Research Philosophies and Approaches

Introduction

Research methodology is not simply a collection of techniques for gathering and analysing data. It is grounded in assumptions about what exists, how knowledge can be obtained, whose interests research should serve, and how evidence should be interpreted. These assumptions influence the formulation of research questions, the selection of data, the role of the researcher, and the standards used to judge research quality.

For business analytics students, philosophical awareness is especially important. Analytics often involves numerical models, automated systems, organisational decisions, and human interpretations. A sales forecast may be assessed through statistical accuracy, but its use also depends on managers’ beliefs, institutional priorities, data governance, and the social consequences of decisions based on it. Consequently, business analytics research can draw on several philosophical traditions rather than relying on one universally correct approach.

This chapter examines five influential research philosophies: positivism, post-positivism, interpretivism, pragmatism, and critical realism. It then distinguishes deductive, inductive, and abductive reasoning and explains how ontology, epistemology, and axiology connect philosophical assumptions to methodological choices.

Learning Objectives

By the end of this chapter, you should be able to:

• explain ontology, epistemology, axiology, methodology, and methods;
• compare positivism, post-positivism, interpretivism, pragmatism, and critical realism;
• distinguish deductive, inductive, and abductive reasoning;
• connect philosophical assumptions with research questions and methods;
• select and justify an appropriate approach for a business analytics investigation;
• recognise ethical, practical, and professional consequences of philosophical choices.

1. The Philosophical Foundations of Research

A research philosophy is a coherent set of assumptions about the world and the production of knowledge. It does not prescribe a single technique. Instead, it provides a rationale for why particular questions, evidence, and procedures are appropriate.

1.1 Ontology: assumptions about reality

Ontology concerns the nature of reality. It asks questions such as:

• Does a social phenomenon exist independently of people’s perceptions?
• Are constructs such as customer loyalty or organisational culture real features of organisations, or are they created through language and interaction?
• Is reality stable and governed by regularities, or is it historically and contextually variable?

A realist ontology generally assumes that reality exists independently of the researcher, although researchers may understand it imperfectly. A relativist or constructivist ontology gives greater emphasis to multiple realities formed through social experience, interpretation, and interaction.

1.2 Epistemology: assumptions about knowledge

Epistemology concerns what can be known and how knowledge can be justified. It addresses the relationship between the researcher and the object of study. For example, a researcher may treat knowledge as strongest when it is based on observable measurements, or may regard participants’ interpretations as essential evidence about social reality.

Epistemological questions include:

• What counts as credible evidence?
• Can the researcher be detached from the research setting?
• Is explanation more important than understanding?
• How should uncertainty and researcher interpretation be addressed?

1.3 Axiology: assumptions about values

Axiology concerns values, ethics, and the role of judgement. Research is never entirely detached from values: researchers choose topics, define concepts, decide what to measure, and interpret results. A positivist may seek procedural objectivity, while an interpretivist may openly acknowledge the researcher’s position. Critical approaches go further by examining power, inequality, and whose interests research serves.

Axiology also has practical importance in analytics. Decisions about which groups to include in a dataset, which outcome to optimise, and what threshold to use in an automated system are value-laden. For example, a fraud-detection model that reduces financial loss but disproportionately flags a particular demographic raises both technical and ethical questions.

1.4 Methodology and methods

Methodology is the overall logic and strategy of inquiry. It links philosophical assumptions to research design. Methods are the specific procedures used to collect and analyse evidence, such as surveys, interviews, experiments, observation, document analysis, regression, classification, or thematic analysis.

The relationship can be represented as follows:

Ontology → Epistemology → Methodology → Methods → Evidence and conclusions

This sequence is not mechanically deterministic. The same method may be used within different philosophical traditions. For example, an interview can be used positivistically to obtain standardised responses, interpretively to explore meaning, or pragmatically as one source of evidence for solving a practical problem. What matters is the justification and use of the method.

