On this page
Databases, from exam results to railway bookings, rest on one simple idea: store each record in a fixed format so a machine can count and sort it. One of the first people to make that work at national scale was Herman Hollerith. This post covers the problem he solved, how his machine worked, a worked example with real contest figures, and what students can learn from him.
What problem was Hollerith trying to solve?
The United States counts its people every 10 years. According to the US Census Bureau's history of tabulation, by 1870 the census had grown so large that clerks could no longer tabulate it well by hand. Even with a tally machine, final tabulations for the 1880 census did not finish until 1887.
Hollerith had worked at the Census Office himself. IBM's history of the punched card tabulator describes him as a mining engineer from Buffalo, New York, who grew frustrated with the slow, error-prone hand counting during a short stint there.
How did the tabulator work?
Each person's census answers were punched as holes in a card, with each hole position standing for one piece of information. A clerk placed a card in a reader that worked like a press. Spring-loaded metal pins came down onto the card. Where there was a hole, a pin passed through into a cup of mercury and closed an electric circuit. That pulse moved one of 40 dials forward by one. Where there was no hole, nothing happened. A separate sorter could drop cards into as many as 24 categories.
IBM's account calls this the dawn of the binary principle: a closed circuit counted as 1, no connection as 0. Punched cards themselves were older. Joseph Marie Jacquard used them to control looms, and Charles Babbage planned to use them in his Analytical Engine; our post on Charles Babbage's engines tells his story. Hollerith's step was to apply them to counting data.
His patent, US 395,782, "Art of compiling statistics", was granted on 8 January 1889.
Worked example: how much faster was Hollerith's system?
In 1888 the Census Office held a contest to choose a system for 1890. IBM gives these times for the test data.
Capturing the data: Hollerith 72.5 hours; rivals 144.5 and 100.5 hours.
Preparing it for tabulation: Hollerith 5.5 hours; rivals 44.5 and 55.5 hours.
Speed-up in capture. 144.5 ÷ 72.5 ≈ 1.99, so about twice as fast as the slower rival. 100.5 ÷ 72.5 ≈ 1.39, so about 1.4 times as fast as the quicker one.
Speed-up in preparation. 44.5 ÷ 5.5 ≈ 8.1 and 55.5 ÷ 5.5 ≈ 10.1. That is where IBM's "up to 10x" comes from.
A common slip. "Up to 2x faster" compares with the slowest rival only. Against the closer rival the gain was smaller. Always ask which two numbers a ratio compares.
Binary check (sample numbers). Suppose one region of a card has 12 positions, each punched or not. Each position has 2 states, so the region can hold 2^12 = 4,096 different patterns. Our guide to binary and other number systems shows how to work with such counts.
What happened after the 1890 census?
The Census Bureau's history says the 1890 count finished months ahead of schedule and far under budget. IBM puts the savings at USD 5 million and more than two years of labour.
In 1896 Hollerith founded the Tabulating Machine Company. The Census Bureau later objected to his rental prices and built its own machines. In 1911 Hollerith sold his company, which merged with two others into the Computing-Tabulating-Recording Company. That firm was renamed International Business Machines, IBM, in 1924. Hollerith died in 1929.
In the late 1940s the Census Bureau commissioned UNIVAC I, an electronic computer built by J. Presper Eckert and John Mauchly, and first used it for the 1950 census. Their earlier machine is the subject of our ENIAC story.
What can a student take from how Hollerith worked?
Start from a real frustration. He saw clerks doing slow, error-prone work and asked what a machine could do instead.
Test on real data first. IBM notes he tried his system on mortality statistics for Baltimore, Jersey City and New York City in 1886.
Let measured results argue for you. He won the census contract on timed tests, not promises.
Reuse old ideas in a new place. Cards came from looms. His contribution was using them for statistics.
If the yes-or-no logic of holes and circuits feels abstract, the Fermi Concept Library covers every concept in physics, chemistry, biology, mathematics and computing, so you can look up any idea that feels unclear.
Try it now: work out how many patterns a card region with 10 positions can hold, then how many positions you need for at least 1,000 patterns.
