What's so Scientific about MS/OR? Author(s): Randolph W. Hall Source: Interfaces, Vol. 15, No. 2 (Mar. - Apr., 1985), pp. 40-45 Published by: INFORMS Stable URL: https://www.jstor.org/stable/25060669 Accessed: 06-09-2018 13:53 UTC JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at https://about.jstor.org/terms INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Interfaces This content downloaded from 128.82.252.58 on Thu, 06 Sep 2018 13:53:31 UTC All use subject to https://about.jstor.org/terms What's So Scientific about MS/OR? RANDOLPH W. HALL Department of Industrial Engineering and Operations Research Etcheverry Hall University of California Berkeley, California 94720 Basic precepts of research in the natural sciences can be used to guide research in MS/OR. To prosper as a science, mathematical elegance is not enough. Attaining an improved understanding of the real world is essential. The Management Science/Operations definition has been addressed by Ackoff [1971], Churchman [1970], Eilon [1975] and Hicks [1973] (summarized in Zahedi Research (MS/OR) profession has often been compared to the natural sci ences. This tradition is rooted in the ideals [1984]). Perhaps a more relevant question of its founders, who sought to introduce is, What can be learned from the natural the principles of the sciences to manage sciences that can add to the success of the rial decisions. And, certainly, the word MS/OR profession? science in the title "Management I will attempt to show how basic pre Science" continues to invite such cepts of research in the natural sciences can be used to guide research in MS/OR. I comparisons. The official definition of operational re believe a greater understanding of these search by the Operational Research Soci precepts can provide needed focus for the ety of Britain reflects this orientation: profession and help resolve the recent de Operational Research is the application of the bates between "practitioners" and methods of science to complex problems in the "theoreticians," which seem to have split direction and management of large systems of the profession. men, machines, materials and money in indus Science try, business, government and defence. The question of whether MS/OR fits this Webster's dictionary defines science as "a possession of knowledge as distin Copyright ? 1985, The Institute of Management Sciences PROFESSIONAL ? MS/OR IMPLEMENTATION 0092-2102/85/1502/0040$01.25 INTERFACES 15: 2 March-April 1985 (pp. 40-45) This content downloaded from 128.82.252.58 on Thu, 06 Sep 2018 13:53:31 UTC All use subject to https://about.jstor.org/terms SCIENTIFIC? guished from ignorance or misun derstanding," or "a department of sys temized knowledge as an object of study." The persons who coined the term force; doubling the mass of both quad ruples the force. Next, it says the force decreases with the square of the distance, the second definition in mind. Business r; the force diminishes quite fast when the two particles are separated. Finally, the model demonstrates that force de schools had been common on university campuses for more than four decades be pends on a universal gravitational con stant, G, which does not depend on the management science most likely had fore the inception of MS/OR, and certainly a considerable body of knowledge regard ing the practice and theory of manage ment had evolved. The key to MS/OR is not the mere possession of knowledge but the systemized attainment of knowl edge. And although different practitioners take different approaches, the three key steps seem to be (1) Modeling, (2) Evaluating, and (3) Deciding. In MS/OR, systemized knowledge is re flected in better decisions. Modeling As in the natural sciences, the purpose behind modeling in MS/OR is to under stand observable phenomena. Effective models abstract the most important facets of a "problem" and present them in a form that is easy to interpret [Friedman 1935]. They make sense out of the detail and disorder underlying the real world. Perhaps one of the most important models of the natural sciences is New ton's Law of Universal Gravitation: We should remember that MS/OR, unlike mathematics, is based on real events. particle composition or anything in the surrounding environment. The density of the material, electrical conductivity, ob jects in the intervening space between the particles, and a myriad of other factors can be ignored. Of course, the model does not account for all details. For instance, it is only ap proximate if the particles are objects with non-spherical shapes and the model ig nores relativity. But even so, no one would argue that it is not one of the most important models in physics. In the MS/OR profession, there is prob ably no model which exceeds the impor tance of the Economic-Order-Quantity (EOQ) model: Cost-**+*-& Q 2 Like Newton's Law in physi is simple and highlights im tures of the real world. It ide Much can be learned from this simple model. First, it shows that the force be tween two particles equals the product of the mass of the two particles. Thus, doubling the mass of either doubles the relationships between the t week and the ordering cost rate, the order quantity, an tory carrying charge. It also single model can be used fo March-April 1985 41 This content downloaded from 128.82.252.58 on Thu, 06 Sep 2018 13:53:31 UTC All use subject to https://about.jstor.org/terms HALL A model should also be verifiable and orders and that the only information needed concerns certain attributes of the based on observable phenomena. It must order (for example, demand rate). Of capture the essence of a problem faced by course, the model ignores details of the practitioners of MS/OR [Dando & Sharp 1978]. And it must include the most im real world which might be considered im portant, such as variations in the demand portant parameters, decision