BMD(ID:145/bmd001)Bio-Medical statistical languagefor BMD, BioMeDical computer programs, also BMDP for BioMeDical Package. Statistical language, first implemented in FORTRAN for the IBM 7090. Dixon et al at Health Science Computation Facility, UCB, 1961. Bought out by SPSS ni 2002 Hardware: Related languages
References: In the last few years the speed of computers has increased so much that computer time is responsible for only a small fraction of the of propramming personnel, particularly those who have some background in physics, chemistry or engineering, which has led many subject-matter specialists to learn programming. This is of course wasteful, in the sense that a number of do-it-yourself activities are wasteful, and also involves an additional loss because of the time taken from the pursuit of the subject matter in which the programmer is a specialist. Although inefficient, it is justified because of the relatively low cost and high speed of modern machines and because of the inaccessibility of professional programmers. The system is unsatisfactory-the scien tist involved becomes more programmer than physicist or chemist and either has to train each new man to code or end up doing his problems himself. This provides the motivation for a general-purpose code to himself. This provides the motivation for a general-purpose code to facilitate communication with the machine, a code which requires as little specialized computer knowledge as possible. For the statistician, general-purpose codes for regression, time series, etc., would probably suffice. There may be other equally good collections of codes, but the programs prepared by the UCLA Biomedical Data Processing Group (BMD) come as near to filling the needs of the statistician as we can reasonably hope to get. Yet in many applications the experimenter wants tables made from the results of curve fitting, plots of auxiliary functions, or auxiliary constants such as air density, virtual temperature, etc. In a number of cases the statistical analysis is a minor part of the calculation and printout. It is a misuse of the statistician's abilities for him to expend his energies on preparing such nonstatistical programs. This desire for a general-purpose coding system in which the machine is programmed by the writing of English sentences has brought forth a number of systems: AARDVARK (Iowa State), BOUMAC (National Bureau of Standards Boulder Laboratories), COGO (MIT), DAM (International Monetary Fund), etc. The following quotations from four of these programs show the strong and consistent motivation toward the utilization of a language oriented toward the subject-matter specialist rather than the machine: COGO - a computer programming systen1 for civil engineering problems ...has the following characteristics: The instructions or commands to the computer, which the engineer uses to express the solution of a problem, are at approximately the same technical language level as instructions which one engineer would use in describing his solution to another engineer. (Miller, 1961, p. 1) The above presentation may be regarded as a Problem Oriented Control Language oriented toward Multivariate Analysis. It is felt that this language is close to the language a statistician would use to express his problem to other statisticians so that the statistician wishing to perform certain multivariate analyses has a minimum of computing rules to learn before he can present his problem. (Cooper, 1963, p. 27) The program, hereafter referred to as DAM (Data processing And Multiple regression), facilitates the preparation of input data to be used in multiple regression analysis. It enables users without specific programming knowlregression analysis. It enables users without specific programming knowledge to write sequences of desired computations essentially in the form and (Boissonneault, 1962, p. I) A formal approach to the routine analysis of kinetic data in terms of linear compartmental systems is presented. The methods of analysis are general in that they include much of the theory in common use, such as direct solution of differential equations, integral equations, transfer functions, fitting of data to sums of exponentials, matrix solutions, etc. The key to the formalism presented lies in the fact that a basic operational unit-called "compartment" - has been defined, in terms of which physical and mathematical models as well as input and output functions can be expressed. Additional features for calculating linear combinations of functions and for setting linear dependence relations between parameters add to the versatility of this method. The actual computations for the values of model parameters to yield a least squares fit of the data are performed on a digital computer. A general computer program was developed that permits the routine fitting of data and the evolution of models. (Berman, Weiss and Shahn,1962,p.289) As useful as these systems are in their specialized fields, a need for a system directed toward the wide variety of mathematical a a need for a system directed toward the wide variety of mathematical and numerical calculations arising in the physical sciences and neering. OMNITAB (Hilsenrath, 1963) was designed to meet this need. in Proceedings of the IBM Scientific Computing Symposium on Statistics 1963 at the Thomas J. Watson Research Center in Yorktown Heights, New York, on October 21, 22 and 23,1963. view details in Proceedings of the IBM Scientific Computing Symposium on Statistics 1963 at the Thomas J. Watson Research Center in Yorktown Heights, New York, on October 21, 22 and 23,1963. view details in Proceedings of the IBM Scientific Computing Symposium on Statistics 1963 at the Thomas J. Watson Research Center in Yorktown Heights, New York, on October 21, 22 and 23,1963. view details in [AFIPS] Proceedings of the 1967 Fall Joint Computer Conference FJCC 31 view details in [AFIPS] Proceedings of the 1967 Fall Joint Computer Conference FJCC 31 view details in R.C. Milton and J A. Nelder (Eds.) "Statistical Computation" Academic, New York, 1969 view details Dixon's on-line statistical programs Dixon (1967) describes the use of on-line displays with packaged statistical programs. In this paper he discusses the usefulness of being on-line with a statistical program and especially the usefulness