15. Doctoral Research and Emerging Frontiers

Introduction and context

Operations and supply chain innovation has become an especially fertile domain for doctoral research because operational systems are simultaneously technical, organizational, social, and ecological. A warehouse algorithm may alter labour practices; a sourcing decision may redistribute environmental risk; and a digital platform may improve visibility while creating new forms of dependence and surveillance. Consequently, doctoral inquiry cannot be reduced to identifying a more efficient process. It must explain how innovations are designed, adopted, governed, measured, and experienced across interconnected systems.

This chapter focuses on two related tasks. The first is the formulation of rigorous original research. The second is the examination of emerging frontiers in which established theories and methods may be insufficient: autonomous supply chains, human–AI collaboration, climate adaptation, and regenerative operations. The central argument is that methodological sophistication is valuable only when it is aligned with a meaningful research problem, a defensible causal logic, and ethically responsible evidence.

Learning objectives

By the end of the chapter, doctoral students should be able to:

1. Formulate significant and researchable operations and supply chain questions.

2. Select and justify design science, case study, survey, experimental, econometric, optimization, or mixed-method approaches.

3. Distinguish description, prediction, explanation, and design objectives.

4. Critically assess causality, construct validity, internal validity, external validity, reliability, and robustness.

5. Design research on autonomous supply chains, human–AI collaboration, climate adaptation, and regenerative operations.

6. Address ethical, governance, privacy, labour, and sustainability implications of operational innovation.

7. Translate research findings into credible theoretical and managerial contributions.

From managerial puzzles to doctoral problems

A doctoral research problem is not simply a topic. “Artificial intelligence in procurement” is a topic; “how explainable AI affects buyer reliance on supplier-risk recommendations under high uncertainty” is a research problem. A strong problem identifies a phenomenon, a tension or gap in existing knowledge, and a reason the answer matters.

Research questions generally pursue one or more purposes. Descriptive questions establish what is occurring, such as the extent to which firms use digital twins. Explanatory questions ask why or how an outcome occurs. Predictive questions estimate future outcomes, while design-oriented questions ask what artifact, intervention, or configuration could improve a situation. These purposes should not be confused. A model that predicts delivery failure may have little explanatory value, and a case that richly explains a transformation may not support statistical generalization.

A useful doctoral proposal connects four elements: a phenomenon, a theoretical mechanism, an empirical setting, and a contribution. For example, a study might argue that algorithmic transparency increases adoption of automated replenishment because it improves perceived controllability, but that this effect weakens when planners fear loss of professional autonomy. The theory identifies the mechanism; the setting provides a meaningful boundary; and the research design determines whether the claim can be tested.

Important concepts and terminology

An ontological position concerns what the researcher assumes to be real, such as objective performance measures, socially constructed meanings, or both. An epistemological position concerns how knowledge can be justified. Positivist studies emphasize measurement and testing; interpretive studies examine meaning and context; pragmatist studies select methods according to the research question and practical consequences.

A construct is an abstract concept such as resilience, trust, or absorptive capacity. An operationalization specifies how that construct will be observed or measured. A mediator explains the process through which a cause affects an outcome, whereas a moderator changes the strength or direction of that relationship. A boundary condition identifies where a theory is expected to apply.

Causal identification is the credible determination of whether a change in an independent variable produced a change in an outcome. Internal validity concerns the credibility of causal claims within the study; external validity concerns transferability or generalizability beyond it. Construct validity concerns whether measures represent the intended concepts. Reliability concerns consistency, although a reliable measure can still be invalid.

Choosing a research architecture

Design science research

Design science is appropriate when the research objective includes creating and evaluating an artifact intended to solve a recognized problem. The artifact may be an algorithm, decision-support system, process model, governance mechanism, or design principle. A rigorous design science project normally includes problem relevance, explicit design objectives, construction, demonstration, evaluation, and communication.

Its scholarly contribution is not merely a functioning prototype. The researcher must explain why design features should produce desired outcomes and what generalizable principles can be learned. Evaluation may assess technical performance, usability, robustness, adoption, fairness, or effects in a realistic setting. A common mistake is to treat implementation success as proof of theoretical validity. An artifact can work in one setting because of local expertise, data quality, or exceptional managerial support.