variables, rate. But like Newton's law, no one would and objectives, accurately illustrating their argue that the EOQ model is not one ofrelationships. We should remember that MS/OR, unlike mathematics, is based on the most important in MS/OR. To be useful, an MS/OR model must real events. It is not our role to develop possess the same qualities as models in and judge models which properly belong the natural sciences, the most important in other disciplines. being that it is Finally, as a criterion for publication, ? Understandable, the phenomena underlying the model ? Verifiable, and must be reproducible. The model must be ? Reproducible. sufficiently general that it applies to con The extent to which a model can be un ditions which occur outside the author's derstood depends on the tools available"laboratory." If the situation is unique, for evaluation. However, there is inherent the model has no broad appeal. Why tell the world about it? value in models which can be interpreted on inspection. In the 1960s, considerable Evaluating The role of evaluating is to extract in formation from the model. Sometimes, The decision maker probably even simple models can be difficult to does not want a single soluunderstand. Understanding more compli tion but information on sevcated models, with many equations, eral alternatives. parameters, and decision variables is even more difficult. For this reason, much of effort was expended by many in the the research in MS/OR has been devoted MS/OR profession to model the behavior to developing tools for interpreting of automobile traffic. The most successful models through evaluation. These range models examined traffic from the macro from drawing graphs, to simulation, to scopic point of view and illustrated impor advanced optimization techniques. Their tant similarities between the flow of traffic common aim is to make sense out of the and the flow of fluids. Less successful model. models examined the behavior of indi Evaluating typically entails two, often simultaneous, activities: vidual drivers with complicated queueing alternatives, and expressions. One might argue that the? Identifying lat ter models are more accurate, yet they ? Calculating objectives. The most familiar technique for identify clearly have less value. They are too dif ing alternatives is optimization. This ficult to interpret. INTERFACES 15:2 42 This content downloaded from 128.82.252.58 on Thu, 06 Sep 2018 13:53:31 UTC All use subject to https://about.jstor.org/terms SCIENTIFIC? process generally yields a single solution which maximizes or minimizes a single objective function. Solving simultaneous equations and many heuristics are similar in aiming to identify a single "solution" to a well-defined mathematical problem. The most commonly used technique for identifying multiple alternatives is sen sitivity analysis. This process can show how the optimum changes when model parameters change or can provide alterna tive solutions in the vicinity of the optimum. The predominance of optimization is disturbing, not just because models are only abstractions of the real world [ Ackoff 1979], but because optimization does not provide adequate information for making a decision. Its aim is to find a single solu tion. The decision maker probably does not want a single solution but information on several alternatives. Although sensitiv ity analysis greatly increases the effective ness of optimization, it too is deficient. It only yields alternative solutions in the vicinity of the "optimum." What the deci sion maker generally needs are truly unique solutions, which offer distinct alternatives. Very little MS/OR research has been devoted to identifying multiple alterna tives. One place to begin may be in the solution process itself. In optimizing the objective, linear program algorithms iden tify many feasible solutions. Each solution is a viable alternative, which the decision maker might prefer to the "optimum." Going one step further, new algorithms might be designed to identify distinct al ternatives. Descent algorithms could be replaced with algorithms designed to jump around to different parts of the feas ible region, thus identifying distinct op tions. There is much to be learned from the optimization process, which might eventually yield better decisions. The second step of evaluating involves calculating quantifiable objectives for each alternative. The purpose is to provide the decision maker with a measure, or meas ures, for comparing the alternatives. Like modeling, it leads to an improved under standing of the underlying problem. Therefore it is crucial that this information Very little MS/OR research has been devoted to identify ing multiple alternatives. be presented clearly and concisely. Too much information is overwhelming, but too little diminishes the model's credibil ity. For instance, management would not be satisfied with a presentation which gave only the profits expected from alter native marketing strategies. Neither would they like to see information on ir relevant factors, such as employee com pensation plans and utility rates. Mana gers prefer to see clear and concise infor mation on the most important criteria, such as cost, availability, and sales. Deciding The aim of MS/OR is to make better de cisions. Therefore, the final step in the process, deciding, is the most crucial. Yet it is also the most elusive. There is hardly an MS/OR analyst alive who has not been dismayed when management failed to implement the "optimal" strategy, or con fused as to why it failed to choose the op March-April 1985 43 This content downloaded from 128.82.252.58 on Thu, 06 Sep 2018 13:53:31 UTC All use subject to https://about.jstor.org/terms HALL owner would benefit from models that tion its own model determined to be best. offer assistance in cutting the carpet. The key to good decisions