2. Major Research Philosophies

2.1 Positivism

Positivism is associated with the view that reliable knowledge should be based primarily on observable and measurable phenomena. It seeks regularities, relationships, and, where possible, causal explanations. The researcher aims to maintain distance from the research setting, define concepts operationally, and use systematic procedures that can be replicated.

Typical positivist features include:

• clearly specified hypotheses;
• structured instruments and predefined variables;
• quantitative data and statistical analysis;
• attention to reliability, validity, and generalisability;
• an aspiration to minimise researcher influence.

A positivist business analytics study might test whether delivery time, price, and product availability predict customer churn. It may formulate hypotheses before collecting data and evaluate them using a statistical model.

Positivism is valuable when concepts can be measured consistently and when the purpose is comparison, prediction, or testing. Its limitations arise when important phenomena are difficult to quantify, when context changes the meaning of behaviour, or when participants’ interpretations are central to the research question. Treating a measurement as a complete representation of a complex construct can create false precision.

2.2 Post-positivism

Post-positivism retains the concern for evidence, systematic inquiry, and empirical testing but rejects the idea that researchers can achieve perfect objectivity or certain knowledge. Reality may exist independently of observation, yet all measurements are theory-laden, incomplete, and potentially fallible.

Post-positivist research therefore emphasises:

• probabilistic rather than absolute conclusions;
• falsification and attempts to identify disconfirming evidence;
• triangulation across data sources or methods;
• recognition of measurement error and researcher bias;
• transparent limitations and alternative explanations.

For example, a post-positivist analyst may develop a churn model but test its performance across customer segments, examine missing-data patterns, compare competing explanations, and report uncertainty rather than presenting the model as universally valid.

Post-positivism is often appropriate for applied analytics because business data are incomplete, noisy, and generated within changing systems. It supports disciplined measurement without assuming that numerical results provide an unmediated picture of reality.

2.3 Interpretivism

Interpretivism holds that social reality is shaped by meanings, language, relationships, and historical context. People do not merely respond to external conditions; they interpret situations and act on the basis of those interpretations. Researchers therefore seek understanding from the participants’ perspectives.

Interpretivist studies commonly use semi-structured interviews, observation, focus groups, case studies, and document analysis. Analysis may involve coding, thematic interpretation, narrative analysis, or discourse analysis. The researcher is not treated as completely separate from the research setting; reflexivity is required, meaning that the researcher examines how their identity, assumptions, and relationship with participants influence the study.

An interpretivist investigation of dashboard adoption might ask how managers understand performance indicators, how trust in data is developed, and why employees resist particular reporting systems. A purely numerical measure of adoption would not capture these meanings fully.

Interpretivism can provide rich, context-sensitive understanding, but its findings may not be statistically generalisable. Quality is assessed through credibility, coherence, transparency, reflexivity, and the adequacy of the interpretation rather than only through replication and numerical reliability.

2.4 Pragmatism

Pragmatism focuses on the consequences of inquiry and the usefulness of knowledge for addressing a problem. It does not require commitment to a single view that either quantitative or qualitative research is inherently superior. The research question determines the appropriate combination of evidence and methods.

A pragmatic study of employee turnover could combine administrative data, a predictive model, and interviews with employees who leave or remain. The quantitative component may identify patterns, while the qualitative component may explain how organisational practices are experienced. Mixed-methods research is frequently associated with pragmatism, although mixed methods can also be justified by other philosophies.

Pragmatism is particularly relevant in business analytics because organisations often need actionable recommendations rather than a philosophical demonstration that one type of evidence is universally best. Nevertheless, pragmatism should not mean using methods opportunistically. A pragmatic design still requires a clear research question, ethical procedures, appropriate quality criteria, and a reasoned account of how different forms of evidence fit together.

2.5 Critical realism

Critical realism distinguishes between reality itself, observable events, and the mechanisms that generate those events. It argues that social and organisational outcomes are produced by interacting structures, powers, and mechanisms, but that these may not be directly observable.

