63 million. Yet at the same time, the administrators’ ambitions relating to the statistics grew, as expressed in the size of the double-entry tables published. This can be observed from the tables with race and sex as one entry and age as the other. In 1850, this table had sixty entries. In 1890, there were 1,696 entries, a massive increase.17 (In 1850, slaves were recorded separately.)
A growing number of clerks were employed. However, this did not solve the problem because of the sporadic nature of the census operations caused by the congressional requirement of separate appropriations for census operations each year. By the end of the nineteenth century, T. C. Martin reported claims that census returns were about to reach a number that would prevent their processing by traditional means—before the following census, ten years later.18 Regardless, any census can be processed manually. It is only a question of obtaining adequate appropriations and creating an appropriate-size organization.
Therefore, the main reason for introducing technical aids was the lack of a permanent census office combined with the growing desire for statistics. In contrast, the British and German statistical offices processed their censuses by hand until 1910. In 1872 and 1890, the census offices introduced mechanical aids for processing the United States census, first the Seaton device and then Hollerith’s first punched-card system.
Census statistics were compiled from returned, completed census forms. Until 1840, enumerators recorded one row for each family, with columns for entering the number of persons of specified categories, such as white male from thirty to thirty-nine years of age. The totals for one schedule page could be obtained by adding the columns and writing the result on the form. Clerks transferred the totals to summary or consolidation sheets. Further additions gave totals for each census district, which were then transmitted to the census office to be aggregated and published.19
The 1850 census form had one row for each individual in a family. Each row had a number of columns indicating the various items of information wanted. Some questions, such as name or occupation, were answered by a name or a noun. The age called for a figure. The remainder, for example, questions on civil conditions and education, could be answered by making a checkmark in the proper position.
As each census form held information on several persons, there were two methods of tabulation. The first method was to have each table summarize the information in one row on a new form, a tallying list, repeating the process until the desired aggregation was achieved. This was a simple, tested method that already had functioned well in processing many censuses. It had the disadvantage, however, that tabulating each table required a separate handling of all the census forms. And, as the number of table entries grew, either the tallying sheets became unwieldy or compiling the table had to be split into sections; both results required separate handling of the census forms.
The second method was to begin processing by copying the information on each individual to a card or slip and compiling all the tables by sorting and counting the cards.20 Here, the initial copying of the schedules gave no immediate benefit, but it eased the subsequent compilation of tables very much. The tallying method was applied in the censuses from 1850 to 1880, and the card method was used from 1890.
Although a substantial leap in the number of statistical units took place at the 1850 census, tabulating the returns required no basic change in the method or technology applied. The same held for processing the 1860 census. The first attempt to introduce new technology in processing the information was made in 1870 by Charles W. Seaton, a manager in the 1870 Census Office. His idea was to improve the way a clerk organized the tallying lists on his or her table. For this purpose, Seaton built a relatively simple mechanical device out of wood. It had parallel rollers by which a roll of paper was unwound so that a number of tallying columns were placed side by side.21 It seems that it enabled tabulation of 160 entries at a time for one or more tables.
Charles W. Seaton’s device for tabulation. This model is from 1880. The drawing is from 1903, when this copy was owed by Herman Hollerith. (W. R. Merriam, “The Evolution of American Census-Taking,” Century Magazine 65 , 836)
The main difference between hand tallying and the Seaton device was that the latter allowed the simultaneous tabulation of up to eight tables or table sections.22 The 1870 census office assessed the Seaton device as a significant improvement over older methods. On the basis of this appraisal, Congress awarded Seaton a sum of money corresponding to about twenty-nine years of a clerk’s salary, which was a substantial congressional reward for solving problems of census processing in the United States.23
The Seaton device was, however, a mere extension of existing manual methods and had limited potential. It reduced the size of tallying sheets that a clerk managed, but a major source of error remained, as the operator still had to oversee the same number of entries.
Several of these devices were used during the compilation of the 1880 census.24 The number of entries in the table having race and sex as one entry and age as the other was comparable to the number a decade earlier, so were the advantages of using the Seaton roller. For the 1890 census, a doubling of entries was planned. This would have been difficult to organize by use of either manual methods or the Seaton device. However, the decision to double the entries for the 1890 census was based on knowledge of the Massachusetts state census’ successful application of single handwritten cards in 1885, and of Herman Hollerith’s successful testing of his first punched-card system on a grand scale.
