جایگزین مدل سازی علی در تحقیق در عملیات : بررسی اجمالی و کاربرد

This paper uses the relationships between three basic, fundamental and proven concepts in manufacturing (resource commitment to improvement programs, flexibility to changes in operations, and customer delivery performance) as the empirical context for reviewing and comparing two casual modeling approaches (structural equation modeling and Bayesian networks). Specifically, investments in total quality management (TQM), process analysis, and employee participation programs are considered as resource commitments. The paper begins with the central issue of the requirements for a model of associations to be considered causal. This philosophical issue is addressed in reference to probabilistic causation theory. Then, each method is reviewed in the context of a unified causal modeling framework consistent with probabilistic causation theory and applied to a common dataset. The comparisons include concept representation, distribution and functional assumptions, sample size and model complexity considerations, measurement issues, specification search, model adequacy, theory testing and inference capabilities. The paper concludes with a summary of relative advantages and disadvantages of the methods and highlights the findings relevant to the literature on TQM and on-time deliveries.

Interest in causal modeling methodologies in the social sciences stems from the desire to establish patterns of regularities or laws analogous to those in the physical sciences. A fundamental appeal of causal modeling is the ability to combine cause–effect information, based on theoretical construction, with statistical data to provide a quantitative assessment of relationships among the studied variables. The purposes for employing causal modeling in the study of operations are to develop an explanation of relationships and to provide a basis for inference. The portrayal, evaluation and summarization of assumed causal relationships are the components of explanation. These relationships are then used to develop inferences for diagnostic reasoning from effects to causes and for the prediction of outcomes that would follow from a policy or procedure intervention. Available modeling methods offer differing functional advantages and limitations. However, any method should have potential managerial usefulness by providing outputs with clear interpretation and the capability to assess the impact of potential changes in the modeled process. Ideally, a causal study would take the form of a randomized controlled experiment conducted over an appropriate time period. Such a research design would minimize construct, internal, external, and statistical threats to validity (Cook and Campbell, 1979), and allow the possibility of causal conclusions to be reached. Unfortunately, randomized controlled experiments can seldom, if ever, be utilized to provide causal knowledge for strategy and policy issues. Thus, causal modeling methods for non-experimental data are of interest. Bayesian networks and structural equation models (SEM) are the causal modeling methods for non-experimental data reviewed and compared in this paper. The paper begins with the central issue of the requirements for a model of associations to be considered causal. This philosophical issue is addressed in reference to probabilistic causation theory. Then, each method is reviewed in the context of a unified causal modeling framework consistent with probabilistic causation theory, and applied to a common dataset. The comparisons include concept representation, distribution and functional assumptions, sample size and model complexity, measurement, specification search, model adequacy, theory testing and inference capabilities. The paper concludes with a summary of the relative advantages and disadvantages of the methods.

The focus of this paper is on methods that view causation from a graphical viewpoint. The notion of a DAG is utilized to represent the conditional independence between any two variables, which implies the absence of a direct causal relationship. Further, assuming the nodes of a DAG represent random variables, the joint probability distribution of these variables can be factored in a causal meaningful manner. These general results are summarized in the specification of a causal model as a structure S conforming to a DAG and a set of associated parameters View the MathML source consistent with the structure. SEM and Bayesian networks are portrayed as specializations of the general causal model specification, View the MathML source. The emphasis of causal modeling applied in operations management has been primarily concentrated on providing increased process understanding by emphasizing psychometric and statistical support for theory-based explanations. SEM have been the most frequently used method for quantifying and evaluating an assumed causal process. The primary objective, under this approach, is to assess whether a postulated theoretical network is a reasonable approximation of the process that generated the study data. Indexes supporting construct validity, measurement reliability, parameter significance, model fit, and causal effects tend to dominate the reported results. Since the analysis is usually based on a high level of theoretical and knowledge domain support, the confirmatory findings typically support the large majority of hypothesized relationships. Thus, conclusions tend to provide value by incremental extension of existing conceptualizations that cannot be extended to prediction of observed outcomes. Bayesian networks, in contrast to SEM, assume the main role of causal modeling is to facilitate the analysis of potential and actual actions, rather than focus on theory confirmation. Indeed, Bayesian networks offer the capabilities to explain system relationships and to predict the impacts of potential actions as alternative structures that can be evaluated by traditional tests of significance or by posterior probabilities or both, as demonstrated in the above application. The probability metric provides non-linear detailed relationship information that should be easily consumable by the managers as well as academics. The modeling effort is concerned with observable variables, not hypothetical concepts. Thus, it is possible to introduce a conceptual intervention and evaluate the expected observable changes. More specifically, the posterior probabilities resulting from the intervention can be compared to the pre-intervention probabilities to provide a quantitative measure of expected change.