By 1870, the census had become so extensive, and the population of the United States so large, that Census Office clerks could no longer effectively tabulate results by expensive and inaccurate hand counting. In 1872 Charles W. Seaton, who became chief clerk in 1880 and superintendent of the census for several years thereafter, developed a machine that enabled several tally sheets to be brought close together, making them easier and faster to count. Seaton's device was first used in 1872, for the 1870 census, and then again in 1880. Seaton also invented a matrix printing apparatus for census work. While the Seaton machine was an improvement, the device was not advanced enough to keep up with the flood of data received for the 1880 census; final tabulations for that census did not finish until 1887.
Herman Hollerith, a former Census Office employee himself, invented a much more effective counting machine. His machine used specially encoded punch cards, each representing an individual's census data. The cards were fed into the counting machine, where the punched holes allowed metal pins to complete an electric circuit. When a circuit was completed, the dial for the corresponding trait would go up. Hollerith's device revolutionized census tabulation. The Census Office leased a fleet of the machines for the 1890 census count, which finished months ahead of schedule and far under budget. In 1896, Hollerith founded the Tabulating Machine Company. Almost twenty years later, after several mergers and management changes, this company became the International Business Machines (IBM) Corporation.
After Hollerith raised the rental prices for his machines to an intolerably high level for the 1900 census, the newly formed Census Bureau decided to create replacements in-house. Census Bureau employees, led by James Legrand Powers, developed a new electric tabulation machine. The new device, with an automatic feeder and card sorter, was an improvement over the machines that Hollerith was offering. In 1911, Powers, who held the patent for the machine that he had designed, left the Census Bureau to found the Powers Accounting Machine Company. Powers's enterprise was soon the most successful automatic tabulation company on the market; it later became part of the Remington Rand Corporation in 1927.
The illustration to the right shows one of the 92,000,000 cards used during the tabulation of population returns for the census of 1910. The holes in the four numbered spaces at the left are arbitrary symbols indicating the state and district in which the person to whom the card relates was enumerated; those in the other "fields" describe his characteristics. Thus, the person to whom this card refers resided in enumeration district No. 924 (Maynard, Middlesex County), state of Massachusetts; was a son of the head of the family in which he lived; mulatto; 20 years of age; native born; single; born in Georgia; father born in the United States; mother born in United States; spoke English; was an agricultural laborer; was out of employment on April 15, 1910; was out of employment between 7 and 13 weeks in 1909; could read and write; did not attend school; and was not a veteran of the Civil War.
In the late 1940's, the Census Bureau commissioned the first electronic computer designed for civilian use. Developed by J. Presper Eckert and John W. Mauchly, UNIVAC I marked a major improvement in data processing-it was the dawn of the "computer age." First used to process results from the 1950 decennial census, the machine was able to tabulate 4,000 items per minute, double the amount that electro-mechanical tabulating machines could process.
During the 1950s, the Census Bureau and the National Bureau of Standards developed a system called Film Optical Sensing Device for Input to Computers (FOSDIC), which took census and survey questionnaires that had been photographed onto microfilm, "read" blackened dots opposite the appropriate answers and transferred that data to magnetic tape. These tapes constituted the input for the Census Bureau's computers. One important result of this process was the elimination of most discrepancies in data records sent for processing. First used to process 1960 census results, FOSDIC played an integral part in the Census Bureau's data processing system into the mid-1990s.
Optical character scanners were used to process returned questionnaires for the 2000 census. The scanners recognized hand-written responses, as well as filled-in ovals or boxes. Using complex software, the scanned images were processed and translated into American Standard Code for Information Interchange (ASCII). These responses were then transmitted electronically over secure lines to Census Bureau headquarters for statistical processing and analysis.