A critical realist analysis commonly distinguishes:

• the real: underlying structures, powers, and mechanisms;
• the actual: events that occur, whether or not they are observed;
• the empirical: events and experiences that are observed or reported.

For example, a company may observe a decline in productivity. The empirical evidence could include dashboard indicators and employee accounts. The actual event is the decline itself. The real level may involve mechanisms such as incentive structures, workload, management practices, skills, or economic pressure. A critical realist researcher asks not only whether variables are associated, but also what mechanisms may explain the association and under what conditions they operate.

Critical realism can support mixed methods and case-based explanation. It accepts that data are fallible and that social systems are open, meaning that outcomes may vary because many causal influences interact. Its challenge is that claims about underlying mechanisms require careful theoretical reasoning and cannot always be verified directly.

3. Comparing the Philosophies

These philosophies should not be treated as rigid categories. Researchers may work across traditions, but philosophical compatibility must be considered. Combining methods is not automatically the same as combining philosophies. A study should explain how its assumptions, research questions, data, analysis, and conclusions fit together.

4. Deductive, Inductive, and Abductive Reasoning

Research approaches describe how ideas and evidence are connected.

4.1 Deduction

Deductive reasoning moves from a general proposition or theory to specific expectations. A researcher identifies a theory, derives hypotheses, collects relevant data, and assesses whether the evidence is consistent with those hypotheses.

A simplified structure is:

1. Theory: perceived service quality influences customer retention.
2. Hypothesis: customers reporting higher service quality are less likely to churn.
3. Data: survey responses linked to subsequent customer behaviour.
4. Test: assess whether the predicted relationship is supported.

Deduction is useful for theory testing and structured quantitative research. However, failure to support a hypothesis does not automatically prove that the theory is false. Measurement problems, inadequate samples, omitted variables, and contextual conditions may also explain the result.

4.2 Induction

Inductive reasoning moves from observations toward patterns, concepts, or broader propositions. Researchers collect or examine data, identify recurring themes or relationships, and develop an explanation grounded in those observations.

For example, interviews with analysts may reveal recurring themes of mistrust, insufficient training, and unclear accountability. The researcher may then develop a conceptual explanation of resistance to artificial intelligence tools.

Induction is valuable when a topic is underexplored or when existing theories do not adequately reflect a particular setting. Its conclusions remain provisional: repeated observations can support a proposition but cannot guarantee that it applies in every future case.

4.3 Abduction

Abductive reasoning seeks the best available explanation for an observation. It often moves iteratively between data and theory. The researcher encounters a surprising pattern, considers possible explanations, gathers additional evidence, and refines the explanation.

Suppose a predictive model performs well overall but poorly for a particular customer group. An abductive investigation may consider data quality, different purchasing behaviour, unequal exposure to promotions, or an inappropriate modelling assumption. The researcher then investigates these possibilities and develops the explanation that best accounts for the evidence while remaining open to revision.

Abduction is especially useful in complex organisational settings where outcomes have multiple possible causes. It differs from deduction because it does not merely test a pre-existing hypothesis, and it differs from induction because it focuses explicitly on explanatory reasoning rather than only on pattern formation.

In practice, a single project may use all three. A researcher may deductively test an established model, inductively analyse unexpected interview findings, and abductively integrate the results into a revised explanation.

5. Connecting Philosophy to Methodological Choice

A defensible design begins with the research problem rather than with enthusiasm for a particular technique. The following sequence is useful:

1. Define the phenomenon and practical or theoretical problem.
2. Decide what kind of answer is required: measurement, prediction, understanding, explanation, or intervention.
3. Clarify assumptions about reality, knowledge, and values.
4. Select a reasoning approach or combination of approaches.
5. Choose a research design and methods that can produce relevant evidence.
6. Specify quality criteria, ethical safeguards, and limitations.

Consider three possible research questions about an analytics dashboard:

• Positivist/post-positivist: Does dashboard use predict faster operational decisions?
• Interpretivist: How do managers interpret and use dashboard indicators in daily work?
• Critical realist: What organisational mechanisms explain why the dashboard improves performance in some departments but not others?
• Pragmatic: Which combination of evidence can guide redesign of the dashboard?