of on-line graphical ability. He says " . . . many experts in data analysis have always used graphical methods to aid their analysis of data. One often hears directives of these experts to their assistants something like, 'Go thou and plot your data.' The plots and charts frequently do not survive the process of report writing and publication, but have played an important part in the analytical process itself." An on-line interactive situation is ideal for the examination of these charts and plots which are discarded during the data analysis process. At the Health Sciences Computing Facility (HSCF), University of California at Los Angeles, the BMD programs, see Dixon (1964), are being rewritten for on-line usage. The programming was initially carried out for an IBM System 360/75 computer with an IBM 2250 display unit with lightpen and special function keyboard as a console. In September 1968 the HSCF computer was replaced by an IBM 360/91, and an operating system called TORTOS was brought into use. TORTOS consists of a number of modifications and extensions to the standard IBM Operating System/360 with the MVT option. TORTOS provides scheduling, memory management (with roll-in/roll-out to drum), and time-sharing for four basic job priority classes: real-time, terminal, background, and batch. The following statistical programs are now in use at HSCF as terminal jobs: a) sampling from distributions, b) data description (simple regression), c) polynomial regression, d) stepwise regression, e) time series spectrum estimation, f) nonlinear regression, and g) nonlinear regression with potentiometer input. Many other interactive graphical routines that are more specific to biomedical problems are also available at HSCF. Programs (b), (d), (e), and (f) are, or will soon be, available from IBM as Type IV library programs. Figures 9 and 10 show the IBM 2250 screen during operation of the time series spectrum estimation program. in [ACM] ACM Computing Surveys 2(4) Dec1970 view details in [ACM] ACM Computing Surveys 2(4) Dec1970 view details This set of statistical programs is perhaps the best known of all the user-written systems, and has been used extensively on large computer systems The programs were originally written in FORTRAN II with some FAP subprograms for the IBM 7090(94), but they have since been converted for use on several other machines. A completely revised second edition is now written entirely in FORTRAN IV; it operates on the IBM 360 series (requiring 256K bytes to compile under FORTRAN H), and has been converted for other third generation machines as well. All of the programs are of a general nature and are quite extensive in their coverage. Great flexibility of input is achieved by means of FORTRAN card format statements read at run time, as well as standard data forms; the output is labeled in a very readable manner. Since this package occupies a special place m any survey of statistical packages, some comments on its features are in order. The special program section contains ten programs, including hfe-table, bio-assay and contingency table, and Guttman-scale analysis. The section on time series analysis covers amplitude and phase analysis as well as auto-covariance and power spectral analysis. The section on variance analysis includes programs for handling one-way designs, factorial designs, multiple range tests, and covariance analysis. Also included is an especially powerful, general linear hypothesis program, which can be used to analyze many designs Almost all of the programs have the capabilities to perform transformations on the input data and to accept data from cards or tape The user documentation for the package is admirably complete and understandable. The BMD X-series programs are described in a supplemental manual published m 1969 These programs have been developed since September 1965, and m some eases supersede earlier BMD programs. The additional capabilities provided by the Xseries programs strengthen the base BMD package considerably. At present, a BMD P-series is under development. in [ACM] ACM Computing Surveys (CSUR) 4(2) June 1972 view details INTRODUCTION In the middle sixties there was a revolution in behavioral science computing brought about by the introduction of software systems, or 'packages', on second and third generation batch equipment (Most notably, BMD, SPSS, OSIRIS, DATA-TEXT). These systems offered the analyst a higher-level language designed specifically for the problems of behavioral science data handling and analysis, thus freeing him from the details of programming, data reformatting and using subroutine libraries. A short time later a number of data-management and analysis systems appeared on time-shared computers. (Most notably, ADMINS, DATANAL, TRACE, IMPRESS, TROLL) These systems seemed to hold further promise for the behavioral scientist wanting to analyze data. An analyst would now be able to interact with his data: to test hypotheses, explore for and formulate new hypotheses, test again and so on. In addition, because of immediate feedback on errors these interactive systems were expected to reduce the learning investment needed to be able to communicate with the computer. Unfortunately, to the broader behavioral science community, the promise of interactive systems is still just that a promise. A number of factors contributed to this situation, among which were: i) time-shared computers were not widely available; 2) the cost of using these interactive systems was high compared with the batch systems; 3) the interactive systems did not, in general, have the breadth of capabilities in both data handling and statistics as the batch systems; and, 4) analysis techniques that took advantage of the power of interactive computing were just beginning to be developed. Going into the middle seventies, we feel the situation is ripe for change. The Cambridge Project is a joint effort by computer scientists, behavioral scientists, and statisticians from M.I.T. and Harvard to bring about the change. Janus is an attempt to provide a powerful interactive data handling and analysis tool for the behavioral scientist. Its design grew out of experience with two interactive systems, ADMINS Mark III and DATANAL, and one batch system, DATA-TEXT. In addition, Janus was influenced by systems and ideas from outside