Statistical Calculation Facilities: 1920-1950s
by Dick Lund
Statistical calculations were tedious and time-consuming during the pre-WWII years. The prior Chapter 1 shows an early 12-digit Arithmometer, circa 1907. The first Monroe hand-cranked rotary calculator of the type we two authors “cut our teeth on” in the 1950s had been available since about 1911. I (Dick Lund) produced my first 4-variable multiple regression on one of these – using the “Doolittle algorithm" for solving the normal equations, cranking away and flipping the carriage to progress through the progressive digits for multiplying and dividing by means of adding and subtracting – working onward hour-after-hour on one of these! The more advanced students and faculty had priority for motor-driven Monroe models.
Major manufacturers of calculator in the U.S. included Friden, Monroe, and SCM/Marchant. Initially, some of these devices relied on a hand-pulled lever crank to do an operation, others a hand crank. Results of calculations were displayed by dials. Nearly all keyboards were full — each digit that could be entered had its own column of nine keys, 1, .., 9, plus a column-clear key, permitting entry of several digits at once. One could call this parallel entry, by way of contrast with ten-key serial entry that was commonplace in mechanical adding machines. Full keyboards had 8 to 10 columns. Most of the machines did not print results on paper. Addition and subtraction were performed in a single operation, as on a conventional adding machine, but multiplication and division were accomplished by repeated mechanical additions and subtractions. The user could find a square root via a popular mathematical algorithm.
Statistical calculations became a popular field for study by the 1930s, as universities and research labs began to acquire the early IBM mechanical punched-card tabulators and to create laboratories for their use. One of the first of such laboratories was founded about 1910 at the University of Michigan by James Glover, a professor who began teaching statistics courses in 1904. Glover's lab inspired Henry Louis Rietz to organize a computing lab at the University of Illinois sometime prior to 1918 at which time Rietz moved to the University of Iowa.
The U.S. Department of Agriculture consolidated all of its equipment into a single statistical office under its founder Howard Tolley in 1923. Tolley devised a practical method to compute correlations that required both a punched-card tabulator and supported a by-hand calculator (the tabulator was unable to multiply or divide).
Several successful companies dealing with tabulators and punched cards decided to merge in about 1911, completing that transition in 1924, and then becoming the one large non-competitive company we now know as IBM. Their 80-column 7 3/8 x 3 1/4 inch “standard” punched-card, so endemic to all data processing for the next several decades, was introduced in 1927. IBM soon brought out its new upgraded product line of equipment to be found in all computing labs and accounting offices across the world. Tolley’s undoubtedly would have assembled an adequate number data-entry key-punch and verification units, card sorters and punched-card driven tabulators in his computing lab.
The need for statistical processing of agricultural data brought together two Iowans, George W. Snedecor and Henry A. Wallace, Snedecor a professor at Iowa State College (before it became ISU) and Wallace a self-taught statistician who published a journal, Wallaces' Practical Farmer. Impressed with Wallace's knowledge of least squares, Snedecor invited him to teach an advanced course on those methods to college faculty. Subsequently, Snedecor and Wallace published a manuscript in 1925 on Wallace’s algorithm for calculating the solutions to normal equations.
The U.S. Department of Agriculture was the only source of government money for scientific research in the 1920s. It was supportive of empirical studies and helped over time to establish the largest and most sophisticated of the statistical laboratories, the Statistics Lab at Iowa State University (ISU). Snedecor acquired tabulators and supporting equipment in 1927, with governmental support. He also received funds to host summer institutes in statistical theory. The first of these was held in 1927 with instruction by R. A. Fisher. In subsequent years, the roster of visitors included many of the most distinguished people in the field. Among the researchers who were eager to learn Fisher’s methods was Henry A. Wallace.
U.S. President F. D. Roosevelt appointed Wallace to be Secretary of Agriculture in 1933. He helped produce the first piece of legislation in FDR's New Deal, the Agricultural Adjustment Act that required the Department of Agriculture to undertake large statistical studies of major farm products. The Act proved to be a boon for land grant colleges and experiment stations because they were asked to do much of the local statistical work. The ISU Statistical Laboratory acquired several government contracts in the mid-1930s, and the laboratory grew in both size and stature.