is knowledge Combining the knowledge obtained from and judgment. Modeling and evaluating modeling with the judgment of the store form a systemized approach for gaining owner would likely provide the best knowledge; judgment is acquired through result. experience. Much of the MS/OR literature is di Perhaps the most important questions management are not so well rected towards problems which do not facing re defined as either the shortest path or cut quire judgment. These are the ones which can be formulated with well-defined ob ting stock problem. Although there might be a related problem which lends itself to jective functions and solved automatically automation, the objective function never seems to capture all of the important fea In the meantime, the carpet store owner would benefit tures of the decision. One example is the assignment problem. A great body of lit from models that offer assiserature exists on assigning employees to tance in cutting the carpet. tasks so that cost is minimized or return is maximized, and there exist efficient al gorithms for solving these problems. Yet with efficient algorithms. Perhaps the best example is the shortest path problem. these In algorithms totally ignore the human choosing a travel route, most people element. The first time I had to assign employees would agree that the objective is to to jobs was when I served as workshift minimize the sum of the costs correspond ing to the arcs of the path. Therefore, manager in a student housing coopera tive. My most important consideration companies are willing to rely on computer software (based on shortest path al was not maximizing a linear objective gorithms) to automatically select their function, but insuring that people as routes. signed to the same job were compatible and that no one felt cheated by his or her In a slightly different category, there are assignment. I am still baffled at the problems which are easy to formulate, but thought of writing an objective function to difficult to solve. Although in principle capture these factors. Yet I feel I did a bet these problems can be automated, there is ter job because I knew how to write an often benefit in allowing a human being assignment matrix and apply the stepping to pick the solution. A carpet store owner stone method to resolve bottlenecks. would not argue with the objective of the The most difficult questions are neither cutting stock problem but may be dissatis well defined, nor is there a related well fied with the solutions provided by avail able software. Whether or not this prob lem can be solved optimally in a reason able computation time is an open ques tion. In the meantime, the carpet store defined problem which can be optimized. One example is the facilities layout prob lem. Although various heuristics exist for arriving at feasible layouts, none are op INTERFACES 15:2 44 This content downloaded from 128.82.252.58 on Thu, 06 Sep 2018 13:53:31 UTC All use subject to https://about.jstor.org/terms SCIENTIFIC? timal in the classic sense. And even the heuristics do not come close to capturing all of the relevant parameters, decision variables, or objective functions. Even so, the models provide valuable information which helps the decision maker identify better layouts. Of the four types of problems men tioned here, only the first lends itself to natural sciences. Further attention by MS/OR theoreticians to the principles of science should not only improve the qual ity of research, but also make MS/OR more effective at addressing the complex problems facing practitioners today. References Ackoff, R. L. 1971, "Frontiers of management science/' The Bulletin (TIMS), Vol. 1, No. 2, pp. 19-24. automation. All others depend on both the judgment of the decision maker and the knowledge gained from modeling and evaluating. Optimization may be a poor Ackoff, R. L. 1979, "The future of operational research is past," Journal of the Operational Re search Society, Vol. 30, No. 2, pp. 93-104. tool for approaching these more difficult No. 2, pp. B37-B53. Dando, M. R. and R. G . Sharp 1978, "Opera tional research in the UK in 1977: The cause problems. It discourages the analyst from fully interpreting the model and tends to mask important trade-offs and principles. In making complex decisions, understand ing trade-offs and principles is far more crucial than knowing the model's optimum. Churchman, C. W. 1970, "Operations research as a profession," Management Science, Vol. 17, and consequences of a myth?" Journal of the Operational Research Society, Vol. 29, No. 10, pp. 939-950. Eilon, S. 1975, "How scientific is OR?" Omega, Vol. 3, No. 1, pp. 1-8. Friedman, M., editor, 1935, "The methodology of positive economics," in Essays in Positive Economics, The University of Chicago Press, Conclusions Chicago. Perhaps the most important similarity between MS/OR and the natural sciences Hicks, D. 1973, "Education for operational re is the goal of systematically attaining an Zahedi, F. 1984, "A survey of issues in the improved understanding of the real world. Mathematical elegance is not enough. To prosper as a science, under standing is essential. Optimization has predominated re search in MS/OR for far too long. It is but one technique for addressing one part of the MS/OR process. Moreover, it is defi search," Omega, Vol. 1, No. 1, pp. 107-116. MS/OR field," Interfaces, Vol. 14, No. 2, (March-April) pp. 57-68. cient in that it does not provide sufficient information for making crucial decisions. The most complex decisions require in formation on many alternatives, as well as understanding of basic trade-offs and principles. Optimization alone does not provide this information. MS/OR has much in common with the March-April 1985 45 This content downloaded from 128.82.252.58 on Thu, 06 Sep 2018 13:53:31 UTC All use subject to https://about.jstor.org/terms