▼ Our work, comprising three essays, examines supply chain agent performance in a variety of decentralized systems under both stochastic and deterministic customer demand. In the first two essays, we develop models for both periodic and continuous review inventory policies when the decision-making rights are split between a supplier and a retailer. The second essay also examines a vendor-managed inventory (VMI) agreement. The last essay proposes a novel approach to workload balancing for a company that faces deterministic nonstationary demand and has little, or no, ability to hold inventory. The first two essays seek to answer the following research questions: 1) when does decentralized decision making result in the greatest loss in supply chain performance and 2) what effect does the distribution of channel power have on system and individual agent performance. Channel power here refers to an agent’s relative ability to control the decision making process and is modeled using a game-theoretic framework. We characterize optimal policies where possible and we use numerical analysis to generate insights. We find that, from a supply chain perspective, asymmetric power decision structures lead to better performance and customer service. Surprisingly, we identify cases where the lowest costs are incurred at the agent level when the agent is a follower and not a leader in the Stackelberg game. Our analysis also identifies the environmental conditions when the penalty from decentralized decision making is largest and shows that concentrating channel power with one of the agents can represent a viable alternative to coordination mechanisms, when the latter are costly to implement. In the third essay, we use Fourier analysis and Walsh basis functions to decompose an input workload profile into a portfolio of recurrent insourcing and outsourcing contracts to better achieve some desired constant workload level. In addition, we develop mathematical programs based on principles from goal programming formulations to answer important practical questions such as: 1) how should a company create a portfolio of contracts to balance workload over time; 2) how should the portfolio be customized to reflect special needs with respect to time, volume, etc., 3) what is the benefit of holding inventory as a supplement to subcontracting. Advisors/Committee Members: Fry, Dr. Michael J.

▼ This dissertation consists of three essays that address issues in production and inventory management. The first essay focuses on inventory management. We study a fixed-reorder-interval, order-up-to (R, nT) inventory replenishment policy in a two-stage serial system with stochastic demand at the lower stage. We develop a simulation based optimization procedure to estimate the long-run average cost and optimal parameter values. The numerical results show that the (R, nT) policy is, on average, 4.4% (5.8%) more expensive than the continuous review (r, nQ) policy (lower bounds). The cost difference is much smaller when the setup cost at the upstream stage and the demand rate are larger. The (R, nT) costs are relatively insensitive to the choice of reorder intervals, T, provided the best corresponding order-up-to level, R, is selected. The second essay deals with production scheduling. We consider the computationally-hard, re-entrant flow, cyclic scheduling problem considered by Graves et al. (1983) and Roundy (1992). We present two problem formulations to minimize job flow time (work-in-process), given a target cycle length (throughput). We describe an efficient optimization method and a new ImproveAlignment (IA) heuristic. Numerical experiments indicate that proposed optimization method was significantly faster than CPLEX-8.0 and solved 40% more test instances to optimality within the specified run time and memory limits. The proposed IA heuristic quickly produced solutions which were, on average, (i) 22% better than those from the Graves' et al. heuristic and (ii) within 14% of the optimal. The third essay focuses on resource planning. We examine a single end-product, discrete-time inventory replenishment problem in a material requirements planning (MRP) environment with demand uncertainty and supply capacity limits on replenishment orders. We develop a simulation-based optimization approach and two novel heuristics. We also evaluate the traditional MRP and safety stock approaches for this problem. Computational experiments show that the two novel heuristics perform very well (on average within 0.06% and 0.66% of optimal, respectively); traditional MRP and safety stock approaches incur higher costs, on average, 45% and 12.05% higher than optimal, respectively. We also provide managerial insights on the effects of different input factors. Advisors/Committee Members: Rao, Uday S.