Case study research

Case studies are valuable for investigating contemporary phenomena in context, particularly when the boundaries between technology, organization, and environment are unclear. A single case may be justified by its critical, revelatory, or extreme character; multiple cases support replication logic and systematic comparison. Case evidence can include interviews, observations, archival records, system logs, and performance data.

Rigour requires a clear case-selection rationale, a protocol, a chain of evidence, rival explanations, and transparent coding. Process research benefits from temporal bracketing: the researcher may examine preparation, implementation, stabilization, and adaptation rather than treating transformation as a single event. Case studies provide analytical rather than statistical generalization. Their purpose is to refine, extend, or challenge theory, not to claim that one organization represents an entire population.

Surveys and experiments

Surveys are useful for estimating relationships across firms, employees, suppliers, or customers. Doctoral quality depends on construct validity, sampling logic, response-bias assessment, scale development, and appropriate model specification. Common-method bias is a concern when the same respondent, instrument, and time point provide both predictor and outcome measures. Remedies include temporal separation, multiple respondents, objective outcomes, careful item design, and statistical sensitivity analysis.

Experiments offer stronger causal identification by manipulating a treatment and comparing outcomes with an appropriate control condition. Laboratory, field, vignette, and online experiments each trade realism against control. In operations research, a vignette might manipulate levels of AI explainability and measure reliance; a field experiment might randomize inventory recommendations across stores. Randomization does not automatically eliminate all threats: noncompliance, attrition, spillovers, weak manipulation, and treatment heterogeneity must be examined.

Econometrics and observational evidence

Operational data are often observational. Firms do not randomly choose suppliers, technologies, or resilience investments. A correlation between digital visibility and performance may reflect reverse causality, omitted managerial capability, or selection into adoption. Econometric designs seek a credible counterfactual—the outcome that would have occurred for the treated unit in the absence of treatment.

Useful approaches include fixed-effects models, difference-in-differences, instrumental variables, regression discontinuity, matching, event studies, and panel models. Each depends on assumptions. Difference-in-differences requires credible parallel trends; instrumental variables require relevance and exclusion; regression discontinuity requires no precise manipulation around the threshold. These assumptions should be argued substantively, assessed empirically where possible, and reported transparently.

Predictive accuracy is not equivalent to causal understanding. A high-performing machine-learning model may exploit stable associations that fail after a policy change. Doctoral research should distinguish prediction from explanation and avoid causal language unsupported by design.

Optimization and simulation

Optimization is appropriate when the research problem concerns the best feasible decision under specified objectives and constraints. Models may address network design, inventory, routing, production scheduling, or capacity allocation. A contribution can arise from a new formulation, an algorithm, a structural result, or an insight into trade-offs.

Model credibility depends on assumptions, parameter sources, sensitivity analysis, computational transparency, and comparison with relevant benchmarks. Robust and stochastic optimization are particularly relevant when demand, lead times, climate hazards, or technology performance are uncertain. Simulation can explore dynamic systems and policies that are difficult to test directly, but its conclusions remain conditional on model structure and input distributions.

Mixed methods

Mixed methods are not simply the addition of interviews to a quantitative study. They require integration. An exploratory sequential design may use cases or interviews to develop constructs and hypotheses before a survey. An explanatory sequential design may use qualitative inquiry to explain surprising statistical results. A convergent design may collect qualitative and quantitative evidence in parallel and compare findings.

The integration point should be explicit: during sampling, instrument development, analysis, or interpretation. Contradictory findings are not necessarily defects; they may reveal boundary conditions, measurement problems, or differences between formal processes and lived practice.

A disciplined logic of evidence

Doctoral research should make its claims proportionate to its evidence. A useful practice is to construct a claim–evidence matrix linking each proposition to measures, observations, identification assumptions, and possible rival explanations. Researchers should pre-specify hypotheses or analysis plans where feasible, distinguish exploratory from confirmatory analyses, and report null or unexpected results.

