In decision analysis we normally consider the value of information to be a constant against which the cost of information is compared. However, when it is possible to buy information sequentially, the value of information is not a constant. Rather, it is a function of the prices of the various pieces of information, or “observables.” When we are faced with a decision and learn one observable, this information not only helps us make the original decision, but also helps us decide if we should pay for more observables. For this reason, the first observable has a value above and beyond that which we would assign if there were no possibility of obtaining additional information. To decide whether or not to buy one observable we must know the prices of all the observables.

A method is presented for the generation of extreme points of regions in the parameter space for which a certain basic solution is optimal. It is shown that for any extreme point the inequalities which determine the adjacent regions are easily generated. A numerical example is worked out in detail.

Conditions of equivalence are established among the following portfolio selection criteria: (1) Mean-Variance (E-V), (2) Second-Degree Stochastic Dominance (SSD), and (3) Mean-Entropy (E-H), for portfolios whose returns are characterized by-Uniform, Normal, and Lognormal probability distributions. We also assume that all portfolios derive from the same family of distributions and consider only the cases where the cumulative density functions intersect. In comparing the three selection criteria, under the three posited probability distributions, we utilize a combination of mathematical and graphical analyses. It is concluded that the three efficiency criteria are equivalent for uniformly and normally distributed portfolio returns. For lognormally distributed portfolio returns, the SSD criterion is optimal. Its efficiency, however, is sufficient but not necessary to establish efficiency for E-V and E-H. On the other hand, given the empirical similarities between E-V and SSD derived portfolios, and the close correspondence between the E-V and E-H criteria, the potential use of E-H becomes more appealing because of its distribution free nature.

Increasing emphasis on the reduction of materials handling costs in the modern plant has led to research into new methods of planning the process type layout in such a way as to minimize these costs. This project compares the performances of three highly rated computer algorithms prescribed for the solution of the plant layout problem with the performances of selected human subjects using the manual and visual methods still used and recommended by industrial engineers for plant layout design. The objective of this comparison is to determine whether there is in fact an advantage to using one of the available computer programs to solve the problem, instead of designing the layout by traditional visual-based methods. These tests, performed under the control of a computer system which accurately recorded the solutions achieved by each subject, show not only that the computer algorithms do not perform better than selected human subjects in the design of plant layouts, but that the human subjects, without the benefit of any prescriptive help from a computer, actually achieve layouts which are stochastically better than those produced by the computer programs.

An approach is presented for the solution of the discrete conveyor model for service time distributions which are general but bounded and where there is no storage at work stations and no recirculation. The model is described by states with the Markovian property and solutions may be derived for moderately sized numbers of operatives.

This paper describes a Bayesian model-discrimination procedure which determines for each consumer the stochastic model of brand choice which is best supported by his purchasing behavior. The Bayesian technique is illustrated by means of two Markov models and two Bernoulli models. We first discuss in some detail how we set priors for the four models. Then we use simulated consumer panel data to demonstrate that the Bayesian technique is a good discriminator among the four models. The technique is then applied to some actual aluminum foil purchase data. We provide estimates of the proportion of aluminum foil consumers in each of the four model segments and the degree to which the size of each segment changes over time. We also show that model discrimination at the individual consumer level has important implications for market segmentation, pricing, and promotion.

A standardized approach to selecting a simple sequencing rule for decentralized application throughout a job shop is developed and illustrated. The sequencing rule is a linear combination of decision factors, each of which is initially assigned a relative weighting. The rule is then used to determine the priority of each job in the queues, and resulting shop costs are determined by computer simulation. The coefficients of the priority function are thereafter modified by a patterned search procedure to find priority coefficients that minimize expected cost per order for a specified cost structure. The cost structure is a combination of multiple response measures for the shop. Rather than leading to a “single best rule” for all job shops, the approach is a “method for finding” a sequencing rule that performs well in any specific job shop situation.

It is well known that when an individual assesses a preference (utility) function, the set of assessed gambles and certainty equivalents is often inconsistent and, if consistent, many preference functions may satisfy the assessments. Mathematical programming is employed to examine properties that might be useful in a sequential determination of the individual's preference function. Specifically, given a consistent set of assessments, if an additional gamble were to be assessed, upper and lower bounds can be found for the probability of the better consequence of the gamble when the certainty equivalent is specified and also bounds for the certainty equivalent of the gamble when the probabilities are specified such that the augmented set of specifications remains consistent. The results are given for general, as well as risk averse preference functions.

The purpose of this paper is to explore the contributions of information systems to the organization. A descriptive model is presented which identifies (1) expected predictors of performance and the use of an information system and (2) the relationship between the use of a system and performance. The results of a study of sales force performance and the use of a sales information system are presented to provide empirical evidence to support the model. The results confirm the general relationships among classes of variables in the model, but specific relations among variables are complex and depend heavily on the environment of the organization.

Two simple variance reduction techniques are discussed, viz. antithetic variates and common random numbers. Their joint application creates undesirable negative correlations between the responses of two simulated systems. Therefore three alternatives are considered: antithetics only, common random numbers only, antithetic and common random numbers combined. No alternative is always best as is shown by analytical results for extremely simple systems and simulation results for simple queuing systems. Therefore a procedure is derived that starts with some pilot runs for both systems and estimates which alternative minimizes the variance; at the same time this procedure allocates the limited amount of computer time to the two systems in an optimal way. Results of the application of the procedure to several queuing systems are presented. Because of certain disadvantages of the procedure we may decide to select alternative 1 (antithetics only) a priori. Then the procedure can still be used for the optimal computer time allocation.