Empirically Grounding Analytics (EGA) Research: Approaches, Contributions, and Examples

Abstract

Empirically Grounding Analytics (EGA) in operations and supply chain management is a research area at the intersection of empirical and analytical studies. Spearman and Hopp (2021) identified it as an underserved research area with great opportunity for input. To clarify what EGA is, we use a quote provided in the JOM editorial on the subject (see de Treville et al. (2023)) as a definition: “an EGA paper combines mathematical, stochastic, and/or economic modeling with empirical data…. Empirically grounding an analytic model creates knowledge by linking analytical insights to what has been observed using empirical methods (such as case studies, action research, field experiments, interviews, or analysis of secondary data) to establish a theoretically and empirically relevant question.”

De Treville et al. (2023) propose a framework for discussing EGA research approaches and assessing contributions, summarized in Figure 1 of their editorial. We will refer to this framework rather extensively in our discussion of work in this Special Issue. We provide a “deconstructed” version of this figure, with some added details, in Figure 1.

Research in “empirical grounding” of analytical models can be conceptually viewed as offering two different ways to drive research and lead to impactful contributions. The “left side” approach has as its end goal to establish analytical models verifiably linked to data and observations reflecting the real operational setting. This approach contributes a “calibrated fit” of the model to the operational decision reality. It requires careful empirical justification of modeling assumptions and parameters. The calibration of model parameters involves collecting representative data from the realistic setting, with any remaining model assumptions and approximations well justified for the real situation. The expectation of these grounded models is a high quality of solutions for the approximated real decision problem.

The “right side” approach pursues empirical assessment of model results, solution quality, and applicability of insights in addressing issues encountered in real practice. It carefully verifies that (a) an effective implementation of the model reasonably and accurately depicts the operational setting and decision situation; (b) the obtained solutions lead to improved performance; and (c) incorporating analytical insights and tools leads to improved managerial practice for this setting.

In most cases, “left side” research leads to well-calibrated models with strong hints for improved solution quality and useful insights to be further tested in the real setting and actual practice. “Right side” research carefully tests and confirms the wisdom of new insights and tools, leading to improved practice in the operational setting. However, such testing and analytical insights may reveal irregularities and complexities not effectively depicted in the models, thus driving the need for new “left side” research and subsequent “right side” research for further validation and testing. After a few such healthy iterations, we expect this EGA research paradigm to lead to the development of well-grounded knowledge and well-tested, impactful theories enhancing operations and supply chain management practice. This systematic iteration, “left side ⇌ right side,” is what we refer to as the “close-the-loop” approach.

In this editorial thought piece, we demonstrate how the above approaches (“left side”, “right side”, and “close-the-loop”) have been used in classic operations literature (e.g., “bullwhip” research), in recent papers (Turcic et al. (2023), Wendt et al. (2025)), and in papers accepted by the Special Issue.

The “bullwhip” research literature is rich in both analytical and empirical studies. In one classic paper, observations of a real phenomenon in supply chains led to insightful modeling work that identifies the “bullwhip” effect's main causes and helps quantify it (Lee et al. (1997)). Efforts to find clear evidence of the “bullwhip” effect in industrial supply chains exposed challenges regarding measures to use and appropriate data to verify it, as described in influential papers by Cachon et al. (2007), Bray and Mendelson (2012), and Yao et al. (2021). Another iteration of modeling research helped address conflicting evidence from aggregate de-seasonal data and measurement differences, and effectively “close-the-loop” in building a well-tested, credible “bullwhip” theory in supply chains. In our opinion, this is the best high-impact EGA research example in the operations and supply chain management field so far.

Spearman and Hopp (2021) and de Treville et al. (2023) expressed the concern that it is extremely challenging to have both analytical model contributions and rigorous empirical testing of obtained insights in a single paper. These challenges notwithstanding, we will demonstrate that it is possible for one paper to “close-the-loop.” We will use the more concise term “Integrated EGA” for research in these papers. We offer an example of Integrated EGA work by discussing in detail the work of Turcic et al. (2023). The authors accurately depict unique complexities in commodity procurement within industrial chains through an innovative and parsimonious model of procurement contracts for these settings. Then, they proceed to test their analytical insights using rich data sets provided by a major automaker extensively involved in commodity procurement. The “close-the-loop” approach ends up with a well-tested procurement contract theory for these industrial chains.

As we mentioned before, most published EGA research (see de Treville et al. (2023)) has either “left side” or “right side” contributions. While it is described as hard to accomplish (see Spearman and Hopp (2021) and de Treville et al. (2023)), we have seen recent efforts of “close-the-loop” research in a single paper (both “left side” and “right side” contributions in the same paper), which we refer to as “Integrated EGA.”

In an effort to present a complete theory on automotive (and similar other industry) procurement contracts, Turcic et al. (2023) accomplish the difficult feat of presenting in the same paper a parsimonious model of the procurement contracting processes, which depicts realities ignored in previous work, and an empirical validation of the generated hypotheses from the model. We will use it as our main example of “Integrated EGA” research, described in Section 3.1. We present a second example of integrated research on capacity trading strategies (Wendt et al. (2025)) in Section 3.2.

