The state wanted to count its citizens. What it got was the invention of the future.
The U.S. Constitution requires a census every ten years, and by 1880 that requirement had turned into a trap. The country was growing faster than officials could process the results of the previous count. Fifty million people, each one a line in an immense paper register, and every line had to be read, classified, and entered into a summary table by hand. Processing the 1880 census dragged on for more than eight years. By the time the final tables were finally ready, the data was already obsolete—people had moved, been born, died, while the government was still totaling up the previous decade.
The reason was not laziness or a shortage of people, but arithmetic: the country’s population was growing by millions each decade, while the speed of manual processing did not change. At that pace, the 1890 census risked becoming meaningless—its results might not appear until the next census. The Census Bureau understood that another method was needed, and in 1888 it announced a competition for a mechanized data-processing system.
Three contestants presented their solutions. A test sample—real population data—became the proving ground. One contender completed the processing in 144.5 hours, another in 100.5. The third participant, a young engineer named Herman Hollerith, finished in 72.5 hours. The truly staggering difference came in tabulation—the actual counting and aggregation of the data: it took the competitors 44.5 and 55.5 hours, while Hollerith did it in 5.5 hours. A tenfold difference. The Bureau did not hesitate.
The idea of recording information with holes in a material was already more than half a century old. Since the beginning of the 19th century, Jacquard looms had used punched cards to control patterns: there is a hole—the thread rises; no hole—the thread stays down. Charles Babbage designed his Analytical Engine around the same principle of input. Punched tape controlled barrel organs and mechanical organs. The hole as a unit of code is older than electricity in everyday life.
Between an idea and a working system lies a chasm filled with engineering decisions. A Jacquard card controlled a loom—one specific mechanism performing one task. Hollerith created a universal system: holes on a card encoded arbitrary attributes of a person—age, sex, birthplace, marital status, occupation—the machine read those attributes electrically, counters accumulated statistics, and a sorting device distributed the cards into categories for further analysis. Roughly speaking, the Jacquard card was sheet music for one instrument, while the Hollerith card was a questionnaire a machine knew how to read.
Hollerith did not invent the punch card. He invented what to do with punch cards when you have sixty million of them.
The system’s technical heart—the reader—was built with the elegant straightforwardness of things designed for real work. Picture a hand press, like a bookbinding press. In the upper section—a grid of spring-loaded metal pins. In the lower one—a plate with sockets, each containing a drop of mercury at the bottom. The operator places a punch card between them.
When the press comes down, the pins descend. Where there is no hole in the card, the paper stops the pin—the spring compresses, no contact is made. Where a hole has been punched, the pin passes through and touches the mercury at the bottom of the socket. Mercury—a liquid metal, an excellent conductor—closes the electrical circuit. A pulse of current runs along a wire to one of 40 dial counters, and the needle on the corresponding dial advances one mark. Each press of the machine—one person. Each hole—one fact about them. Each click of the needle—one unit in a statistical summary.
What fascinates me is the physicality of the process. Data here is literally holes. Reading is literally metal touching metal through paper. Electricity does not compute; it merely transmits a signal, but that is enough to lift the monstrous burden of counting from human eyes and fingers. The operator does not need to read the card—the machine does it for them, turning the geometry of holes into counter readings.
The sorting device deserves its own admiration. When a particular circuit closed, an electromagnet opened the lid of one compartment in the sorting box. The operator dropped the card into the compartment that had opened; the lid snapped shut. That way, cards were automatically grouped according to the desired attribute—say, everyone born in Ireland went into one compartment, everyone born in Germany into another. The sorted stack could then be run through the reader again, this time for a different attribute. Cross-tabulation—counting one parameter within a subset selected by another—became a routine operation.
The 1890 census covered more than sixty-two million people. A punch card was made for each of them. The cards were created with a pantograph—a device that allowed an operator to punch a hole in the required position by pressing a lever. One pantograph operator could prepare around 500 cards a day. That means punching the entire dataset required hundreds of people and months of work—the data-entry process remained manual and labor-intensive.
