restricted access An Early Statistical Calculation
In lieu of an abstract, here is a brief excerpt of the content:

An Early Statistical Calculation

The first operational electronic computer in Canada was the Ferranti Ferut computer at the University of Toronto. (C.C. Gotlieb has described the introduction and initial use of this computer in an earlier work.1) The first major computation with the Ferut involved the solution of systems of equations related to the design of the St. Lawrence Seaway. Another early project was the inversion of a 64 × 64 matrix, which occurred in the design of the AVRO Arrow, Canada's ill-fated fighter aircraft.

When listing other kinds of numerical calculations performed on the Ferut, the words "statistical calculations" occur at the end of the list. In their 1958 book High-Speed Data Processing, Gotlieb and Pat Hume give a six-page discussion ofmultiple regression as an important application of electronic computers, although they do not reference a specific regression program.2 In an informative and delightful memoir, B.A. Griffith includes several references to multiple-regression problems, the first being a matrix-inversion program for the IBM 602A punched-card equipment used before Ferut became operational.3 Griffith continued this work with the Ferut by writing a matrix-inversion program subsequently used in a regression program that helped eliminate independent variables that were not statistically significant.

Given the importance of regression models in statistics and other disciplines such as economics, it is not surprising that references to regression programs occur in accounts of the uses of the first electronic computers. Another example of a substantive statistical calculation on one of the first electronic computers, although of limited application compared with regression analysis, was a program written in 1956 for the statistical design known as a Latin square. Here, I discuss the computer and programming environment in which this calculation was done and give an example of the program's rather obscure output.

NCR Computers

In 1954, I began my computing career with Computing Devices of Canada, a small electronics firm based in Ottawa. When I joined the company, it occupied the upper floors of a small building in the Westboro suburb, but a year or two later, the company moved to its own more spacious and comfortable quarters at Bell's Corners a few miles west of the city.

Most of the firm's early business was in the development of electronic equipment for various branches of the Canadian government. It also handled the Canadian sales of the NCR 102 computers manufactured by the National Cash Register Company of Dayton, Ohio. Two NCR 102A computers were installed in Canada, the first at A.V. Roe (Canada), an aircraft company in Malton, Ontario, adjacent to what is now Pearson Airport, and the second at the Royal Canadian Air Force Station at Cold Lake, Alberta. Computing Devices also acquired a later model, the NCR 102D, for its Ottawa offices. (Information on both the NCR 102A and 102D models is available in a late 1950s survey of domestic electronic computers.4)

The 102A consisted of the "computer proper" (as it was called in the programming manual), which was approximately six feet high, six feet deep, and two feet wide with a console consisting of a modified Flexowriter typewriter with a paper-tape reader and punch and a small control panel. The input speeds of the reader and punch were 10 characters per second. The basic system cost approximately US$82,000. Specifications stated that it required 7.7 kilowatts of power, occupied 250 square feet, weighed 2,700 pounds, and had 400 tubes of 12 different types and 8,000 crystal diodes. It also required air conditioning and a separate power supply, and it had a magnetic drum memory of 1,024 42-bit words and an additional buffer memory of eight words for input/output purposes. (The main memory's capacity, in units customarily used today, measured 0.0000054 Gbytes.) The internal number system was binary with addition and subtraction times of 7 to 20 milliseconds depending on access and multiplication and division times from 25 to 38.5 milliseconds. NCR produced 16 102A computers in all.

The 102D was a decimal version of the 102A.5 The internal number system was binary-coded...