Systems for Processing the 1890 Census
The Hollerith punched-card system introduced for processing the 1890 census represented a momentous change. While Seaton’s machine had organized the tallying lists, the punched-card system broke the tallying lists into single cards and comprised complex mechanical artifacts. These could not yet be described as machines, as they were exclusively operated by hand and had no moving parts. At the 1890 census, punched cards became the cornerstone of a technological system consisting of the producer (Hollerith), the machine shops building his hardware, the printing shops supplying the cardboard cards, and the Census Office as the only user of the technology.
This new technological system derived from three basic principles: representation of the information of each census unit on a single card or slip, mechanical tabulation, and electric card reading. John S. Billings conceived the first two in 1880 and was an experienced manager and statistician in the 1880 Census Office.25 He mentioned the idea to Herman Hollerith, also a staff member, who developed this idea during the remainder of the 1880s.
Billings had introduced the representation of a census unit on a card for processing the 26,000 deaths reported in an 1880 investigation.26 The 1885 Massachusetts state census also used single cards.27 Further, mechanizing production processes was a principal feature of that age. One historian has drawn attention to the fact that, on several notable occasions, academics stimulated the problem choice of independent inventors.28 This is another example of that stimulation, although Billings was trained as a medical doctor and had received no formal education in statistics. His ideas were based on his experience with library card files and fatality statistics, not on his formal education. Herman Hollerith adopted the electrical card reading in the innovation process.
The superintendent of the 1890 census chose Hollerith’s first punched-card system for processing the census after a competition in September 1889. Three systems sought the contract: Hollerith’s and two others presented by the Massachusetts statisticians Charles F. Pidgin and William C. Hunt. The task was to transcribe accurately and tabulate as quickly as possible the data on 10,491 people collected in the preceding census. The major argument for introducing Hollerith’s punched-card system was its speed of processing and the consequent savings in labor.29 Punched cards also offered more possibilities in terms of tallying, but a written card of a similar size could hold more information. For example, a figure of several digits could be used to represent a criterion. Representing multiple digit figures as holes required more space than writing by hand.
The competition and the three systems presented indicate the scope of census-processing improvements within statistical thinking in the United States in the late 1880s. Pidgin and Hunt designed their systems based on their experiences in the Massachusetts state census in 1885, where the returns were processed by use of one card on each inhabitant. Both Pidgin and Hunt used cards sorted by hand, and they only differed in the way the census data were transcribed. Pidgin used specially designed cards, printed in different colors, to facilitate transcribing, sorting, and counting. Hunt transcribed the data onto slips of paper by use of various colored inks.30 Really, the three systems offered were very similar. All were based on the unit card concept. The difference lay in the way the data were represented on the cards—and in Hollerith’s mechanized counting. Furthermore, the systems’ originators, Billings, Hollerith, Hunt, and Pidgin, were all members of the statistics community and connected to the federal and the Massachusetts state censuses.
In the competition only speed and accuracy mattered, while the costs of acquiring the method or system chosen did not count. Five hundred eighty thousand dollars in wages was estimated to be saved using punched cards.31 The costs of using the proposed systems or methods were only calculated for Hollerith’s punched-card system, subsequent to the test on speed and accuracy. Hollerith was paid $230,390, or about 40 percent of the estimated the wage savings.32 The total cost of the 1890 census was $5.8 million, 143 percent more than the 1880 census in real dollars. This was far more than the population growth during the decade of 25.5 percent, which leaves the increasing ambition regarding statistical details as the main reason for the higher costs.33
The layout of the punched card for the United States census in 1890 with the abbreviations in each position. Columns 1–4 were used to indicate the number of the enumerations district. The remaining positions were used for coding the information of an individual. (Leon E. Truesdell, The Development of Punched Card Tabulation in the Bureau of the Census 1890–1940, Washington, DC: Government Printing Office, 1965, 47)
The 1890 punched-card system consisted of three components: punched card, punch, and tabulator.34 The card, made of thin cardboard measuring 6⅝ by 3¼ inches (16.8 × 8.3 cm), featured twenty-four columns having twelve punching positions each. These positions could record all the information returned in the census on one individual, organized in a compact layout specially designed for the information collected in the 1890 census.