But tabulation—the actual counting—proceeded at a speed unimaginable for the time. 81 clerks worked with the readers, and on peak days the system processed up to 556,346 cards a day. The first overall results of the census—the population of each state—were obtained within weeks of processing beginning. For comparison, the equivalent stage of the previous census had taken months.
An honest caveat is needed here. The myth that the entire 1890 census was processed “in a few weeks” is a simplification that wanders from article to article. The total population count really was obtained quickly. But the detailed statistical tables—the distribution by age, occupation, literacy, national origin, and dozens of other parameters—continued to be compiled for years. Hollerith’s system radically accelerated every individual stage, but there were many stages, and each required a new pass of the cards through the readers, new sorting, new adjustment of the counters. A census is not one question—“how many are we?”—but thousands of questions about who we are.
Still, the effect was undeniable. The Census Bureau saved millions of dollars. Data became available to policymakers and planners far earlier than it would have through manual processing. And Hollerith gained something greater than a government contract—proof that his system worked on the scale of an entire country.
After the success of the census, Hollerith followed the path logical for an inventor with a head for business. He did not sell the machines—he leased them and sold the punch cards, creating the model that half a century later would be called “razors and blades.” The governments of other countries—Canada, Russia, Austria—became interested in the system for their own censuses. Railroads and insurance companies discovered that tabulators were useful for accounting and record-keeping.
Around 1896, Hollerith founded the Tabulating Machine Company. The business grew, though not without friction: by the next cycle, the Census Bureau considered Hollerith’s prices excessive and began looking for alternatives. The inventor’s relationship with his chief customer deteriorated. In 1911, Hollerith’s company became part of the Computing-Tabulating-Recording Company (CTR) conglomerate, which brought together several manufacturers of counting and measuring equipment.
In 1914, Thomas J. Watson Sr. joined CTR—a talented executive trained in sales at National Cash Register. Watson was not an engineer and had nothing to do with the invention of the tabulator. His genius was of another kind: he knew how to build corporations. Under his leadership, CTR became a machine for selling machines—with aggressive marketing, a cult of loyalty, and the famous slogan THINK, hanging in every office. In 1924, Watson renamed the company International Business Machines — IBM.
By then, Hollerith had stepped away from active management. He died in 1929, having lived to see his tabulating workshop begin its transformation into one of the world’s largest corporations, but not long enough to see the era of electronic computers, which his punch cards would feed with data for decades to come.
There is one aspect of Hollerith’s system that is easy to miss amid admiration for the mercury contacts and dials. Before the 1890 census, population data existed in the form of records—lines in ledgers filled out in a census taker’s handwriting. A record can be read, but it cannot be processed mechanically. It is bound to language, handwriting, the format of a particular form.
The punch card turned data into a physical object with a machine-readable structure. A hole in position 5 of the third row means “female”—and that meaning does not depend on the operator’s language, literacy, or mood. A card can be run through a reader in Washington or Vienna, and the machine will read the same thing. Cards can be sorted, grouped, recounted in a new order—essentially, you can do with them what we now do with database queries.
I think it matters not to stretch the analogy too far. Hollerith’s tabulator was not a computer. It did not execute a program, store data in memory, or make conditional decisions. It was a counting machine—fast, reliable, scalable, but fundamentally limited. And yet, in the gap between the census ledger and the punch card, a shift occurred whose consequences we still feel: information separated itself from the paper on which a person had written it and began to exist in a form accessible to a machine.
IBM’s 80-column punch cards—descendants of Hollerith’s cards—remained the standard medium for entering data into computers well into the 1970s. Programmers of the first generations wrote code by punching holes in cardboard. When, in the 1960s, cards carried the warning “Do not fold, spindle or mutilate”—do not bend, pierce, or crumple them—they were protecting a data format invented by a man solving the problem of counting a population seventy years earlier.
The state wanted to know how many people lived in the country. An engineer came up with a way to encode the answer as holes in paper and read it with a drop of mercury. From that grew an industry that now processes more data in a second than the entire 1890 census contained—yet at its deepest level, it is still doing the same thing: turning facts about the world into signals that can be counted.