Validity is multidimensional. Internal validity may be threatened by omitted variables, maturation, history, selection, and contamination. External validity may be limited by unusual industries, crisis conditions, or highly capable focal firms. Measurement validity may be weakened by vague constructs or scales transferred across cultures without adaptation. Reproducibility requires documented data preparation, code, model specifications, and decision rules, subject to confidentiality and privacy constraints.

Triangulation is strongest when different sources address the same theoretical claim in complementary ways. For example, interviews may reveal why planners distrust a system, system logs may show override behaviour, and performance data may indicate whether overrides improve outcomes. Triangulation does not mean that all sources must agree; disagreement can itself be theoretically informative.

Emerging frontier I: autonomous supply chains

An autonomous supply chain uses sensing, analytics, algorithms, and automated execution to make or implement decisions with limited human intervention. Autonomy is better understood as a continuum than a binary condition. At one end, systems recommend actions; at the other, they negotiate, order, schedule, reroute, and learn with minimal approval.

Research questions concern the appropriate allocation of decision rights, the conditions under which autonomy improves resilience, and how organizations govern exceptions. Autonomy can reduce response time and routine workload, but it may amplify errors, create correlated decisions across firms, and produce brittle behaviour when conditions depart from training data. A theoretically useful model may distinguish automation of execution from autonomy of goal-setting. The latter raises questions of accountability and value conflict that technical performance metrics cannot resolve.

Research designs should examine rare events, escalation protocols, human override, and system recovery rather than average performance alone. Scenario experiments, longitudinal case studies, agent-based simulation, and field data on interventions can be combined to study these issues.

Emerging frontier II: human–AI collaboration

Human–AI collaboration is not simply a question of whether an algorithm is more accurate than a person. Operational outcomes depend on complementarity, reliance, contestability, learning, and division of labour. An AI may detect patterns that planners miss, while planners contribute tacit knowledge, contextual interpretation, and ethical judgment.

Important variables include explainability, confidence calibration, task structure, expertise, workload, time pressure, and perceived autonomy. Over-reliance occurs when users accept incorrect recommendations; under-reliance occurs when useful recommendations are rejected. Research should therefore measure reliance quality rather than adoption alone. A randomized experiment can vary recommendation accuracy and explanation type, while interviews can examine why participants accept or reject advice.

Governance should preserve meaningful human responsibility. “Human in the loop” is insufficient if the human lacks time, authority, competence, or access to relevant information. Effective collaboration requires escalation rights, audit trails, training, model monitoring, and procedures for challenging automated outputs.

Emerging frontier III: climate adaptation and regenerative operations

Climate adaptation concerns adjustments that reduce vulnerability to hazards such as floods, heat, drought, wildfire, and infrastructure disruption. It differs from mitigation, which addresses emissions, although the two are connected. Supply chain research can examine adaptive capacity, supplier diversification, inventory positioning, flexible contracts, facility redesign, early-warning systems, and financial instruments.

A key doctoral challenge is to avoid treating resilience as an unlimited good. Redundancy can improve continuity while increasing cost, emissions, or extraction. Adaptation investments may also shift risk to suppliers, workers, or communities. Distributional effects should therefore be included in the analysis.

Regenerative operations go beyond reducing harm toward restoring ecological and social systems. Examples include restorative agriculture, circular material flows, watershed regeneration, repair and reuse ecosystems, and procurement that strengthens local capabilities. The concept requires careful operationalization. A firm should not claim regeneration merely because it recycles a fraction of its waste. Research must specify the ecological or social baseline, the time horizon, the system boundary, and evidence of net improvement.

Multi-objective optimization, lifecycle assessment, participatory case research, and impact evaluation can help investigate these questions. Researchers should be alert to rebound effects, burden shifting, greenwashing, and the difference between relative efficiency and absolute environmental improvement.