Gathering census information on the population involved completing one form per household, with one row of data for each member, compiling age, sex, marital status, place of birth, occupation, and so on. First, the clerk punched the information on each individual onto a card by use of a pantograph punch (described below), and, at the same time, handwrote a serial number on the card, enabling verification of the punched information against the census list.
A pantograph was a device invented in the seventeenth century for replicating drawings. The pantograph punch used a similar method to replicate items punched on a perforated board, which contained abbreviations noting the positions for entering each item of data, onto cardboard cards. Over the punch board swung an arm, rather like that of a record player, with a needlelike “finger” at the end. In the manner of a pantograph, this finger’s movement was followed along at the rear of the machine by a punch device. Hence, when the finger was pressed down into a depression in the plate in front indicating a punching position, the punch at the back created a hole in the cardboard card. The arm the finger was attached to worked as a lever and, thus, lessened the amount of pressure required for each punch operation.
The tabulator, with its electric card reading capabilities, was Herman Hollerith’s main contribution to the punched-card machine. It had three main parts: the press, the counters, and the sorting box. The manually operated electric circuit closing press had a hard rubber plate containing 288 holes, or pockets, each corresponding to the full set of the card’s punching positions. Each pocket was partly filled with mercury. The card to be read was placed between a series of spring-loaded pins and the rubber plate with the mercury cups. In each position containing a hole, the corresponding pin made contact with the mercury and closed an electric circuit that included the counter. Activated by an electromagnet, the counter moved ahead one unit. It displayed a hundred divisions and had two hands, like a watch. The big hand counted units, the small hand hundreds, allowing counting up to 9,999. Before reaching that figure in a tabulation, the operator read the counters, noted the figure on a sheet, and reset the device.35 An operator could read up to about forty cards a minute.36
The third part of the tabulator was a sorting box, easing manual sorting of cards for a subsequent tabulation. The sorting box was divided into compartments, each closed by a hinged lid. The lid, which would be opened by a spring, was normally kept closed by a pin. The pin could be released by an electromagnet connected by a wire to the press. For example, in a classification by sex, race, and marital status, it was possible, at the same time, to sort the cards in age groups in preparation for a following count of this distribution. When the operator closed the press, the lid of the compartment containing the relevant age group opened. The operator placed the card in the open bin, closed the lid, and was ready for the next card. If there was no age indication on the card, no lid opened and an error was indicated. In this way, sorting became a part of the counting process. When sorting could not be combined with a counting operation, it was easier to sort by hand than by the sorting box.
Cross-section of the tabulator’s circuit-closing press, which was Hollerith’s first card-reading device and was built for processing the U.S. census in 1890. Herman Hollerith, “An Electric Tabulating System,” School of Mines Quarterly 10 : 302)
The total population figure was tabulated in two separate ways. First, clerks took a gross count based on the census forms that each contained information on a household. A special keyboard facilitated this work; it had keys numbered from one to twenty, as well as twenty-one counters on the tabulator. The operator took a form and struck the key indicating the number of persons in the household. This actuated the counter of that number, and the twenty-first counter recorded the number of households. In this way, households with up to twenty members could be processed mechanically. A few exceeded this and were noted down by hand. Finding the total population number simply involved multiplying the number of households in each group by the number of persons per family. This quick gross population figure was produced to check the subsequent punched-card calculation. Once the accuracy of the subsequent punched-card processing had been established, this special keyboard was discarded.37
When the total population figure of nearly 63 million was announced, some people who believed the Union to have at least 75 million inhabitants criticized the punched-card method as being inaccurate.38 As the calculation had been performed twice, first directly from the census forms and then using punched cards, this figure hardly contained significant inaccuracies. Punching was not verified, but clerks checked small samples to monitor the key punch operators.39 In statistics processing, punched cards were hardly ever verified simply because of the cost; verification required nearly as many hours of work as the punching operation.