▼ Two empirical phenomena motivate our study. First is managers’ ongoing pursuit of effective change management technology despite the gigantic normative body of knowledge on the subject. This pursuit is intensifying as planned changes become more complex, and as pressure to execute mounts. Second, an increasing number of organizations are experimenting with “self-assessment” – a process of evaluating the effectiveness of organizational systems with little or no outside help. The quality movement in particular appears to have stimulated organizational interest in the self-assessment approach. In this investigation we cross these two concepts and consider self-assessment as structure for managing change and its implementation. The opportunity for theoretical contribution is significant, since the literatures on both self-assessment and change implementation are relatively thin. In the first part of the study, we develop and validate a model of change process. Such a model is important since it defines what aspects of change process get assessed. Few change process models have been validated rigorously in the literature. We investigate the model’s conceptual validity by comparing its content to a number of research streams. Then, we evaluate the unidimensionality and predictive validity of an associated measurement model by employing various quantitative methods. We find both the conceptual model and the measurement model to exhibit considerable validity. With a change model defined and understood, we then investigate the self-assessment process in which the model is used. This research is mostly exploratory, and utilizes qualitative methods to define the theoretically salient features of the self-assessment process and its consequences. We also identify contextual variables that suggest the conditions under which self-assessment is likely or encouraged. An integrative model of self-assessment is generated from 13 theoretical propositions. The model is compared to theory in managerial control systems. It is proposed that self-assessment provides information that facilitates managerial control. Further efforts to integrate managerial control system concepts into change management research should be valuable. Advisors/Committee Members: Evans, James R.

▼ This research presents a novel model of the railway master scheduling problem, applies it to a theoretical study of railway line capacity under representative conditions, and offers a method for applying this model to practical railway scheduling problems. Contemporary market demands for different classes of rail transportation service are not adequately served because of the inability to assess the cost to the network of disparate services, as well as the inability to determine optimal scheduling. Disputes over cost and scheduling also arise between entities managing disparate services that must share common infrastructure. This new model represents railway schedules as binary multicommodity flows on discrete time scales, mapped on a hypergraph. The objective function calculates network value as a linear combination of viable train paths, en route delay, and destination tardiness. Studies are presented that estimate the cost of imposing a non-conforming, high speed train on an existing congested, homogeneous railway network flow. The analysis is provided for two formats of single track railway and for one common format of double track railway. A significant benefit is demonstrated when track networks are configured to allow complex train interactions that are not limited to pairwise meets and passes. In addition, a novel claim is made that under some conditions of single track operation, a higher speed non-conforming train is less costly to the network. A method is proposed and demonstrated for applying this model to a practical railway network scheduling problem. When applying the model, the primary decisions are the choice of track segment boundaries and the dimension of the discrete time unit. Advisors/Committee Members: Magazine, Dr. Michael J.

▼ This study defines Volume Flexibility as: the ability to profitably increase or decrease aggregate production (output) in response to changes in customer demand. We use a triangulated approach to measure and relate volume flexibility to firm performance. Part 1 uses secondary data to measure volume flexibility. Other researchers use variability in sales to measure volume flexibility and conclude that small firms are more volume flexible than large firms are. But, variability in sales essentially measures diversity in the environment, and therefore, it may not be a valid measure of volume flexibility. Our measures consider the combined impact of the firms technology and environmental diversity by incorporating process properties such as inventory levels and costs incurred in meeting sales variation. Using 20 years (1979-1998) of Compustat data on 550 firms in the capital goods industries (SICs 3510-3590), we identify key sources of volume flexibility that give competitive advantages to small firms. But, when we simultaneously account for environmental uncertainty, production technology, and performance, we find that large firms are more volume flexible than small firms are. We also revalidate these findings with a second data set representing 20 years (1979-1998) of data on 2,100 firms in 93 industries. In part 2, we conduct case studies of three small firms in the capital goods industries. We document and assess the drivers and sources of volume flexibility. Our key findings identify drivers of volume flexibility in two categories: external market forces and internal strategic choices. We also identify key sources of volume flexibility and categorize them into a taxonomy of short-term and long-term sources as well as internal and external sources of volume flexibility. Finally, in Part 3, we conduct a field survey of 750 APICS managers to understand the leverage that volume flexibility provides across small and large firms. Our results validate that the short-term and long-term sour ces have a positive impact on a firms volume flexibility. In addition, the results show that volume flexibility has a positive impact on delivery performance and financial performance. Advisors/Committee Members: Raturi, Amitabh S.