of the behavioral science tradition; for example, the relational data work of S. D. McIntosh and D. M. Griffel and that of E. F. Codd. Janus is one of the subsystems being developed for the Cambridge Project Consistent System (CS). The CS also contains other data analysis programs and subsystems, modeling programs, an urban-planning subsystem, an econometrics analysis subsystem and others. in [ACM] Proceedings of the 1972 Annual Conference of the ACM view details in [ACM] CACM 17(06) (June 1974) view details in [ACM] CACM 17(06) (June 1974) view details INTRODUCTION Social scientists are a large group of potential users of computer graphics. They make heavy use of computers in both research and teaching. When they use computers they are accustomed to sophisticated software and unsophisticated devices. They rarely write special purpose programs and only a small subset use special hardware. Instead, they rely on high quality, multipurpose-software packages of canned programs. These and other characteristics have implications for the design of graphics software and hardware. These potential users will expect graphic software to be easy to use, powerful, inexpensive and device independent. If computer graphics evolves in these directions social scientists will become major consumers. Computer graphics can aid the social scientist who is involved in data analysis in two distinct ways. First, computer graphic displays can be used after the research process to communicate research results to others in publications, in presentations and in teaching. This use requires that graphs be clearly labelled and adjustable in format. Second, computer graphics can be used during the research process to communicate results to the primary researcher. This can be accomplished by embedding the graphics within data analysis software. This use requires that the graphics be obtained easily in addition to the traditional printout of results; ease of use is more important than flexibility of format or elegance of labelling. We have developed graphic software to satisfy the perceived basic needs and expectations of the social scientists and which have been well received and utilized at the University of Michigan. It is embedded in the interactive statistical system, MIDAS,[1] and appears as a stand alone interactive prompter, GRAPH.[2] In the future this software may be available more generally over a national computer network. Our graphic software includes a set of modules which correspond to the visual components of a display, such as an axis with tic marks and labels or a legend. This modularity provides extensibility of the set of available displays with minimal effort. Our software in turn employs the University of Michigan Integrated Graphics library [3] to define a visual component as a set of device-independent lines or text strings. The Integrated Graphics routines handle all graphic device communications including that for storage tube terminals and the CalComp ploter. We are not attempting in this paper to describe the internal design of the software we have produced. Our purpose is to describe the social science computing environment and the external form of the graphic software designed both to fit that environment and to introduce social scientists to computer graphics. in Proceedings of the 3rd annual conference on Computer graphics and interactive techniques, July 14-16, 1976, Philadelphia, Pennsylvania view details in Computer Graphics 10(2) Summer July 1976 view details in Computer Science and Statistics: Tenth Annual Symposium on the Interface, pp. 265-70. Edited by D. Hogben and D. W. Fife. (National Bureau of Standards Special Publication 503.) Washington, DC: National Bureau of Standards; 1978 view details in Computer Science and Statistics: Tenth Annual Symposium on the Interface, pp. 265-70. Edited by D. Hogben and D. W. Fife. (National Bureau of Standards Special Publication 503.) Washington, DC: National Bureau of Standards; 1978 view details The BMD programs were developed some 25 years ago under the initial direction of University of California, Los Angeles, (UCLA) Department of Biostatistics faculty members Frank Massey, Jean Dunn, and W.J. Dixon. Early development was funded by grants from the National Chemotherapy Service Center of the National Cancer Institute, the National Institutes of Health, and later. by the National Science Foundation. Because the early work on BMD programs was supported by the U.S. government. the earlier BMD programs are now in the public domain, and BMDP no longer supports or maintains them. Many universities, however. have taken the old programs and modified them, added their own subroutines, or both. Since BMD was the first widely accepted statistical package, most Systems developed subsequently have been based, at least in part, on our original work. BMDP came into being at UCLA in 1975. when we completely reworked the System, combining the best of the BMD programs and our experimental series, the BMDX programs. The "P" was added to emphasize that BMDP would no longer use fixed-format program control. By adding a Parameter language, we greatly increased the ease of use and unified the program presentations. At this time, the U.S. government insisted that we copyright the programs and become self-sustaining. so the BMDP programs are not in the public domain. Thus BMDP is a registered trademark of BMDP Statistical Software; BMD is not. BMDP Statistical Software continually supports and enhances the BMDP programs by including new computing and statistical routines, making it an extremely reliable and efficient statistical package. The recent addition of regression diagnostics to the 2R program is an example of our staying abreast of the latest findings. Great Progress has been made in statistical research since the early days of the BMD's. BMDP has always had a commitment to original research. and many staff members and associates have contributed to program development, including Alan Forsythe. Robert Jennrich. John Hartigan. Ray Mickey. R.L. Andersen, and of Course. W.J. Dixon. in Computer Science and Statistics: Tenth Annual Symposium on the Interface, pp. 265-70. Edited by D. Hogben and D. W. Fife. (National Bureau of Standards Special Publication 503.) Washington, DC: National Bureau of Standards; 1978 view details |