THE STATISTICAL SOFTWARE NEWSLETTER 191

Computational or Statistical Computing,

-° ~* is that the question.

Carlo Lauro

Dip. Matematica e Statistica, Universita di Napoli, Via Cintia, 1-80126 Napoli, Italy

Past President of IASC

During last COMPSTAT Symposia, a recurrent cal Computing. The latter, founded in 1977, in- matter of debate regards the meaning, the contents eluded the previous European group known as and the future of this recognized and successful COMPSTAT, which started this series of biannual meeting that has come by now, to its XII edition. scientific meetings with the first one being held in A crucial point of this debate concerns the defini- Vienna in 1974. tions of Computational statistics and statistical The Statute of the section of Statistical Computing Computing. Actually, the lack of an agreement on of ASA reports as objectives "Encouraging the ap- their meanings allows for confusion in the debates plication of computer hardware, software and sys- on the contents of both COMPSTAT and the IASC tems to statistical problems .... Encouraging re- journals. search in statistical computing and communication The two terms are often used as being inter- of the results". The objectives of IASC are "to changeable or mutually inclusive. Sometimes, they promote the theory, methods, and practice of sta- are understood as being different, with the former tistical computing and to foster interest and knowl- meaning a more theoretical or academic approach edge in effective and efficient statistical computing of the European School, with respect to the more through international contacts.". applicative one of the latter, followed mainly by As it can be seen, both associations rely to the ap- the North American School. plication of computers to statistics but IASC, dif- From an historical point of view, the birth of the ferently from ASA, puts a major accent on the de- conference activity in the area of interest may be velopment of the theoretical and methodological dated in 1967 when the Interface Symposium of aspects of Statistical computing. Does this point and Statistics took place in At- suffice to justify the names COMPSTAT and lanta. The term Statistical Computing has been CSDA (official journal of IASC) as an expressed later adopted by the two major scientific associa- preference of IASC towards more theoretieal as- tions in this area, i.e. the Statistical Computing pects (Computational Statistics)? Such a feeling section of the American Statistical Association would not be supported by the term recently founded in 1972, and the corresponding section of adopted by the Eastern Asian Regional Section of ISI known as International Association for Statisti- the IASC (EARS) which refers to its conference, 192 THE STATISTICAL SOFTWARE NEWSLETI'ER corresponding to the COMPSTAT of EARS, as with some brief references to programming lan- STATCOM. guages and techniques). Therefore, do we have to believe that Computa- While browsing on Intemet, I noticed that the tional Statistics and Statistical Computing are con- Ph.D. courses in this area mostly refer to the term sidered as being interchangeable by IASC? It Computational Statistics. The content of the would be so if we looked just at the definition courses fairly corresponds to the mentioned text- given by Norbert Victor, who states the books with some additions regarding most recent "Computational statistics is the examination of techniques like methods and statistical possible solutions and simplification of the statisti- visualization. When the adopted term is Statistical cian's problem by the application of computer Computing, the relative courses are mainly ori- and/or the method of Computer Science" ented to meta-languages and statistical software. (Computational Statistics - Tool or Science? SSN, I believe the contents of textbooks and Ph.D. 3, 1984, 105-116). courses offer a first idea of what can be meant with Computational statistics, as the result of the inter- This view is no more supported when looking at action between Statistics and Numerical Analysis the most accepted general purpose journals in the that favourized the implementation of statistical area, Computational Statistics and Data Analysis methods as well as the development of new ones, (S. Azen), Computational Statistics (W. Hitrdle), that is, according with B. Efron, to think he un- and Statistics and Computing (D.J. Hand). Their thinkable. respective aims and scopes, but for the specific reference to Data Analysis of CSDA, shows a list This view of mine is close to what Andrew West- of very similar topics ranging from the design of lake (1994) proposed during a debate on the IASC algorithms to implement technical methods, to the list-server: ,,Computational Statistics relates to the development and evaluation of statistical systems, advance of statistical theory and methods through so the newest hardware and software environments the use of computational methods. This includes in Statistics, as well as the application in Statistics both the use of computation to explore the impact of the most recent technologies of Computer Sci- of theories or methods, and development of algo- ence. rithms to make these ideas available to users." It is easy to understand that the above topics have Murray Cameron, in the same occasion, offers a an impact on the different areas of Statistics. They Statistical computing "broad view" including all contribute to define, besides any opposition "most modem applied statistics as well as some between Computational Statistics and statistical aspects of theoretical statistics (e.g. Bootstrap, Computing what I like to call "Statistics in Com- MCMC)" as well as a "narrow view" regarding the puter Age" (see my discussion to the paper of N. "Computational aspects of statistical methods such Victor in SSN, 3, 1984, 119-120). as languages, environments, libraries, packages, to support this work." Nevertheless, the titles of the very few reference This definition, mostly referred to the activities of books in the area are helpful in clarifying the de- users of Statistics, neither seems general enough bate. Kennedy and Gentle (Marcel Dekker, Wiley, nor takes into account the evolution of the Com- 1980), in fact, prefer the term Statistical Comput- puter Science tools which offer new and more effi- ing, Thisted (Chapman and Hall, 1988) adopts the cient ways for an effective Statistical computing. term "Computational Statistics", whereas Main- Although Computational statistics and Statistical donald (J. Wiley, 1984) is for "Statistical Compu- Computing so far argued show some overlappings tation". Although he different titles, the books are characterized by a strong common structure mainly (see Figure 1) in their possible definitions, as the based on numerical methods and the algorithms for latter develops from the former, it deserves a more solving statistical problems (random numbers, specific definition. computational ,methods in linear algebra, optimi- Actually, Statistical computing represents the natu- zation methods, smoothing and density estimation ral evolution of the traditional computations in the THE STATISTICAL SOFTWARE NEWSLEYfER 193 activity of statisticians. This activity now takes ad- nitions can not be sharply drawn bust be necessar- vantage from the new computer and software envi- ily flexible in order to take into account the con- ronments thus increasing the use of Statistics in tinuous changes and improvements characterizing other fields. The new Computer Science technolo- Statistics in computer age. gies extend Statistical Computing to other basic After clarifying the terms of the question, it seems tasks of Statistics like data collection and dissemi- to me that the next COMPSTAT meetings as well nation (EDI, Networks), their organization and as the official IASC journals should refer to the management (SDBMS), interpretation of results Statistics in Computer Age at large, including, in a and reporting (AI), disclosing new perspectives for proper balance, themes of Computational Statistics Data Analysis though computer graphics, neural and Statistical computing without forgetting the networks and so on. applications to different disciplines according with I therefore prefer to define "Statistical Computing the mentioned IASC objectives. as the application of Computer Science to Statis- tics", and, on the other hand, "Computational sta- tistics as aiming at the design of algorithm for im- plementing statistical methods on computers, in- cluding the ones unthinkable before the computer age (e.g. Bootstrap, Simulation), as well as to cope with analytically intractable problems". Figure 1 highlights these definitions in terms of interactions between disciplines showing also a significant area of multidisciplinar research. It must be noticed that even if the emphasis of the IComlputati above definitions is on Computer Science or/and Numerical Analysis, the context is always statisti- cal. Obviously, the contours of the previous deft- Figure 1: Statistics in Computer Age

* Editorial: Statement presented during COMPSTAT '96