The motivation for soliciting research for the Special Issue on “Empirically Grounding Analytics (EGA) in Operations & Supply Chain Management” was to bridge the widely observed gap between modeling research of operational environments and the rigorous empirical work in verifying models and theories using data from real operational settings. We viewed the widely cited EGA framework in de Treville et al. (2023) as an opportunity to bring new EGA research to prominence, offer incentives to contributing authors to apply it in new settings, and then classify new contributions along the dimensions of the framework. Through the Special Issue, we hope to further motivate sound EGA research in the operations and supply chain management field. Furthermore, we hope to move beyond “one sided” (either “right-side” or “left side”) EGA contributions, which represent the majority of published research, and highlight the opportunity and strength of contributions via an “Integrated EGA” research paradigm. We consider achieving progress toward the latter point a major accomplishment of our Special Issue.

We received 22 submissions as a result of the call for the Special Issue on EGA. Unfortunately, many of the submissions, while representing worthwhile research papers by solid credential and expertise authors, misinterpreted the term “empirically grounding analytics.” Some of the papers were strong on the modeling/analytics side but lacked the formulation of the research question, appropriate data, and focused effort to use methodologies for empirical research in their studied topics. Other contributions, while appropriate in fitting the empirical grounding analytics scope, they were lacking the focus on providing insights for enhanced decision-making in an operations management context. While we do understand that some of this may sound subjective, the generalizability, novelty and quality of the insights gained are the main criteria for the inclusion of the selected papers in this Special Issue. We ended up selecting three papers for the Special Issue.

The first of the papers is “An Empirically Grounded Analytics (EGA) Approach to Hog Farm Finishing Stage Management: Deep Reinforcement Learning as Decision Support and Managerial Learning Tool” by Kouvelis et al. (2025). The study considers the weekly farmer planning decisions on how to dispose finished hogs to long-term contract markets with downstream meat packers and short-term transactional open market, with the goal to maximize the farm's value over a given operating horizon. The authors use deep reinforcement learning (DRL) (an actor-critic concept as in Sutton and Barto (2018)) for the Markov decision process (MDP) model of the problem. With respect to EGA-appropriate “left side” modeling contributions, they use proprietary data from one of the top hog farmers in the U.S. and price data from the Chicago Mercantile Exchange and U.S. Department of Agriculture (USDA) to estimate price and inventory distributions. They use these distributions to generate extensive synthetic training data and apply the trained DRL agent to demonstrate near-optimal decision making. The provided solution over performs existing practice (in the 20–25% improved performance range). Furthermore, the study uses classification trees (Bertsimas and Dunn 2017) to derive “managerial learning” from the DRL solution by easing-out key themes in the DRL agent's decision-making process. The paper “closes-the-loop” on improved decision-making in their environment and is an example of “Integrated EGA” research.

The article “Encounter Decisions for Patients with Diverse Sociodemographic Characteristics: Predictive Analytics of EMR Data from a Large Chain of Clinics” by Mukherjee et al. (2025) presents a novel EGA application: toggling between outcomes and the analytic model (“close-the-loop” approach). The authors propose a decision framework for the optimal allocation of limited primary care encounter capacity based on predictive and analytical modeling. In this study, socioeconomic and demographic information enhanced individuals' diabetes risk prediction significantly; the authors also show that more encounters can substantially improve diabetes outcomes and reduce health risks. Furthermore, for a test set, the authors compare the predicted encounter frequency with the observed data. Implementation heuristics for the practical implementation are proposed. The backdrop of this study is the healthcare sector, where there is a persistent shortage of workers and budgets are tight. This situation is frequently encountered in many operations and service settings as well.

Using a contingent claims framework, the article “Survive the Economic Downturn: Operating Flexibility, Productivity, and Stock Crash” by Li et al. (2025) empirically investigates how operating flexibility may be a “more fundamental factor” in curtailing the risk of stock crashes than suggested by the current literature. The paper draws heavily from the existing literature as a foundation upon which to build and empirically validate three hypotheses. The authors test these hypotheses by defining various variables (factors or measures) deemed relevant for this study. Using data from 1961 to 2020, the authors conduct cross-sectional tests and show a significant negative association between operating flexibility and crash risk and the mechanism underpinning this association. The findings of this paper have practical implications for corporate managers and investors and deviate from the conventional notion of bad news hoarding or bad news withholding. From an EGA perspective, this paper falls into the category of the “right side” approach, where the theoretical aspect is the starting point. According to the authors, their main contribution is to provide equally strong theoretical and empirical evidence.

There are many opportunities for EGA research with “left side,” “right side,” “complementary in close-the-loop,” and “Integrated EGA” contributions. Our overall assessment, and one could accuse us of some personal bias in it, is that the more productive and impactful opportunities are in “right side” and “Integrated EGA” contributions. We will elaborate this point below.

