Census 1921 - Processing the Data

As family historians, almost our only interest in the census is in the family details recorded in the household returns, but this did not enter into the thinking of those who carried out the census. Their only concern was to produce an accurate statistical analysis. How many people fell within various age ranges? How many were married and how many children did their family contain? How were they employed? Who these people were was of no concern and, indeed, the household returns were to be locked away for 100 years to make sure that this information remained confidential for the (anticipated) lifetimes of those named.

Up to 1901, this analysis was a painstaking manual process. Clerks extracted the information they needed from the returns line-by-line and manually produced the totals they required for the eventual published report. You can see signs of this when you look at the images of census pages. An occupation may be annotated with the employment category into which it falls, for example a "Labourer-Excavator" becoming a "Navvy" and "Tube & Fitting Warehouse Man" rather imprecisely as a "Plumber". You will also see various ticks and other marks where, for example, a birthplace is outside the county concerned.

This changed in 1911, when partial mechanisation was introduced, but it was not until 1921 that the process was mechanised throughout, from the initial coding of the household returns through to the calculation and printing of the results.

The breakthrough which made mechanical analysis possible has its origins in 1725 when Basile Bouchon invented a method for controlling the raising of a loom's warp threads using a paper roll with punched holes in order to create complex patterns. This was developed by the better-remembered Joseph Marie Jacquard (1752-1834) who in 1804 replaced Bouchon's paper roll with more robust punched cards linked into a continuous chain. Sensor rods pressed against the card and mechanical linkages raised the appropriate threads. The card is, in effect, a readable representation of the pattern to be woven. The principle, although now translated into computer-driven, electro-mechanical form, continues to be used to drive looms to this day.

The wider application of Jacquard's invention was realised by German immigrant to the USA, Hermann Hollerith (1860-1929). Hollerith recognised that the principle of coding information onto a card could be applied to calculation. The punching could represent any parameter and cards could be sorted and counted according to the information they contained. This had wide applications in business and its potential was soon recognised as a means to automate much of the tedious work involved in analysing the census. Consequently Hollerith's machines were used to process both the 1890 and 1900 US census and a decade later, the idea crossed the Atlantic to be used to analyse the 1911 UK census.

Hollerith's machines were, however, of limited capability. Although the company produced machines to punch and sort/count the cards, results had to be manually copied from the counters. The Hollerith machines were also viewed, at least in the USA, as excessively expensive. It fell to another immigrant to the USA to move the process forward. James Legrand Powers (1871-1927) was born in Odessa, Russia and emigrated to the USA in 1889. Trained as an engineer, Powers established his own company, the Powers Tabulating Company (later the Powers Accounting Machine Company), producing machines similar to Hollerith's. However, unlike Hollerith, his machines provided the tabulation and printing functionality needed to fully automate the process. Following successful application of Powers' machines to the Cuba census of 1908-9 the US Government chose Powers for the 1910 census, a decision echoed a decade later in the UK.

It is often the case that success does not always go to the best technology. Betamax, was generally regarded as of better quality than VHS and yet it was the latter format which prevailed. So it was with card-readers. Hollerith, ahead of his time, developed optical sensors to detect the presence or absence of a punch-hole. Powers was prevented from following suit by Hollerith's patents and so developed a mechanical reader, an update on Jacquard's mechanism of a century earlier.

In the UK, Powers' machines were marketed by the Accounting and Tabulating Machine Company (the "Acc and Tab" as it was commonly known). In 1920, the UK Government contracted the company to provide machines for the processing of the following year's census. There were essentially three, electrically-driven, machines involved in the process.

Automatic Key Punch: This was a machine used to convert the data from the household schedules into punched holes on cards. The punch was worked by an operator sitting at a simple numeric keyboard. The various parameters to be analysed were coded numerically before being punched onto the card. The card, similar to the example illustrated below, used a 45 column format with the columns divided into groups to represent, for example, location information such as the town, sub-district and enumeration district and personal information such as sex, age, condition (single, married, widowed, divorced) and occupation. One card was punched for each person enumerated.

Reader-Sorter: The punched cards were fed through this machine, which 'read' the punched information and sorted the cards by mechanically diverting them into one of a dozen pigeonholes. Sorting might be, for example, by sex, marital status or occupation. The machines could sort cards at the rate of 15,000 to 20,000 per hour.

The Punch Card used for the 1911 Census

Printer-Tabulator: The final stage was to feed the cards through this machine where they were once more mechanically 'read' and counts made of the selected parameters. The totals were then tabulated and printed onto paper using what amounts to a cross between an electric typewriter and an early version of a computer line-printer

The census processing centre, which was established in the old Lambeth Workhouse building in London, employed around 400 men and girls. Priority in the recruitment of male workers was given to former servicemen. One estimate puts the cost of the undertaking at £500,000 and suggests that had the information been processed manually, the cost would have been as high as £10 million.

It might be reasonable to expect mechanisation of the process would produce results more quickly. This was not, however, the case. Preliminary population totals were released at the end of August, about nine weeks after census day. By comparison the preliminary results for the census held on 1 April 1901, and the last to be analysed manually, were published on 14 June, about six weeks after census day.

What we see in this process is the precursor to elements of early computing. The use of punched cards for data entry was common for computers in the 1960s, though Hollerith's optical sensing method prevailed over mechanical reading. The printer-tabulator was the chosen output device before video displays emerged. These developments are not altogether surprising given the later history of the rival companies.

Hollerith's company was consolidated with two others in 1912 to create the Computing Tabulating-Recording Company (CTR), which in 1924 adopted the name International Business Machines Company, better known today as IBM. In an almost direct parallel, Powers' UK outlet, the Accounting and Tabulating Company, merged in 1929 with SAMAS (Societé Anonyme des Machines a Statistiques), Powers' French subsidiary, to create Powers-SAMAS. In 1959 a further merger with another competitor, the British Tabulating Company (BTC), created International Computers and Tabulators (ICT), which changed its name in 1968 to International Computers Ltd. (ICL), and became another important player in the development of computers. It was on an ICL-1901 mainframe computer that I was to run my first (though far from last!) computer programs in the late 1960s.