The first major experimental comparison of approval voting with regular plurality voting occurred in the 1985 annual election of The Institute of Management Sciences (TIMS). In approval voting a person votes for (approves of) as many candidates as desired, the winner being the candidate with the most votes. By permitting more votes than the number of positions to be filled, approval voting collects more information from the voter than does plurality voting. This can make a difference, for example, when three candidates compete for a single office. In such situations two candidates with wide but similar appeal sometimes split a majority constituency so that, under plurality voting, a minority candidate is elected. By contrast, approval voting is likely to identify the candidate who is most broadly acceptable to the electorate as a whole. In the TIMS experiment society members received an experimental approval ballot along with their official plurality ballot. Two contests involved three candidates running for a single office and a third, five candidates for two positions. Surprisingly, in two of the three contests, approval voting would have produced different winners and neither of the changes was of the type usually emphasized in the approval voting literature. The experiment demonstrated the practicality of approval voting and showed that it can elect a set of candidates different from that which plurality voting would. Direct comparison of ballots makes it possible to determine why the experimental switches occurred. It is shown that in each reversal the approval winner had broader support in the electorate than the plurality winner. The experiment also provided empirical data on how voters distribute approvals across candidates and indicated that, in this case, their behavior was roughly, but not exactly, consistent with theoretical analyses of voting efficacy.

A heuristic scheduling policy is introduced for multi-item, single-machine production systems facing stochastic, time-varying demands. The policy, which we term the dynamic cycle lengths heuristic, integrates feedback control based on the monitoring of inventory levels with the maintenance of economic production cycles. The policy is applied time period by time period to make decisions concerning which items to produce in what quantities during the next time period. These quantities reflect production cycles revised each time period in response to differences between projected and actual inventory levels and in response to changes in demand rates. We extend the basic scheduling policy in order to integrate its use with tactical planning. We also discuss the results of simulation tests of the heuristic against other scheduling policies.

This paper considers the problem of scheduling n jobs, each having a processing time, a due date and a weight, on a single machine to minimize the weighted number of late jobs. An O(n log n) algorithm is given for solving the linear programming problem obtained by relaxing the integrality constraints in a zero-one programming formulation of the problem. This linear programming lower bound is used in a reduction algorithm that eliminates jobs from the problem. Also, a branch and bound algorithm that uses the linear programming lower bound is proposed. Computational results with branch and bound algorithms that use this and other lower bounds and with a dynamic programming algorithm for problems with up to 1000 jobs are given.

We consider the use of a nonhomogeneous Poisson process in modeling time-dependent arrivals to service systems. In analyzing a set of actual arrival times corresponding to epochs of calls for on-line analysis of electrocardiograms, we found that approximating the rate function by an exponential polynomial (or exponential trigonometric polynomial) suggested in the literature inadequate. This was caused by the nature of our data which contained frequent cyclic abrupt changes in arrival rates. In this paper, we propose the use a piecewise polynomial to represent the rate function. We present two maximum likelihood estimators for estimating the parameters of the piecewise polynomial—one based on arrival times, and another based on aggregated counts, and a numerical method for carrying out the computation. We use a procedure based on thinning for generating arrival times from such a process. For hypothesis testing, these are combined to produce critical values for the Kolmogorov Smirnov statistic by a Monte Carlo simulation. While our approach is presented and discussed in the context of a specific case, the results are applicable and observations relevant to many other systems sharing similar patterns of variation.

To analyze simulation runs which use the same random numbers, the blocking concept of experimental design is not needed. Instead, this paper applies a linear regression model with a nondiagonal covariance matrix. This covariance matrix does not need to have a specific pattern such as constant covariances. A simple example yields surprising results. The paper proposes a new framework for the error analysis. This framework consists of three factors (namely, common random numbers, replication, model validity), each with three levels.

Bultez and Naert (Bultez, A. V., P. A. Naert. 1988. When does lag structure really matter … indeed? Management Sci. 34(July).) provide an analytical example and several simulations that show profits to be more sensitive to advertising lag structure misspecification the higher the advertising goodwill elasticity. We explain how this result is consistent with our (Magat, McCann, and Morey [Magat, W. A., J. M. McCann, R. C. Morey. 1986. When does lag structure really matter in optimizing advertising expenditures? Management Sci. 32(February) 182–193.]) result that a given amount of advertising error induced by misspecification causes greater reductions in profits the lower the advertising goodwill elasticity. We also use their simulation results to identify the pattern of prior advertising used in estimation as an important determinant of when misspecification matters in estimating advertising lag structures.

This paper generalizes Stein's (Stein, C. 1973. Estimation of the mean of a multivariate normal distribution. Proc. Prague Sympos. Asymptotic Statistics, September 1973.), Rubinstein's (Rubinstein, M. 1973b. A comparative static analysis of risk premiums. J. Bus. 46(October) 604–615; Rubinstein, M. 1976. The valuation of uncertain income streams and pricing of options. Bell J. Econom. Management Sci. 7(Autumn) 407–425.), and Losq and Chateau's (Losq, E., J. P. D. Chateau. 1982. A generalization of the CAPM based on a property of covariance operator. J. Financial and Quant. Anal. 17(December) 783–797.) covariance formula to the case where both variables are functions of multivariate normal random variables. The resulting formula is extremely useful for either implicit functions of, or nonpolynomials of, multivariate normal random variables, such as exponential functions. An application of the use of the generalized Stein/Rubinstein covariance formula to the estimation of real systemic risk is provided to illustrate the results.

Despite the increasing popularity of decision support systems (DSS), effectiveness of such systems remains unproven. Past research claiming usefulness of the DSS has relied largely on anecdotal or case data. The relatively few laboratory experiments report mixed results regarding the effects of a decision aid. This study reviews the results of prior investigations and examines the effectiveness of DSS-aided decision makers relative to decision makers without a DSS over an eight-week period. An executive decision making game was used in two sections of a business strategy course. Three-person teams in one section used a DSS while the teams in the other section played the game without such an aid. Various measures of decision quality were recorded. Overall, the groups with access to the DSS made significantly more effective decisions in the business simulation game than their non-DSS counterparts. The DSS groups took more time to make their decisions than the non-DSS groups at the beginning of the experiment. However, the decision times converged after the third week. The DSS teams reported investigating more alternatives and exhibited a higher confidence level in their decisions than the non-DSS groups, but these differences were not statistically significant.

The purpose of this paper is to compare the abilities of two competing analytical techniques—mathematical programming and statistical regression—to shed light on the union/nonunion productivity differential in U.S. surface coal mining. The programming approach has the virtues of being nonparametric (and thus extremely flexible) and of being able to provide a decomposition of productivity differentials into three components—differences in technical efficiency, differences in scale efficiency, and differences in congestion. Identification of these three components provides an aid to management in its search for the sources of, and remedies for, productivity gaps. The econometric approach, on the other hand, is neither flexible nor does it provide such a decomposition; its chief virtue lies in the fact that, being stochastic, it allows for the presence of noise and measurement error that plagues most if not all empirical data. Thus, the two approaches have complementary virtues. The two techniques are used to investigate productivity in two samples of U.S. surface coal mines. The programming approach finds a large and significant positive union/nonunion productivity differential. This differential is due primarily to greater congestion occurring in the smaller nonunion mines in one sample, and to scale inefficiency in the smaller nonunion mines in the other sample. The econometric analysis finds nearly the same union/nonunion productivity differential, but sheds no light on the composition of, and thus the cure for, the differential.