A similar laboratory was founded in 1927 by H. T. Davis at Indiana University, and other campuses followed as resources allowed, until the time when computer centers replaced calculator-based statistical laboratories.3
Fisher visited ISU several more times in the 1930s, often for several weeks at a time, lecturing and consulting. He thereby brought “modern statistics” to the staff and to students graduating from ISU. These students in turn were influential in developing similar statistical programs at several other Land-Grant universities. One person of note, a woman Gertrude Cox, about 1940, created the statistics program at North Carolina State University.6 Further over time, we can point also to the statistics programs at Kansas State, Oklahoma State, and Wyoming Universities, since these are in the background for Tiahrt, Lund and Hamilton in developing the statistics program at MSU in the 1970s. Dr. Bernard Ostle, an ISU graduate and short-time ISU faculty member, created a complete statistics program at MSC well before our arrival (see Era 4: 1944-1963).
While Snedecor did little original research in statistics, he was a great synthesizer other’s ideas. He first published his book Statistical Methods in 1937. Future publications continued through the 8th edition in 1989, the latter three editions coauthored by distinguished statistician.
I (Dick Lund) was “awarded the special privilege” (?) of helping read the galley proofs for the sixth edition of Snedecor's Statistical Methods, while a graduate student at ISU. I relied largely on Statistical Methods when teaching statistical methods classes in Mexico, 1967-67 and at MSU, 1969-81. The manual for my interactive statistical package MSUSTAT used examples from Statistical Methods to demonstrate nearly all procedures (MSUSTAT is discussed in Era 6: 1971-1976).
IBM had developed a new computer, the IBM 650, by the mid-1950s, To promote its use in academia, IBM created and funded the Western Data Processing Center (WDPC) sited at the University of California Los Angles (UCLA) in 1956. Several western universities were invited to associate with the Center and were given some free computer processing time. The University of Wyoming joined. I was selected to represent Wyoming and was provided a graduate assistantship by IBM.
I was expected “to contribute to programs of instruction related to data processing, assist persons at their own institution who may wish to make use of the facilities of the Center and possibly do some computer programming.” We did learn much about programing (called coding now) at that first workshop at UCLA in summer of 1957 – though it was held before the computer had actually arrived! It was in this assignment that I learned much about the many punched-card IBM machines available at both UCLA and Wyoming (Wyoming had installed the equipment in the registrar and business offices for accounting purposes, and only occasionally was it used for processing research data). I did help process some “real” research data from agricultural experiments on them.
Programs for the IBM 650 were written initially in SOAP (symbolic optimum assemble programming). I didn't have access to a compiler for Fortran (IBM's Mathematical Formula Translating System) which was first delivered to select locations that year. I developed a SOAP user’s guide for use at UW. I wrote one program to process data from a balanced lattice experimental design and another for a matrix inversion approximation routine for use on heavy diagonal symmetric matrices. I was able to debug these on another trip to UCLA. But I doubt that anyone ever used them.
Marty Hamilton said that, after I moved on, he held the IBM fellowship at Wyoming.
He was sent by WDPC to UCLA in the summer of 1961 to attend a short course on Fortran. It was not of immediate use to him because a Fortran compiler was not available for Wyoming's Bendix computer. The campus upgraded to an IBM, Fortran capable, machine soon after Marty graduated. Because Fortran quickly became popular among scientists, he necessarily acquired practical experience with Fortran coding by 1970 when he arrived at MSU.
Computing moved ahead somewhat faster in Bozeman. MSC acquired an IBM 650 in 1958 and installed it in Ryon Labs. Management of the computer was assigned to the Mathematics Department and MSC became a participating institution in WDPC in December that same year.
For AY 1961-62, the official representatives of the WDPC institutions included Dr. F. S. McFeely of MSC and Prof. J. C. Routson of UWyo; the IBM Research Assistants included Kenneth M. Lochner, Mathematics, MSC and Martin A. Hamilton, Statistics, UWyo. The map shows locations of the 66 WDPC Participating Institutions.9
- Dick Lund
(Last revised: 2021-04-18)