The individual, written identification number provided a safety precaution that enabled each punched card to be checked against the original entries on its census form. The 1900 census also used the individual identification number, but then it was dropped.40 This indicates that, like the original gross count, the individual identification number had been a safety precaution designed to ease the acceptance of the punched cards. The adjustment of the tabulator for each count—“programming” in today’s usage— was done by making wire connections using screws between the relevant contact points on the counters, circuit closing press, and sorting box.41
Born in Buffalo, New York, in 1860, Herman Hollerith moved to New York with his family in 1869.42 At that time, American industrialization was gaining force and speed, which meant opportunities for hundreds of thousands of workers. The industrial cities grew rapidly. Nine years before Hollerith moved to New York, the city had 1.17 million inhabitants. The year after his arrival, it held 1.48 million.43 The city’s noise and its immense construction work proved both noticeable and attractive to a teenage boy. Industrialization called for skills in devising, designing, and managing technologies.44
At the age of sixteen, Herman Hollerith began studying mechanical engineering at the School of Mines, in today’s Columbia University, founded only in 1864. In the 1870s, the school offered three curricula: mining and mechanical and civil engineering.45 Hollerith graduated at the age of nineteen, an average student who met the requirements solidly but without class honors. Education at Columbia laid great emphasis on school subjects, and less on workshop experience, a significant shaping factor in Hollerith’s career.46 Later, his inventions would apply electricity as a prominent feature, but he attended no classes in electricity. Electrical engineering was only born with Thomas A. Edison’s invention of the incandescent lamp and the installation of the Pearl Street electric power system in New York in 1882. Electrical engineering at Columbia appeared in 1889, ten years after Hollerith’s 1879 graduation.47
Professor William P. Trowbrige of the Columbia School of Mines, already a special agent at the Census Office, recruited Hollerith for the coming census. Hollerith joined another student and several instructors to perform this short-lived task. The professional census employees he encountered offered this young graduate the possibility to get acquainted with various members of the network who would shortly be dispersed across the country. Hollerith’s assignment in the 1880 Census Office was to work on statistics of manufacturing.48 By late 1881, the Census Office started to downsize because of a funding shortage.49 In 1882, through the census network, Hollerith took a job as an instructor in mechanical engineering at the Massachusetts Institute of Technology (MIT).50 Hollerith left the following year to become an assistant patent examiner in the United States Patent Office in Washington, D.C. This shift could have had several reasons behind it: Hollerith’s choleric temperament was hardly compatible with teaching; he wanted to live near Washington, D.C., and he earned more in the Patent Office.51 The shift could also be related to his activity as an inventor, a vocation he had pursued only in his spare time. By the early 1880s, his first statistics machine was approaching the patent phase. In 1884, Hollerith resigned from the Patent Office and six months later filed a patent application on his statistical machine, thereby embarking on his main career as an independent inventor and entrepreneur, which lasted until 1911. This profession lay outside the census network, which Hollerith left temporarily. A few years later Hollerith would return to the network to seek “employment” for his invention.
Thus Hollerith’s choice to become an independent inventor and his subsequent accomplishments placed him in the company of the heroes of the U.S. age of invention. In 1876 Alexander Graham Bell invented his telephone and Thomas Alva Edison opened his Menlo Park laboratory. An average of 20,260 patents applications were filed a year in the 1870s. In the 1880s this figure rose to 32,277.52 At that time, there was no graduate training in engineering but Hollerith’s employment in the census of 1880 and at the Patent Office had fulfilled this role in his career. At the census he collected and processed statistics on power and machinery used in manufacture. In the Patent Office, he learned about all varieties of machinery components.
The basic problem in the 1880 census had been the gigantic task of tabulating the population. During the 1880s, Hollerith developed John S. Billing’s original idea and several times discussed his suggestions and constructions with him. The census office management was also keen to devise a way to mechanize tabulating population statistics, but Hollerith received no financial support. However, he did learn the duties of a clerk during his working hours through a temporary reassignment, and he actually operated the Seaton device.53
Hollerith’s original tabulation system used a continuous roll of paper, somewhat like Seaton’s, and consisted of a punch and a reader. Two lines of holes across the width of the paper made by a primitive hand punch represented the record for each individual. Holes in successive lines were staggered not to weaken the paper. The blank paper was fed from one roll to another, running under a metal template with a hole in each punch position through which the hand punch made perforations to provide the record. On the reader, the perforated roll of paper ran between a metal cylinder and one pin or pointer for every punching position across the paper. A hole in the paper enabled a pin to touch the metallic cylinder and an electric circuit was closed that, by use of an electromagnet, moved a simple mechanical counter.54
Although Billings provided the ideas for punch representation and mechanization, Hollerith developed the actual embodiment of these ideas. Though he had received no college training in electricity, he chose basic electromechanical technology, which was well-publicized through books and periodicals.55 In the machine’s first version a paper strip replaced Billings’ single cards. The electric telegraph transmitter, the Seaton device, or the Jacquard loom could all have been the inspiration, as all of these used strips. The French punch-strip-controlled loom was developed during the eighteenth century, and Joseph Jacquard built the definitive design in 1804. This loom was widely used in the United States in the 1880s.56 Another consideration might have been that it was more difficult to read single cards than a strip, and that Hollerith’s strip-reading construction allowed a mechanized feed.