Illustrative case: Meridian Foods’ autonomous replenishment project

Background

The following is a fictional illustrative case. Meridian Foods is a regional packaged-food distributor operating several distribution centres and serving supermarkets, hospitals, and independent retailers. It plans to deploy an AI replenishment system that forecasts demand, recommends orders, and eventually submits purchase orders automatically.

Situation and problem

Pilot tests show a reduction in stockouts and planner workload, but planners frequently override recommendations for products affected by promotions, local events, and supplier unreliability. Senior management proposes full automation. The procurement director is concerned that the model favours historically reliable large suppliers and may disadvantage smaller regional producers. A sustainability manager also notes that the system optimizes availability and cost but does not account for food waste, transport emissions, or supplier vulnerability to heat and flooding.

The doctoral research problem is not simply whether the system improves forecast accuracy. It is how decision autonomy, explanation, human expertise, and multi-dimensional objectives influence operational performance and distributional consequences.

Research approach

A mixed-method design is appropriate. First, a longitudinal case study can document implementation, planner adaptation, supplier responses, and governance changes. Second, a quasi-experimental comparison can examine stores adopting the system at different times, using pre-intervention trends and controls to estimate effects on stockouts, waste, and emergency shipments. Third, a factorial experiment with planners can vary explanation quality, recommendation confidence, and time pressure to test reliance behaviour. Finally, a multi-objective optimization model can compare policies that weight service, cost, emissions, waste, and supplier concentration differently.

The study should predefine outcomes and examine heterogeneous effects. For instance, automation may improve stable products but perform poorly for promotional or climate-sensitive products. Qualitative evidence can explain these differences and identify when human intervention adds value.

Governance and solution

Meridian adopts graduated autonomy. Stable, low-risk items can be ordered automatically within limits; volatile, perishable, or socially sensitive categories require planner approval. The system records recommendations, explanations, overrides, and subsequent outcomes. A cross-functional committee reviews fairness, supplier concentration, waste, and climate exposure each quarter. Suppliers receive a communication channel through which data errors and exceptional circumstances can be challenged.

Likely outcome and learning points

The central outcome is not maximum automation but better allocation of human attention. The case illustrates that operational performance, legitimacy, and sustainability may require different objective functions and different levels of autonomy. It also demonstrates why a single method would be inadequate: quantitative designs estimate effects, experiments examine mechanisms, optimization explores trade-offs, and qualitative inquiry explains organizational adaptation.

Ethics, governance, and professional responsibility

Operations researchers work with sensitive commercial, employee, supplier, and location data. Consent, confidentiality, data minimization, secure storage, and lawful processing are basic requirements, not administrative details. Researchers should consider whether participation is genuinely voluntary when employees are recruited through managers, and whether publication could expose vulnerable suppliers or communities.

Algorithmic research raises additional concerns. Models may reproduce historical discrimination, encode unequal bargaining power, or make decisions that affected parties cannot contest. Fairness cannot be assessed without defining the relevant population, outcome, and notion of fairness. Privacy-preserving techniques do not eliminate all risks of re-identification or surveillance. Researchers must also disclose conflicts of interest, funding influence, commercial restrictions, and model limitations.

Ethical review should include foreseeable consequences of deployment. A technically successful optimization model may intensify work, eliminate discretion, or transfer environmental burdens. Responsible research therefore asks who benefits, who bears risk, who has voice, and who can appeal.

Common failure modes and best practices

Frequent doctoral errors include selecting a fashionable technology before identifying a problem, treating access to a firm as a research design, equating correlation with causation, using unvalidated scales, reporting only successful experiments, and claiming generalization from a single case. Another error is measuring resilience as a single index without specifying the disruption, time horizon, or recovery criterion.

Better practice involves:

• beginning with a theoretically consequential problem rather than a tool;
• stating the unit of analysis, temporal scope, mechanism, and boundary conditions;
• aligning claims, data, and identification strategy;
• testing alternative explanations and conducting sensitivity analyses;
• combining performance measures with behavioural and distributional evidence;
• documenting analytical decisions and separating exploratory from confirmatory work;
• involving practitioners and affected stakeholders without allowing access partners to dictate conclusions; and
• designing governance and evaluation before operational deployment.

