Herman Hollerith and his first tabulator from 1890
In 1879 the bright and unmanageable teenager Herman Hollerith (see biography of Herman Hollerith) graduated with distinction at School of Mines at Columbia University in New York (but with low marks only in bookkeeping and machines) and went to work as a special agent for the US Census Office in Washington, D.C. One of his professors of engineering, William P. Trowbridge (1828-1892), engaged for the Census as an expert special agent, got the job for him. The 1880 census was about to begin, and the Census Bureau needed employees with mathematical and engineering ability.
Although the 1880 census had taken only a few months, the work of tabulating and analyzing the data promised to drag on for years. By the time it was done, the census reports would be hopelessly out of date; the government would be lucky enough to finish in time for the next census in 1890. Since the country’s population, swelled by immigration, was growing by the millions, the 1890 census undoubtedly would take even more time and money. The situation was getting out of control and the Census Bureau was casting about for a solution.
In his spare time, Hollerith helped Dr. John Shaw Billings (1838-1913), a surgeon and head of the division of vital statistics of the Census, compiling his reports. Billings appreciated the young man’s help, and invited him to dinner one Sunday night in August of 1881. That proved to be a cornerstone event in the life of Hollerith, as he recalled later:
One Sunday evening, at Dr. Billings’ tea table, he said to me there ought to be a machine for doing the purely mechanical work of tabulating population and similar statistics. We talked the matter over and I remember… he thought of using cards with the description of the individual shown by notches punched in the edge of the card…. After studying the problem I went back to Dr. Billings and said that I thought I could work out a solution for the problem and asked him would he go in with me. The Doctor said he was not interested any further than to see some solution of the problem worked out.
Tackling the problem on his own, Hollerith decided to study the Census’s procedures. In the first step of the count, enumerators called at every household unit and recorded the answers to their questions on large sheets of paper (known as schedules). The completed schedules were sent back to Washington, where an army of clerks transcribed the answers to tally sheets. For example, for every white male on a schedule, a slash mark was placed in a small box on a tally sheet, five slashes to a box. It was easy to add up the slashes on a tally sheet, since the form was divided into large boxes that contained a specific number of small boxes. The clerks totaled up the completed large boxes and noted the number of slashes at the bottom of the sheet. In the next step, the tally totals were transferred to consolidation sheets, whose figures were combined to yield the population of the county, state, and finally, of the whole nation.
The 1880 enumeration required six tallies, one for every major statistical classification. In the first tally, the Census broke down the population by sex, race, and birthplace; in other tallies, it collated these statistics with literacy, occupation, and other characteristics. Every time a tally was called for, the clerks had to sift through the schedules all over again, and there were millions of schedules. The process was remarkably slow and expensive, not to mention prone to error. Moreover, it prevented the Census from performing sophisticated analyses of the data.
Almost everything post- and pre- census work was done by hand. The only mechanical aid was a simple contraption called the Seaton device, invented by Charles W. Seaton, the Census’s chief clerk. It consisted of a continuous roll of tally sheets wound on a set of spools in a wooden box. By zigzagging the roll around the spools, it brought several columns of a sheet close together, making it easier for the clerks to enter the slashes. Completed rolls were removed from the box, cut into separate sheets, and consolidated numerically. Even with the Seaton device, the 1880 census took nearly the entire decade to tabulate and publish.
In 1882, Hollerith became an instructor in mechanical engineering at Massachusetts Institute of Technology, where he started building his first tabulating system (a year later he returned to Washington to become an examiner for the Patent Office). Initially he made a mistake, deciding to use punched tape instead of cards. The tape was run over a metal drum, under an array of metal brushes; whenever the brushes passed over a hole, electrical contact was made with the drum, advancing a counter. A separate counter was set up for each statistical category, and the totals were displayed by a number on the counter. Anyway the first system was a huge improvement over tally and consolidation sheets; once the data on the schedules had been converted into punched tape, many items could be tabulated in a single, fast run of the tape, in contrast to the one, two, or three items that could be collated on a tally sheet at any time.
Although this system was a big step forward, Hollerith soon realized that he had made a serious mistake: paper tape was a flawed medium, severely limiting the tabulator’s speed and flexibility because of its serial access. For example, if you wanted to retrieve from a tape a particular piece of information, or related pieces of information, you’d have to sift through the (entire) reel. Moreover, once you found the data, there was no way to isolate it for future reference (other than cutting the tape into pieces). It seems Hollerith came to a dead end, forgetting for the Billings’ idea of punch cards. However, once he was traveling in the West and as he recalled later “…and I had a ticket with what I think was called a punch photograph…. The conductor… punched out a description of the individual, as light hair, dark eyes, large nose, etc. So you see, I only made a punch photograph of each person.” (Punched photographs discouraged vagrants from stealing passengers’ tickets and passing them off as their own.) Thus Hollerith rediscovered the idea it on his own.
First tabulator from 1890
In effect, Hollerith cut his tape into sections (cards, which measures 3.25 by 7.375 inches, the same size as the 1887 US paper currency because Hollerith used Treasury Department containers as card boxes), and the result was a quick and versatile tabulating system. Once one had transcribed the information on the schedules to the cards, one could manually or electromechanically isolate any card or class of cards (so instead of serial access you have random access.) For instance, you could set aside a pile of cards representing nothing but farmers and perform any statistical analyses of this group you wished. You did not have to sift through the schedules all over again. Therefore, if you wanted to know how many farmers owned more than ten hundred acres and earned their income from wheat, you had only to run through the farmer cards, setting up the counters on the tabulator to match the appropriate holes in the cards.