Herman Hollerith’s first statistics processing system used punched paper tapes. The top drawing show details of his punch, and the two lower drawings depict his tabulator. (Herman Hollerith, U.S. Patent 395,783, filed and issued 1889. Reference numbers deleted.)
The basic electromechanical technology provided simple construction as well as programming flexibility. It is often simpler to construct a prototype device using electromechanical technology than to use an exclusively mechanical solution. An example is the transmission of information from reading of a hole to the counter. In mechanical technology a physical connection would be established, for instance, by use of levers—as in traditional typewriters. Hollerith just used electric wires to establish the desired connections. If different connections between holes and counters were needed for different statistical compilations, it would have been easier to change wiring than to rearrange levers. The simple technical construction enabled Hollerith to bypass his limited workshop experience.
The price of programming flexibility was that an operator had to check all electrical connections for each program or machine setup. This problem continued to plague the electromechanical punched-card machines for as long as they were used. Initially, Hollerith solved this problem by doing all the programming himself.
Hollerith resigned his job as patent examiner in March 1884 to devote his time to the innovation of his machines for the projected mid-decade state-level censuses in June 1885. A federal project was mounted to establish a national census in 1885 based on censuses in the many states using the 1880 schedules. The states were to be persuaded through receiving a federal grant-in-aid if they deposited a copy of their return. Federal tabulations would be based on this material.57 In addition, Hollerith aimed at applying his system to finish the tabulation of the 1880 census, which was still under way even if at a low key. In September 1884, Hollerith was ready to file a patent application. His brother-in-law loaned him the money to cover the expenses and to build a version of his census machine to try out on mass data.58 Soon after receiving the money, he filed to patent his invention.59 In any case, the plans for a mid-decade census vanished, and the office of the 1880 census was abolished in March 1885, yielding no income for a prospective supplier before 1890.60
The dwindling chances of census tabulations in 1885 caused Hollerith problems. Through a relative, he got a job as manager at a Missouri company producing railroad brakes, which he held until 1887. During those years, Hollerith also worked on a small keyboard-driven mechanical adding machine designed by Tolbert Lanston. From 1865 to 1887 Lanston had been a clerk in the Pension Office in Washington, D.C. Meanwhile, Lanston studied law and qualified as an attorney, while retaining his interest in mechanics and making several inventions in various fields.61
In the 1880s, adding machines were not common. The slide-based Thomas de Colmar calculators had been batch-produced since 1821, and about 1,500 examples had appeared by 1878.62 The first keyboard-driven adders emerged in 1850, but no big producer had yet been established. Burroughs only started production in 1885 and the next year followed Felt and Tarrant, who were known for their comptometers.63 Most adding machines were related to bookkeeping, but the Lanston machine originated from statistics.
Charles W. Seaton of the 1880 Census Office sponsored Lanston’s work on the adding machine in the early 1880s. The objective was to facilitate the compilation of statistics based on more than units, that is, on more than statistics requiring addition. This, and the simultaneous support for Hollerith’s early punched-card work, shows the importance attached by the Census Office, especially by Charles Seaton, for further mechanization of statistical production than the Seaton device offered.
In 1882, Hollerith tried a working model of the Lanston adding machine in Charles Seaton’s office. Shortly afterward Lanston lost interest in this calculator, as he became greatly occupied with machines for composing type and secured several patents in that field. In 1887, Lanston resigned from his clerical job and set up a company to develop and produce his machines for type composition. He then undertook the task of converting his patented idea into a practical machine for commercial use. For ten years he labored and, finally, in 1897 he launched a highly successful machine, the Lanston Monotype.