▼ Products and services can be thought of as bundles of attributes that are each made up of many levels. Consumers consider tradeoffs between various attributes and levels before making a purchase decision for a product. A popular marketing research technique called conjoint analysis is used to study these tradeoffs. The data on consumer preferences is generally decomposed using a regression procedure into utility values that the consumer associates with each level of every attribute. In the design of a new product, it is critical to determine a judicious combination of attribute levels that is likely to perform well in a market containing competitor products. Using these utility measures, a firm can design products that would best meet consumer expectations and the seller’s objectives. A popular way to measure the success of a new product is to calculate the amount of market share it is expected to capture in a competitive market. Researchers have proposed the share-of-choice problem whose objective is to maximize the number of respondents from a conjoint study who prefer a new product to their status-quo products. The share-of-choice problem is NP-hard. The linear-programming based branch-and-bound algorithm is inefficient in solving real world instances of the share-of-choice problem. Several heuristic procedures to solve the share-of-choice problem have appeared in the literature. In this research, an exact algorithm is proposed to solve real world instances of the share-of-choice problem to optimality. An experiment is designed to test the run-time performance of the exact algorithm using a set of simulated problems that mimic real world problem characteristics. We also suggest potential extensions of the exact algorithm to solve modified versions of the share-of-choice problem, and also to solve the related product line design problem. Advisors/Committee Members: Camm, Dr. Jeffrey D.

▼ Spline curve fitting has attracted a great deal of attention in recent years. It is a powerful method for fitting nonlinear models when the true mean functions are unknown and need to be estimated. This dissertation consists of two essays in Bayesian penalized splines estimation. In the first essay, we propose Bayesian Adaptive Penalized Splines (BAPS), combining features of both penalized splines and regression splines. We first outline a hierarchical Bayesian approach to penalized splines using Markov Chain Monte Carlo for fixed locations of spline knots. We then propose Bayesian Adaptive Penalized Splines, employing a reversible jump Markov Chain Monte Carlo algorithm to adaptively and Simultaneously select the number of knots, the locations of knots, and the smoothing parameter. BAPS is applied to real examples and compares favorably with competing methods in simulation studies. The second essay provides a Bayesian approach to estimating Treasury and corporate term structures with a penalized spline model. First, we estimate the Treasury term structure with a Bayesian penalized spline model, considering both normal and double exponential disturbances. We then estimate the corporate term structure by adding a spread to the estimated Treasury term structure, incorporating the knowledge of positive credit spread into our Bayesian model as informative priors. This is the first work using a Bayesian approach in the term structure literature and several advantages for adopting such an approach are presented. Advisors/Committee Members: Yu, Dr. Yan.

▼ The goal of the printed circuit board job-batching (PCB-JB) problem is to minimize the total manufacturing time (setup time and processing time) required to process a set of printed circuit board jobs on an insertion machine. PCBs are processed on a single-head, concurrent, pick-and-place machine that places components onto a board. The PCB-JB problem is a combinatorial optimization problem that is NP-hard thereby, in general, restricting optimal solution techniques to small instances. We have developed four heuristic approaches to solve the PCB-JB problem: a cluster analysis approach (clustering), a best-fit-decreasing bin-packing approach (BFDJB), a sequencing genetic algorithm approach (GASPP), and a grouping genetic algorithm approach (GGA). We randomly generated 80 problems and performed an experimental design to characterize the performance of these heuristics. Results show that there is not a best heuristic for all circumstances. Clustering obtains the best average solution quality and fastest execution time. For a small number of jobs in the set to be partitioned, the grouping genetic algorithm finds the best solutions often finding the optimal solution. For problems with a large number of jobs, clustering is preferred for problems with a small job size variance and the BFDJB heuristic is preferred for problems with a large job size variance. The execution time for the BFDJB heuristic is close to the clustering algorithm. The two genetic algorithms are slower. GGA requires over 30 hours for a problem that takes less than 18 seconds for the clustering heuristic. Advisors/Committee Members: Magazine, Michael J.