The decision to use cards led Hollerith to redesign his initial system. He designed a special puncher (a pantograph punch consisting of a template and two connected punches); when the operator punched the template, the second puncher perforated the card. The card reader (see the nearby image) was a small press made up of an overhead array of pins and an underlying bed of tiny cups of mercury; when the operator slipped a card into the press and pulled down on the handle, the pins passed through the holes into the mercury, closing electrical circuits that advanced the counters (each completed circuit caused an electromagnet to advance a counting dial by one number), 40 simple clock-like dials set into a wooden table. When the bell signaled the card had been read, the operator recorded the data on the dials, opened the card reader, removed the punch cards, and reset the dials.
The sorter (see the nearby image) was simply a box with several compartments, positioned next to each tabulator. When a card with a desired set of characteristics passed through the reader, a box on the sorter opened up, and the operator slipped the card into it, then reset the dials, and positioned a new card to repeat the process. An experienced tabulator clerk could process 80 punch cards per minute.
Hollerith intended to power his tabulators with batteries and recharge them through the power outlet.
In 1884 he applied and in 1889 received his first patent for a tabulator (pat. No. US395783), for a tabulator with a paper tape. On the same date he received his first patent for a tabulator with punched cards (pat. No. US395781).
The Census Office was impressed with Hollerith’s work, but it decided to conduct an official test of the system before making a commitment. The trial in 1888 pitted Hollerith’s machines against the “chip” system of Charles Felton Pidgin (who invented the first electro-magnetic calculating machine in the world) and the “slip” system of William C. Hunt, both Census officials. In the chip system, data from the schedules were transcribed to colored cards; in the slip system, the information was written onto slips of paper in colored inks. In both cases, the cards and slips were counted by hand. The competition called for the transcription and tabulation of a thick sheaf of schedules, compiled during the 1880 census, covering 10491 people in St. Louis. There were two parts to the trial: the time required to transcribe the schedules and the time required to tabulate the data. Surprisingly, Hollerith’s system smashed the rivals. It showed its greatest advantage in the tabulation portion of the test, completing the job eight to ten times faster than the hand-counted slip and chip methods.
Pleased with the results, the Census ordered 56 tabulators and sorters, and Hollerith was in big business.
Hollerith’s machines went to work in July 1890, shortly after the completion of the head count of the census. The first task was a general tally of the population, and Hollerith devised a special counter for the job, a typewriter like device equipped with twenty keys, numbered 1 to 20. The clerks read the schedules, which represented only one family per sheet, then pressed the key signifying the number of people on the schedule. Some operators handled 9200 schedules, listing 50000 people, in a single day. By August 16, only six weeks after the count had begun, the Census had a tally: 62,622,250. With great pride and fanfare, the figure was officially announced in October, and everyone was suitably amazed.
Hollerith’s punch card system received a great deal of attention in the popular and scientific press in the USA and abroad and was featured on the front tape of the 30 August 1890 issue of Scientific American (see it below).
Compared to the 1880 census, which had taken nine years and cost $5.8 million, the 1890 count was completed in fewer than seven years, but it had cost $11.5 million, twice as much. Under the circumstances, there was some controversy about the benefits of automation. The Census, which had paid only $750,000 in rental fees for Hollerith’s equipment, ascribed the financial disparity to the expense of running a far more careful and thorough statistical analysis of the raw data. Indeed, the Census estimated that it had actually saved about $5 million in labor costs. From the current perspective, even then became clear, that the hardware is cheap, but the software is expensive 🙂
The 1890 tabulator was capable only of counting. Subsequent models, developed by Hollerith, were also able to add, thus broadening their scope to accounting, warehousing, and shipping applications. Between 1902 and 1905, Hollerith also developed an automatic card feed and a method for reading cards in motion and settled on a standard card format.
Hollerith’s system was promptly adopted all over the world. In late 1890, Austria, then Russia, Canada and France ordered several tabulators and sorters for their census. After some initial resistance, private industry began using them too. Swamped with paperwork, large companies like the Chicago department store, the New York Central Railroad Company, and the Pennsylvania Steel Company moved the equipment into their accounting and inventory departments.
In 1892 Hermann Hollerith moved his fledgling tabulating machine business from downtown Washington, D.C., to a former cooper’s shop in the Georgetown section of the city. Seen in the nearby image the following year 1893, the two-story building (later expanded) housed Hollerith’s card manufacturing plant, assembly plant, repair shop and development laboratory. Three years after this photograph was taken, in 1896, Hollerith incorporated his business as the Tabulating Machine Company.
By the early 1900s, Hollerith’s firm, the Tabulating Machine Company, had more customers than it could handle. However, because the firm leased rather than sold its equipment, which provided a steady and quite profitable stream of income but produced a thinner cash flow, the company was always short of capital. Moreover, Hollerith was a brilliant engineer, but not a brilliant businessman and he gradually lost a large part of his business. The best solution seemed to be a merger, and in 1911 Tabulating Machine Co. joined with three other outfits to become the Computing-Tabulating-Recording Co., which eventually in 1924 became the International Business Machines Corporation (IBM).