Hollerith interest in the Lanston adding machine grew. In 1884, while drawing up the first patent for his own census machine, he bought the rights to it. He went to the Pratt and Whitney Company of Hartford, Connecticut, to explore production options, but it proved too expensive.64 The Lanston episode demonstrated that Hollerith was already concerned with mechanizing the compilation of statistics requiring addition at the time he filed his first patent in 1884. In the office of the census from 1880 to 1882, he worked on statistics requiring addition. Machines for addition are more complex than those for counting. None of Hollerith’s inventions or innovations outside census processing was quickly commercialized. He returned to his census machine and, in 1887, moved back to New York.65
Even during his job at the railroad brake company, Hollerith tended his census machine, introducing two major changes. The first was to replace the paper strip with individual cards. This required a new card-reading device but enabled sorting. For this purpose, he added a sorting box. This transformation grew out of discussions with Billings. Major arguments in favor of the changes were the use of single cards in the Massachusetts State census of 1885, and the ability, in a simple way, to compile statistics on small social groups, like the 105,000 people of Chinese descent in the United States in 1880.66
The second innovation was the prototype of an adding tabulator.67 In some statistics only units are counted. This is the case for the greater part of population census tabulations. The remaining statistics require the addition of larger numbers than one. Hollerith’s first census system only could count, but he recognized the importance of machines that could add. Perhaps he devised the adding tabulator for the processing of the 1890 agricultural census, but the census first used adding tabulators ten years later.68
Hollerith’s innovations of the statistics system were not confined to his drawing board and workshop. As early as 1885, he sought returned and completed census forms to use in his developmental experiments. Experience processing actual returns could prove useful in getting his system accepted. Hollerith tried, in vain, to obtain contracts to process the 1885 Massachusetts state and Baltimore city censuses.69 In 1886, he used his system to tabulate mortality statistics for Baltimore and Jersey City.70 During these trials, punching the cards proved a weak point. Hollerith perforated the cards with a conductor’s punch and strained his arm. Further, the conductor’s tool could reach only two rows of punching positions along the edges. Therefore, Hollerith designed the lever-based pantograph punch, which was easy on the arm and enabled the operator to perforate the full card.71 Its design resembles an early kind of typewriter.72
In 1889, the Surgeon General of the Army applied Hollerith’s system for statistics on illness and the city of New York used it for mortality statistics.73 The army discontinued their application within a few years. With 27,000 military personnel on active duty in 1890, a system based on transcribing by hand the information from a report card for each incident to statistics sheets was entirely satisfactory. This system remained in place even as the number of personnel grew to 99,000 by 1914.74 The Surgeon General then revived punched-card processing of illness statistics, as the army planned to introduce draft service and an army of a million men in the First World War. The commercial contracts with the Surgeon General and with the city of New York contained a provision of long-standing importance to the Hollerith company, and later to IBM, as the tabulators were rented out, not sold. IBM continued this practice until they were compelled by an antitrust case in 1956 to sell tabulators. Two reasons can explain this practice.
The first reason was technical: To Hollerith, leasing the tabulators gave the advantage of assuring their maintenance, which was crucial. There had to be mercury in the card reader’s pockets, and all electrical circuits had to be wired correctly and to function properly.75 Electricity provided flexibility and simple constructions, but it also required checks and maintenance. The first Eastman Kodak standardized camera provides a contemporary example in which the company also took care of a crucial process: loading and taking out the film. Kodak’s camera was sold loaded with film for a hundred exposures. Once the owner took these pictures, he returned the camera to the Eastman factory, which developed the film, printed the pictures, and reloaded the camera.76
The other reason was financial: Renting the tabulators saved customers the full expense of buying the machines. This was important in the ad hoc Census Office, which only used these machines for a couple of years, as well as for the first application in the Office of the Surgeon General of the Army and in New York. Further, saving money was a major argument in the official report on the three systems or methods proposed to process the 1890 census.77 Cost considerations were not uncommon when new machines were introduced in the nineteenth century. For example, George Corliss accepted a percentage of the fuel saved as rent for his steam engines, and telephone rental to the customers funded Bell Telephone’s growth from the 1870s.78 While the financial reason originally appears to have been the most important, the technical reason of securing the maintenance of the machines later became fundamental to the famous IBM sales organization, as it involved regular visits to the customers by IBM personnel.