The subject is similar, but the assumptions, evidence, analysis, and claims differ. A survey and statistical model may be suitable for the first question. Interviews and observation may be more suitable for the second. Comparative case analysis may help with the third. The fourth may justify a mixed-methods intervention-oriented design.

Quality criteria across philosophies

Quality cannot be reduced to one universal checklist. Common criteria include:

• Reliability: consistency of a measurement or procedure;
• Validity: whether a measure or inference adequately represents what it claims to represent;
• Generalisability: the extent to which findings apply beyond the study sample or setting;
• Credibility: the plausibility and evidential support of an interpretation;
• Transferability: the usefulness of findings in another context, supported by detailed contextual description;
• Dependability: the transparency and consistency of the research process;
• Confirmability: the extent to which findings are grounded in evidence rather than hidden researcher preference;
• Explanatory adequacy: whether the proposed account plausibly explains how and why outcomes occur.

A quantitative study may prioritise reliability, construct validity, and generalisability. An interpretive case study may emphasise credibility, reflexivity, and transferability. A critical realist study may emphasise theoretical coherence and explanatory power. A pragmatic project may evaluate whether the design produces trustworthy and useful decisions without neglecting ethical standards.

6. Illustrative Case: Predicting and Understanding Employee Turnover

The following is a fictional illustrative case.

Background

Northstar Services, a customer-support organisation, experienced rising employee turnover. The analytics team was asked to identify the causes and recommend action. The organisation possessed human-resources records, employee surveys, performance data, and exit-interview transcripts.

Problem

A purely predictive solution might identify employees with a high probability of leaving. However, management also needed to know whether predictions were fair, which organisational conditions contributed to turnover, and what interventions were likely to help.

Analysis through multiple philosophies

A positivist design could specify variables such as tenure, workload, absence, pay, promotion history, and survey scores. A model could test whether these variables predict subsequent turnover. This would provide a measurable basis for targeting further investigation.

A post-positivist design would assess data quality, examine missingness, test performance across demographic and departmental groups, compare alternative models, and report uncertainty. It would avoid treating a prediction as a fact about an individual.

An interpretivist component could use interviews to explore how employees experience workload, supervision, scheduling, and career opportunities. Two employees with similar recorded workloads might interpret their situations very differently because of managerial support or family responsibilities.

A critical realist analysis would connect observed patterns to mechanisms. High turnover in one department might reflect not merely workload but a combination of unstable scheduling, weak supervision, limited progression, and a local labour market offering better alternatives.

A pragmatic project could integrate these components to support intervention. The organisation might use the model to identify broad risk patterns, interviews to understand mechanisms, and a pilot programme to test changes in scheduling and supervisor training.

Approach and safeguards

The analytics team should not automatically label individuals as likely to leave or use the score to penalise them. It should establish a legitimate purpose, restrict access, explain the use of data, examine disparate impacts, and offer supportive interventions rather than punitive action. Employees should not be reduced to predictions, and the organisation should evaluate whether interventions improve working conditions.

Outcome

The combined investigation found that turnover was associated with workload, but interviews showed that workload became especially problematic when schedules changed unpredictably and supervisors offered little control or feedback. A pilot involving stable scheduling, supervisor training, and clearer progression information reduced reported dissatisfaction and improved retention indicators. The model remained useful for monitoring patterns, but it was not treated as a causal explanation or an automatic decision rule.

Learning points

• A prediction is not the same as an explanation.
• Quantitative and qualitative evidence can answer different parts of a problem.
• Philosophical assumptions influence what counts as a satisfactory conclusion.
• Analytics systems require ethical governance, not only technical validation.
• Abductive reasoning can help connect surprising patterns to organisational mechanisms.