Current directions for doctoral inquiry

Several frontiers deserve sustained investigation. Digital twins may connect real-time data with simulation, but questions remain about model drift, interoperability, and organizational trust. Federated analytics may enable collaboration without centralizing data, while introducing new problems of incentives and comparability. Generative AI may support planning and knowledge work, but its reliability, provenance, and accountability in high-consequence decisions require careful study. Climate volatility will increasingly test whether supply chain resilience is adaptive or merely reactive. Finally, regenerative operations invite researchers to rethink the purpose of operations from efficient throughput toward long-term system vitality.

The most valuable doctoral contributions will not be those that simply attach a new technology to an old outcome. They will clarify mechanisms, expose trade-offs, improve decision artifacts, and expand the moral and ecological boundaries of operations research.

Chapter Summary

Doctoral research in operations and supply chain innovation begins with a significant problem, a defensible theoretical mechanism, and a research design suited to the intended claim. Design science develops and evaluates artifacts; case studies explain context and process; surveys and experiments test relationships and causal mechanisms; econometrics addresses observational counterfactuals; optimization and simulation analyze constrained decisions; and mixed methods integrate complementary evidence. Rigour requires attention to construct, internal, external, and ecological validity, as well as transparency and robustness.

Emerging research on autonomous supply chains must address decision rights, exceptions, and accountability. Human–AI research should examine reliance quality and complementarity rather than adoption alone. Climate adaptation and regenerative operations require explicit system boundaries, time horizons, distributional analysis, and evidence of genuine improvement. Across all frontiers, ethical governance, stakeholder voice, and proportionate claims are essential to responsible scholarship.

Key Terms

Design science research: Research that constructs and evaluates an artifact intended to solve a relevant problem and generate transferable design knowledge.

Construct validity: The extent to which an operational measure represents the theoretical concept it is intended to measure.

Internal validity: The credibility of a study’s causal conclusions within its research setting.

External validity: The extent to which findings can be transferred or generalized to other settings.

Causal identification: Establishing whether an intervention or exposure produced an observed outcome.

Mediator: A variable explaining the process through which one variable affects another.

Moderator: A variable that changes the strength or direction of a relationship.

Boundary condition: A circumstance defining where a theoretical relationship is expected to hold.

Difference-in-differences: An econometric method comparing changes over time between treated and comparison groups.

Instrumental variable: A variable used to identify causal effects when treatment is confounded, subject to relevance and exclusion assumptions.

Common-method bias: Bias arising when predictors and outcomes are measured using the same source or method.

Triangulation: Use of multiple methods, data sources, or perspectives to examine a claim.

Autonomous supply chain: A supply chain in which digital systems make or execute decisions with limited human intervention.

Human–AI collaboration: Joint performance in which people and artificial intelligence contribute complementary capabilities.

Climate adaptation: Adjustments that reduce vulnerability to climate-related hazards and variability.

Regenerative operations: Operational practices intended to restore or improve ecological and social systems.

Counterfactual: The outcome that would have occurred for a unit under an alternative treatment condition.

Graduated autonomy: Allocation of different levels of automation according to task risk, uncertainty, or reversibility.

Review Questions

Short Answer Questions

1. How does a research problem differ from a research topic?

2. Define construct validity and explain why it matters in supply chain research.

3. What is the difference between a mediator and a moderator?

4. Why is predictive accuracy not equivalent to causal explanation?

5. What is meant by graduated autonomy?

Long Answer Questions

1. Compare design science, case study, and econometric research for investigating autonomous supply chains.

2. Explain the major threats to causal validity in observational operations data and how researchers can address them.

3. Discuss how human–AI collaboration changes the meaning of operational performance.

4. Critically evaluate the relationship between climate adaptation and regenerative operations.

5. Explain how mixed methods can produce a stronger doctoral contribution than a single-method design.

Analytical / Application Questions

1. Design a research study testing whether AI explainability affects planner reliance on replenishment recommendations. Identify the treatment, outcomes, mechanism, and validity threats.