The rent for a tabulator was $1,000 a year, which probably was reasonable, as an urban clerk or wage earner then earned about $550 a year.79 At the same time, it was good business for Hollerith. Two decades later, when the tabulator was more highly developed and a separate mechanical sorter could be supplied, a tabulator plus a sorter cost about $1,200 to build, and the rent was $600 a year.80 Further Hollerith’s business and private accounts show that he had a substantial income during the processing of the 1890 census. Although the savings argument was important in 1889, Hollerith’s total income from his tabulators became a major objection to using his machines once the Census Bureau became permanent in 1902.
In the spring and summer of 1889, Hollerith still had no major customer, and only in December of that year would he gain the contract to process the population census for 1890. Therefore, he worked to find other customers, but his efforts were confined to public statistics on individuals. In June 1889, he demonstrated his first punched-card system machine before a convention of United States Labor Commissioners.81 Earlier that year, in April, he traveled to Europe and exhibited the machine in Berlin and Paris, which caught attention of census officials in Europe. In contrast, an approach from an insurance actuary in March 1889 led only to a demonstration the following year.
The Problem of U.S. Census Processing and the Shaping of the First Punched-Card System
The United States changed fundamentally during its first century. In 1776, there were thirteen states hugging the Atlantic coast and in 1790 a total population of 3.9 million. At the dawn of the twentieth century, the country stretched from the Atlantic to the Pacific and held 76 million inhabitants. The rising population and greater territory caused the scale of the censuses to grow and, simultaneously, their scope rose as well. The absence of a permanent census institution hindered the transmission of information from one census to the next, which in turn hampered controlling the censuses’ growing scope and operations. Every decade, the management had to reinvent how to organize collecting, processing, and publishing the information. Each census operation in the United States had an individual history in the nineteenth century. France, Germany, and Great Britain had established permanent statistics institutions for processing census returns. For example, in England about 20 percent of the staff was re-employed from previous censuses, thus transferring prior knowledge.82
The problem of organizing processing the census returns was only slowly recognized in the United States. A historian has used the concept of a reverse salient to uncover the process of recognizing a problem. He borrowed the concept from military historians, who delineated those sections of an advancing line or front that had fallen back as reverse salients. Having identified them, the strategists analyzed them as critical problems. The important distinction is between the existence of a reverse salient and the identification of the critical problems.83
This distinction is useful for analyzing the history of the censuses in United States. From 1790 to 1840, the reverse salients in population statistics were the statistical unit applied and the decentralized processing. These limited the statistics’ degree of detail and accuracy. Both introducing the inhabitant as the statistical unit and centralized processing in 1850 opened the way for more detailed statistics, but these steps made the processing the reverse salient. This problem might only have been identified in 1872, when the Seaton device was built and introduced. The Seaton device helped, but it was not a definitive answer. During the census operation in 1850, Charles W. Seaton looked for better aids, and he later encouraged Herman Hollerith’s endeavor to develop John Shaw Billings’ idea of mechanizing the tabulation based upon a single card representation of every individual’s census information.
In planning for the 1890 census, Hollerith, Hunt, and Pidgin offered different processing methods. The census chose Hollerith exclusively on the consideration of minimizing processing time. The scope of these systems was remarkably narrow. They were all based on the representation of the information about every individual on a single card, which allowed sorting. All three eased the division of labor between the many operators and the few statisticians. The operators had limited training, and they compiled the tables from the census returns. The experienced—later trained— statisticians took on the complex assignments. They designed the census forms and the tables to be published, and they wrote the comments to the tables. The narrow scope of the methods offered can be explained by the fact that Hollerith, Hunt, and Pidgin were all members of the same networks of statisticians. The importance of such networks will fully emerge later.
This does not explain why other alternatives did not come into consideration. Why did the agenda not include a reduction of the scope of the census operation to something more manageable or the establishment of a permanent census institution? The elimination of the “relevant” alternatives is essential to understanding the shaping of this technology as is the actual choice of one of the relevant alternatives. That the United States eliminated alternatives can be explained through arguments based on institutional dynamics. Any reduction in the scope of the censuses was to be avoided, as the censuses had developed into a tool to demonstrate the capabilities of the United States as a great power. On the other side, Congress repeatedly resisted the establishment of organizations requiring permanent government funding. A permanent census institution would have had more focus on costs, as did the U.S. Bureau of the Census when it was established in 1902. The choice of Hollerith’s method over those of Hunt and Pidgin was motivated by speed. A decision based on cost would have turned out differently.