7. Ethical, Governance, and Professional Implications

Philosophical choices have consequences for participants and organisations. Researchers should obtain informed consent where appropriate, protect confidentiality, minimise harm, and comply with relevant data-protection and institutional requirements. Secondary organisational data may still contain sensitive personal information and should not be assumed to be ethically unrestricted merely because it already exists.

Analytics researchers should distinguish correlation from causation, communicate uncertainty, document assumptions, and avoid overstating generalisability. Automated or statistical systems should be assessed for bias, unequal error rates, explainability, and effects on vulnerable groups. Data minimisation, access controls, secure storage, retention limits, and responsible disclosure are important governance practices.

Artificial intelligence increases the importance of philosophical reflection. A high-performing model may still be unsuitable if the target variable is a poor proxy for the real outcome, if historical data encode discrimination, or if deployment changes behaviour. Post-positivist testing, interpretive engagement with affected users, critical realist analysis of institutional mechanisms, and pragmatic evaluation of consequences can all contribute to responsible AI research.

8. Contemporary Developments and Best Practice

Contemporary business research increasingly uses mixed methods, computational social science, digital trace data, real-time experimentation, and machine learning. These developments create opportunities but also intensify concerns about privacy, reproducibility, platform power, algorithmic bias, and the opacity of proprietary systems.

Good practice includes:

• aligning the research question, philosophy, reasoning, design, and methods;
• stating assumptions explicitly rather than presenting methods as neutral;
• preregistering hypotheses when appropriate while allowing justified exploratory analysis;
• documenting data preparation, model choices, exclusions, and limitations;
• using triangulation when different evidence sources illuminate different aspects of a problem;
• involving relevant stakeholders, especially people affected by data-driven decisions;
• reporting null, negative, and unexpected findings;
• separating prediction, explanation, interpretation, and intervention claims;
• revisiting conclusions when new evidence or changing contexts challenge them.

Philosophy should therefore be understood as a practical resource. It helps researchers make defensible choices, identify limitations, and communicate the scope of their conclusions.

Chapter Summary

Research philosophy provides the assumptions underlying research design. Ontology concerns the nature of reality; epistemology concerns knowledge and evidence; axiology concerns values and ethics; methodology links these assumptions to a research strategy; and methods are the specific procedures used to collect and analyse data.

Positivism emphasises measurement and testing. Post-positivism retains empirical discipline while recognising uncertainty, fallibility, and the limits of observation. Interpretivism focuses on meaning, context, and socially constructed experience. Pragmatism selects methods according to the research problem and intended practical consequences. Critical realism distinguishes observable events from deeper mechanisms and structures that generate them.

Deduction moves from theory to evidence, induction develops concepts from observations, and abduction seeks the best explanation for surprising or complex findings. These forms of reasoning can be combined. In business analytics, philosophical awareness improves methodological fit, strengthens interpretation, and supports ethical governance of data and AI systems.

Key Terms

Ontology: Assumptions about the nature of reality.

Epistemology: Assumptions about knowledge and how it can be justified.

Axiology: The study of values, ethics, and value-related choices in research.

Methodology: The overall logic and strategy of inquiry.

Methods: Specific procedures for collecting and analysing evidence.

Positivism: A philosophy emphasising observable measurement, regularities, and empirical testing.

Post-positivism: A fallibilist approach that accepts empirical testing while recognising uncertainty and bias.

Interpretivism: A philosophy focused on meaning, context, and participants’ understandings.

Pragmatism: An approach that evaluates inquiry by its practical consequences and usefulness.

Critical realism: A philosophy that explains observable events through underlying mechanisms and structures.

Deduction: Reasoning from general theory or propositions to specific expectations.

Induction: Reasoning from observations toward patterns, concepts, or theory.

Abduction: Reasoning toward the most plausible explanation of evidence.

Reflexivity: Critical examination of how the researcher influences the research process.

Triangulation: Use of multiple data sources, methods, researchers, or theories to strengthen inquiry.

Reliability: The consistency of a measurement or procedure.

Validity: The extent to which an inference or measure represents what it claims to represent.