“Left side” EGA research opportunities continue to exist in the operations management field, as many of our model-based theories have been empirically tested and calibrated only to a limited extent. In top journals, investigation using these theories often receives the underappreciated label of “applied research,” and is performed only for certain industries of immediate applicability or having easy access to data for the researchers at that time. In almost every case, researchers were limited by the data limitations of the application setting at the time of the project (with many of those projects going back 10–20 years, with more infrequent periodic and discrete data and lots of data gaps that drove the need of synthetic data with hypothesized distributions). In the absence of incentives to publish calibrated models in these previously examined application areas, despite the increased availability of richer, real-time and complete data sets, we see that very limited EGA research has been published in that space. For example, applied queueing models research offers reasonable model calibration for applications in production systems, healthcare settings, and telecommunications and call centers, but there is not much reported effort to calibrate the use of these models in product development funnels. Even in the applications where models were tested, there is no effort to form a rigorous EGA approach in collecting, processing, and calibrating data so that meaningful model insights can be generated to “meta-theorize” on the effective use of these models in their application environments.

“Right side” and “close-the-loop” research opportunities for EGA research are the most promising ones. The limited extensive empirical testing of many of the analytical insights of operations management theories and their quantifiable impact on improved practice across many industries creates opportunities to structure research agendas for doing so. In many cases the anecdotal evidence, often summarized in company case studies, suggests that these insights and practices improved operational effectiveness (e.g., focused factory principles, risk pooling inventory management, postponement practices in product line design, multi-echelon inventory methods, etc.). However, a rigorous cross-industry study to determine the right level of product aggregation—something similar in extent and magnitude to the research for the “bullwhip” phenomenon—has not been performed. In addition, as we had an opportunity to see in our coverage of “closing-the-loop” in “bullwhip” research, new questions will arise on unanticipated inconsistencies between what our analytical insights and model-based theories predict and what industry data and actual practices reveal. This might eventually lead to going back to the drawing board for new assumptions and models to explain and further test them.

As Spearman and Hopp (2021) and de Treville et al. (2023) implied in their coverage of our field's research, our field has not yet reached the maturity of empirical testing that other fields (e.g., economics, behavioral sciences, etc.) have achieved. Quite a lot of this “close-the-loop” EGA research will be done in a complementary fashion, and in most cases will test the magnitude of effect and improvements in practices of analytical insights driven by assumptions not tested in an evolving technology and applications environment. At the same time, in this environment, research teams that are trained on both empirical and modeling skills, or simply appropriately sized and composed, have the unique opportunity to pursue “Integrated EGA” in ambitious research projects. Our Special Issue has highlighted the potential of such research, and we believe the opportunities for more of it are there.

We would like to close our future EGA research opportunities discussion by emphasizing how emerging data processing and analytics technologies, in particular machine learning and other advances in artificial intelligence (especially deep reinforcement learning and generative AI, with caveats on permissible use, www.jom-hub.com/submissions/ai-policies), can reshape the EGA research agendas and capabilities. To avoid unnecessary overlap, we refer to recently published work for deeper understanding on how supervised machine learning will affect theory building and testing for the operations management field (Chou et al. (2023)). In our research experience, and through submitted work to the Special Issue, we discovered that among supervised learning techniques, classification trees (a type of decision tree used to predict categorical outcomes by recursively partitioning data based on input features and aiming to create “pure” nodes representing distinct classes) can be extremely valuable in understanding the performance of practices in operational environments—for example, grouping practices according to when and under what conditions they work effectively. In other instances, classification trees can increase the forecast accuracy of operational variables with unstructured data sets (e.g., forecast lead times for engineered-to-order products ordered from global suppliers based on unstructured data on engineering and manufacturing). At the same time, we saw researchers using classification trees, as reported in one of the Special Issue papers, as a managerial learning tool to understand patterns of effective decisions pursued by opaque, but highly effective computational decision models (e.g., dynamic decision instances of deep reinforcement learning neural network algorithms).

Deep reinforcement learning (DRL) is a machine learning (ML) methodology that uses algorithmic approaches to make decisions to optimize a set of goals. At first glance, it is hard to see how DRL will contribute in an EGA research framework. However, we highlighted the Kouvelis et al. (2025) Special Issue paper as an example of ways DRL in applied decision-making settings can lead to insightful EGA research contributions, and we are hoping to see more work along these dimensions. We expect that such contributions will go beyond a pure “left side” contribution in fitting and demonstrating the test value of the DRL model for the decision-making environment with the use of empirical data. We would like to see machine learning (ML) techniques, as also discussed in Chou et al. (2023), in combination with DRL to drive managerial learning to improve practice within an operating environment. The congruence of three main factors—richness of available application data, due to recent technologies (e.g., sensors, blockchain, vast storage and cloud services, computing power and real-time control, etc.); the sophisticated use of algorithmic approaches to dynamic decision making, with both numerical and unstructured data (e.g., neural networks, generative AI, etc.), provided they adhere to clear and shared standards regarding AI-use in research (cf. AI-policy at www.jom-hub.com/submissions/ai-policies); and the advancements in data sciences in understanding patterns (e.g., classification trees, regression trees, etc.)—creates the opportunity for highly impactful “Integrated EGA” research in operations management.