▼ Companies rely heavily on new products as a source of profit. New products are usually introduced in several configurations to comprise a product line. A product line enables a firm to satisfy the heterogeneous preferences of today’s customers. Important in practice, the product line design problem has also drawn tremendous interest from academia. Since the 1970s researchers have been working on the problem of constructing an optimal product line using partworth data obtained from conjoint studies. In this dissertation, we focus on solving the share-of-choice product line problem. The objective is to determine the product attributes of a set of products for a firm to offer so as to maximize the projected market share provided by the set of products in the line; i.e., the number of respondents for whom at least one new product’s utility exceeds a specific hurdle. Previous contributions to this NP-Hard problem include a series of heuristics. We present a new branch-and-price algorithm that embeds a column generation procedure within a branch-and-bound process to obtain exact optimal integer solutions to the product line problem. The dissertation provides a great deal of detail about the issues encountered in implementing the branch-and-price algorithm (e.g., pricing scheme, column management and tree traversal strategy). Computational results using real and large simulated datasets demonstrate that the algorithm is capable of identifying provably optimal solutions very quickly. We design an experiment to isolate the factors to which the solution times of the branch-and-price algorithm are most sensitive. The effectiveness of previously published heuristics has only been evaluated on small problems for which complete enumeration is possible. We benchmark the performance of the branch-and-price algorithm against two heuristics (i.e., the segment-by-segment approach and the sequential approach) in the context of preference heterogeneity using test problems of realistic size. All published heuristics treat estimated partworths as errorless instead of statistical estimates, leaving the robustness of heuristics to partworth uncertainty unknown. We analyze the robustness of the branch-and-price algorithm and the sequential method to within-person variation in estimated partworths by conducting an experiment with a test phase and a validation phase. The increase of unique of product lines indicates that the sequential method is less robust to the partworth uncertainty than the branch-and-price algorithm. Advisors/Committee Members: Camm, Dr. Jeffrey D.

▼ Single-index models, in the simplest form E(y|x)=g(xTb), generalize linear models by allowing flexible curvatures captured by the unknown function g(.), and at the same time, retain the same easy interpretability as in linear models, given the index parameter vector b that form the linear index xTb. In addition, compared with fully nonparametric models, single-index models avoid the “curse of dimensionality”. This dissertation consists of two essays on single-index models. The first essay is concerned with estimation of single-index varying coefficient models. Varying coefficient models assume that the regression coefficients vary with some threshold variables. Previous research focused on the case of a single threshold variable. It is common for the coefficients to depend on multiple threshold variables but the resulting model is difficult to estimate. Single-index coefficient models alleviate the difficulty by modeling each coefficient by a function of an index. Existing estimation approaches employ kernel smoothing or local linear approximation of the coefficient functions (Xia and Li, 1999; Cai, Fan and Yao, 2003) which entail heavy computational burden. Also, implementation of different bandwidths for different coefficient functions to allow different smoothness is difficult for local approaches. We propose a penalized spline approach to estimating single-index coefficient models that not only allows different smoothness for different coefficient functions but also is computationally fast. Asymptotic theory is established under dependency. Numerical studies demonstrate the proposed approach. The second essay is on single-index quantile regression. Nonparametric quantile regression with multivariate covariates is often a difficult estimation problem due to the “curse of dimensionality”. Single-index quantile regression, where the conditional quantile is modeled by a nonparametric link function of a linear combination of covariates, can reduce the dimensionality of the estimation problem while retaining both the flexibility of a nonparametric model and easy interpretability of a simple linear model. We extend the local linear approach of Yu and Jones (1998) to estimation of single-index quantile models and introduce an iterative algorithm. Large sample properties of estimators for both the nonparametric part and the parametric part are studied. Simulation results together with real data applications show promise of the new approach. Advisors/Committee Members: Yu, Yan.