2. A firm reports lower emissions after relocating suppliers but higher supplier failure rates and community unemployment. Analyze the claim using a regenerative-operations perspective.

3. Using the Meridian Foods case, propose a governance framework for deciding which products may be ordered autonomously and justify the criteria.

22. Multiple Choice Questions with Answers

Question 1

A doctoral researcher wants to determine whether AI explanations cause planners to rely more appropriately on replenishment recommendations. Which design offers the strongest direct test of this mechanism?

A. A cross-sectional survey of planner attitudes

B. A randomized experiment varying explanation quality and measuring reliance quality

C. A single interview with the system designer

D. A descriptive analysis of annual stockout rates

Correct Answer: B. A randomized experiment varying explanation quality and measuring reliance quality

Question 2

Which statement best distinguishes a mediator from a moderator?

A. A mediator changes the sample, while a moderator changes the measurement scale

B. A mediator explains how an effect occurs, while a moderator changes its strength or direction

C. A mediator is always qualitative, while a moderator is always quantitative

D. A mediator establishes external validity, while a moderator establishes reliability

Correct Answer: B. A mediator explains how an effect occurs, while a moderator changes its strength or direction

Question 3

In a difference-in-differences study, which assumption is especially important?

A. The treatment group must be larger than the comparison group

B. All variables must be normally distributed

C. Treated and comparison units would have followed parallel trends absent treatment

D. The intervention must affect every unit equally

Correct Answer: C. Treated and comparison units would have followed parallel trends absent treatment

Question 4

Why might a highly accurate machine-learning model still provide weak causal evidence?

A. Prediction models cannot use operational data

B. Accuracy proves that the model is biased

C. The model may exploit associations that would change after an intervention or policy change

D. Causal studies cannot include algorithms

Correct Answer: C. The model may exploit associations that would change after an intervention or policy change

Question 5

Which is the strongest justification for using a case study in doctoral operations research?

A. The organization is convenient to access

B. The case is theoretically critical, revelatory, or suitable for process explanation

C. The researcher wants to avoid collecting quantitative data

D. A single case automatically represents the industry

Correct Answer: B. The case is theoretically critical, revelatory, or suitable for process explanation

Question 6

What is the principal concern with treating supply chain autonomy as a binary condition?

A. It prevents measurement of delivery speed

B. It ignores different levels of decision authority, human intervention, and risk

C. It assumes all algorithms are inaccurate

D. It eliminates the need for governance

Correct Answer: B. It ignores different levels of decision authority, human intervention, and risk

Question 7

Which finding would most clearly indicate over-reliance on an AI system?

A. Planners reject all recommendations

B. Planners accept incorrect recommendations despite contradictory contextual information

C. Planners request explanations before acting

D. Planners override recommendations and improve outcomes

Correct Answer: B. Planners accept incorrect recommendations despite contradictory contextual information

Question 8

Why must regenerative-operations research specify a baseline and system boundary?

A. To ensure that any recycling activity is classified as regenerative

B. To determine whether claimed improvements represent genuine net ecological or social restoration rather than burden shifting

C. To remove stakeholders from the analysis

D. To focus exclusively on short-term cost

Correct Answer: B. To determine whether claimed improvements represent genuine net ecological or social restoration rather than burden shifting

Question 9

Which feature best characterizes responsible doctoral research involving operational employee data?

A. Collecting every available variable to maximize prediction

B. Disclosing identities to improve managerial accountability

C. Using data minimization, appropriate consent, secure handling, and assessment of potential harms

D. Allowing the sponsoring firm to approve all conclusions

Correct Answer: C. Using data minimization, appropriate consent, secure handling, and assessment of potential harms

Question 10

In the Meridian Foods case, why is mixed methods particularly appropriate?

A. No single method can produce any valid evidence

B. Different methods can estimate effects, examine reliance mechanisms, explain adaptation, and explore trade-offs

C. Qualitative evidence is required to replace all performance metrics

D. Optimization models automatically establish causal effects

Correct Answer: B. Different methods can estimate effects, examine reliance mechanisms, explain adaptation, and explore trade-offs