Generalisability: The extent to which findings apply beyond the study setting or sample.

Mechanism: A process or power that helps generate an observed outcome.

Mixed methods: The planned integration of quantitative and qualitative evidence in one study.

Review Questions

Short Answer Questions

1. What is the difference between ontology and epistemology?
2. State two characteristics of post-positivist research.
3. Why is reflexivity important in interpretivist research?
4. What is the main purpose of abductive reasoning?
5. Distinguish methodology from methods.

Long Answer Questions

1. Compare positivism and interpretivism in relation to reality, knowledge, researcher role, evidence, and quality criteria.
2. Explain how critical realism can improve the analysis of organisational performance data.
3. Discuss the philosophical assumptions underlying a pragmatic mixed-methods study.
4. Explain why correlation or prediction should not automatically be treated as causal explanation.
5. Evaluate the advantages and limitations of using post-positivism in business analytics.

Analytical / Application Questions

1. Design a research approach for investigating why employees do not trust an artificial-intelligence recruitment tool. Justify the philosophy, reasoning approach, data, and ethical safeguards.
2. A churn model performs accurately overall but poorly for one customer group. Use deductive, inductive, and abductive reasoning to propose a research process for investigating the finding.
3. An organisation wants to use employee risk scores to decide who receives promotion opportunities. Analyse this proposal using ontology, epistemology, axiology, and critical data-governance principles.

Multiple Choice Questions with Answers

Question 1

Which concept concerns assumptions about the nature of reality?

A. Ontology

B. Epistemology

C. Axiology

D. Methodology

Correct Answer: A. Ontology

Question 2

Which philosophy most directly emphasises participants’ meanings and contextual understanding?

A. Positivism

B. Interpretivism

C. Post-positivism

D. Critical realism

Correct Answer: B. Interpretivism

Question 3

What is a defining feature of post-positivism?

A. It rejects all empirical measurement

B. It assumes knowledge can be completely certain

C. It recognises fallibility and uncertainty in empirical inquiry

D. It requires only qualitative methods

Correct Answer: C. It recognises fallibility and uncertainty in empirical inquiry

Question 4

Which reasoning approach moves from a general theory to specific hypotheses?

A. Induction

B. Abduction

C. Interpretation

D. Deduction

Correct Answer: D. Deduction

Question 5

A researcher investigates an unexpected pattern and develops the explanation that best accounts for it. This is primarily:

A. Abduction

B. Deduction

C. Operationalisation

D. Replication

Correct Answer: A. Abduction

Question 6

Which philosophy focuses most explicitly on practical consequences and useful solutions?

A. Pragmatism

B. Positivism

C. Interpretivism

D. Structuralism

Correct Answer: A. Pragmatism

Question 7

In critical realism, underlying organisational powers and processes are usually treated as:

A. Methods

B. Mechanisms

C. Samples

D. Variables only

Correct Answer: B. Mechanisms

Question 8

Which statement best distinguishes a method from a methodology?

A. A method is philosophical, while methodology is a dataset

B. A method is a specific procedure, while methodology is the overall logic of inquiry

C. A method is always quantitative, while methodology is always qualitative

D. There is no meaningful distinction

Correct Answer: B. A method is a specific procedure, while methodology is the overall logic of inquiry

Question 9

Why should a predictive model not automatically be treated as a causal explanation?

A. Prediction never uses data

B. A prediction may identify association without identifying the mechanism producing the outcome

C. Causal explanations cannot use statistics

D. Models cannot be evaluated empirically

Correct Answer: B. A prediction may identify association without identifying the mechanism producing the outcome

Question 10

Which practice is most consistent with responsible analytics research?

A. Hiding unexpected findings to simplify the report

B. Using sensitive data without access controls because it already exists

C. Reporting assumptions, uncertainty, limitations, and potential group impacts

D. Treating a model score as an unquestionable fact about an individual

Correct Answer: C. Reporting assumptions, uncertainty, limitations, and potential group impacts