17,814 Cards

On 2 October 1891, Mary Wasner handled 17,814 cards at the United States Census Office. Between batches she stopped to read the machine’s dials and write down its totals. Her day included five such readings. The cards went under the press one at a time.

Wasner’s figure appears in a table published by the electrical journalist T. C. Martin after a visit to the Washington offices. Maurice Alexander had managed 19,071 cards in a day; Mrs Z. Richardson’s best recorded weekly average was 11,351 a day. There were names behind the extraordinary numbers being claimed for Herman Hollerith’s new tabulating system.

The equipment combined a cabinet of clocklike counters, a lever-operated press and a box with a row of lids. A card could make several counters move together. It could also make one lid spring open, telling the clerk where to put it. Across millions of repetitions, those small movements would produce a much more detailed account of the country than a head count alone.

Information left on the page

The United States took a national census every ten years. Hollerith had worked on the 1880 census and encountered a problem that grew as officials asked more questions. Collecting answers did not ensure that anyone could afford to turn them into useful tables.

The schedules already contained answers about marriage, widowhood and divorce. The expense lay in combining them with everything else.

In his own technical paper of 1889, Hollerith complained that the 1880 returns still held unused information about the nation’s marital condition. A simple total would be useful; tables connecting it with age, sex or occupation would be more useful again. Under a tally-sheet system, each additional inquiry could require another laborious journey through the returns.

Hollerith proposed transferring each person’s particulars to a separate card. The card could be counted, sorted and used again for another table. In 1889 the Census Office tested his system against two other proposals, using returns from four districts in St Louis. William C. Hunt offered written slips; Charles F. Pidgin used coloured chips. Both also separated the individual record from the original return.

Making Hollerith’s cards took 72 hours and 27 minutes. Tabulating them took another five hours and 28 minutes. Hunt’s tabulation alone took more than 55 hours, Pidgin’s more than 44. The electrical system won, though even its preparation occupied far more time than its counting.

Building the equipment required other workshops. Pratt & Whitney developed and made the keyboard punches under George M. Bond’s supervision. Western Electric supplied the electrical apparatus. Hollerith’s arrangement with the government included keeping the machines in working order.

One family, many holes

In 1890 enumerators used a fresh schedule for each family. Alongside names came ages, occupations, birthplaces, school attendance and many other particulars. The household summary at the top supplied a quick route to the first population total: clerks entered the family-size figures through a small keyboard connected directly to Hollerith counters.

Martin reported that this preliminary operation took about six weeks. It did not involve feeding more than sixty million individual punched cards through the presses. Those cards were needed for the detailed work that followed.

To make one, a punching clerk read the schedule and moved a stylus over a labelled board. The pantograph—a linkage that copied the stylus’s movement—positioned a punch over the card. Pressing into a chosen hole in the board punched its counterpart in the manila stock.

Some answers needed one hole. Others required a combination: age was divided between a group of years and a particular year within it. Classification lists helped clerks turn places and occupations into the prescribed codes. A list told them which German territories belonged under the code for Germany, rather than France, Austria or Denmark. A job written in ordinary language had to become an approved statistical occupation.

Government had also decided which distinctions were worth counting. The racial categories included the supposed fractions of ancestry called “quadroon” and “octoroon”. Those classifications entered through instructions and human judgements before a circuit could count them. Electrical precision at the desk could reproduce the choices already made on the form.

Geography was punched in batches.

The clerk making personal holes left a strip at one end for the enumeration district, the area assigned to a census enumerator. A separate gang punch carried the district’s pattern through three to six cards at once. Each card also had an identifying number. The completed cards stayed together by district, packed upright in labelled tin boxes.

Experienced punchers were reported to make around 1,100 cards between nine and four. Each of those cards still had to be read by the tabulating apparatus, often more than once.

A circuit beneath the paper

Solid paper pushed a pin back. A hole let it continue.

The clerk laid a card against the press’s guides and brought down the handle. Pins that passed through holes reached small cups of mercury in the hard-rubber bed. The completed electrical connections operated electromagnets, advancing the corresponding counters by one.

Each counter had two hands, one for units and another for hundreds. Wiring determined which particulars it counted. Relays could make a counter respond only when several conditions coincided, allowing combinations of birthplace, sex and other recorded characteristics to be counted together.

The same stroke could sort the record for another inquiry. If the sorting box was connected for age groups, the appropriate compartment opened as the card was read. The operator released the handle, lifted the card with her right hand, put it into the open compartment and closed the lid. Her left hand brought the next card into place.

The sorting box indicated the destination; the clerk made the journey. Once a district’s cards had passed through, one set of figures was ready on the dials and the cards themselves were already divided for further work. They could be regrouped into larger geographical areas and passed through again, yielding new combinations without another reading of every family’s handwriting.

A closer look · 7:44 · German original

Punch, read and sort

Play film

When the bell did not ring

Martin described the sound of monitor bells in the workrooms. A bell rang when a card was counted; silence could mean a missing required punch or a card incorrectly placed. Connected for one district, the machine could also refuse a card carrying another district’s geographic pattern.

There was a correction room where rejected cards were checked against the original schedules. Those returns remained close at hand, stored in an old church across the street from the Interocean Building where the cards and machines were housed.

The checks had limits. Martin distinguished an inconsistent record, such as one lacking sex, from a plausible but incorrect occupation code. A policeman punched as a poolroom keeper was his example of the second kind. The circuit would have no reason to object to a valid hole in the wrong occupational position.

His defence of the apparatus was emphatic. Newspaper attacks on the census had implicated the new machines, and he examined two published examples of supposedly defective cards. Both lacked the geographic gang punches. According to Martin, these were incomplete cards discarded before they ever reached the tabulators.

That answered a particular accusation against the office machinery. It did not settle whether enumerators had found everybody or recorded their answers correctly. Even the editorial accompanying Martin’s enthusiastic report allowed that local enumeration could be criticised.

Back into handwriting

During Martin’s visit, 81 machine clerks handled a reported 556,346 cards in a day, counting some and sorting others. Finishing a district also meant reading 38 dials, copying their totals onto prepared result slips, adding, verifying and making corrections. The cabinet produced no printed table of its own.

The slips supplied the figures that could be consolidated for publication. Part I of the census population report appeared in 1895, with tables on subjects including birthplace, parentage, marital condition, dwellings and families. Part II followed in 1897, covering detailed ages, schooling, literacy, citizenship and occupations.