Research Methodology & Technical Writing

Chapter 4
Research Philosophy, Theory, and Design

Bachelors Program in AI & ML
Research Methodology & Technical Writing • Bachelor

Research Philosophy, Theory, and Design

Introduction

Research is not simply the collection and analysis of data. It is a reasoned process for producing defensible knowledge. Every research project rests on assumptions about reality, knowledge, evidence, values, and the relationship between the researcher and the researched. These assumptions influence the research questions that are considered legitimate, the data that are collected, the methods used to analyse them, and the claims that can responsibly be made.

For students of Artificial Intelligence and Machine Learning, these issues are especially important. An AI project may involve measuring model accuracy, interpreting user experiences, investigating algorithmic bias, or collaboratively redesigning a system with affected communities. Each purpose may require a different philosophical perspective and research design. A technically sophisticated model can still produce weak research if its question, evidence, and interpretation are poorly aligned.

This chapter examines four influential research perspectives—positivism, interpretivism, critical approaches, and pragmatism—and connects them to theory, conceptual models, and common research designs. It also considers how researchers select between quantitative, qualitative, and mixed-method approaches, and between cross-sectional, longitudinal, case-study, and action-research designs.

Learning Objectives

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

1. Explain the relationship between ontology, epistemology, methodology, and methods.

2. Compare positivist, interpretivist, critical, and pragmatic perspectives.

3. Distinguish theories, conceptual frameworks, constructs, variables, and hypotheses.

4. Select research designs appropriate to different questions and purposes.

5. Explain the strengths and limitations of quantitative, qualitative, and mixed-method research.

6. Compare cross-sectional, longitudinal, case-study, and action-research designs.

7. Develop a coherent alignment between a research question, paradigm, theory, design, and evidence.

8. Recognise ethical, governance, and validity issues in AI and ML research.

1. The Foundations of Research Thinking

1.1 Ontology and epistemology

Ontology concerns assumptions about reality. A realist ontology generally assumes that a world exists independently of the researcher, although it may be observed imperfectly. A more relativist ontology emphasises that social reality is shaped by language, culture, history, and human interpretation. For example, “model fairness” may be treated either as a measurable property of prediction outcomes or as a contested social judgement that differs among stakeholders.

Epistemology concerns how knowledge can be obtained and justified. A positivist epistemology gives priority to systematic observation, measurement, and testing. An interpretivist epistemology views knowledge as emerging through interpretation of meanings and experiences. Critical epistemology asks whose knowledge is recognised, whose interests are served, and how power influences what is accepted as truth.

Methodology is the overall logic linking philosophical assumptions to research procedures. A methodology is broader than a method. Surveys, interviews, experiments, observation, and document analysis are methods; a quantitative, qualitative, mixed-method, case-study, or action-research strategy is part of a methodological design.

1.2 Values and reflexivity

Research is influenced by values even when researchers seek objectivity. Values affect the selection of a topic, the definition of a problem, the categories used in a dataset, and the interpretation of findings. Positivist research attempts to control value influence through standardisation and transparency. Interpretive and critical researchers make their positionality explicit through reflexivity—the systematic examination of how their background, assumptions, and relationships influence the research.

Reflexivity is highly relevant to AI. A researcher developing a facial-recognition dataset may assume that demographic categories are natural and fixed, although those categories may be socially constructed and technically consequential. A reflexive account would explain how categories were chosen, who participated in that decision, and what forms of identity or harm may remain invisible.

2. Major Research Perspectives

2.1 Positivism and post-positivism

Positivism holds that reliable knowledge is generated through systematic observation, measurement, and logical analysis. It typically favours clearly defined variables, replicable procedures, and the testing of hypotheses. In an AI study, a positivist design might investigate whether a feature-selection method improves classification accuracy under controlled conditions.

Typical features include:

operational definitions of concepts;
numerical measurement;
statistical estimation and hypothesis testing;
efforts to control confounding variables;
separation, as far as possible, between researcher and object of study; and
emphasis on reliability, validity, and replicability.

Strict positivism assumes that observations can provide direct access to facts. Post-positivism, more common in contemporary research, accepts that observation is fallible and that knowledge is provisional. Measurement error, sampling limitations, researcher assumptions, and model dependence are acknowledged. A post-positivist researcher does not claim absolute certainty; instead, the researcher evaluates the strength of evidence and alternative explanations.

Positivist approaches are appropriate when the purpose is to estimate prevalence, compare groups, predict outcomes, test causal explanations, or evaluate performance under defined conditions. They are less suitable when the central question concerns subjective meaning, unobserved power relations, or the lived experience of technology use.

2.2 Interpretivism and constructivism

Interpretivism argues that human action cannot be understood fully without examining the meanings people assign to their circumstances. Constructivist approaches further emphasise that knowledge is constructed through social interaction and interpretation. Rather than seeking one universal account, an interpretive study may examine multiple perspectives and the context in which they arise.

An interpretive AI study could explore how nurses understand recommendations produced by a clinical decision-support system. Interviews and observation might reveal that users do not interpret a confidence score as a probability, but as an institutional signal about accountability. The important outcome is not merely whether users follow the system, but how they make sense of it and how organisational routines shape that response.

Interpretive quality is assessed through credibility, depth, transparency, coherence, and attention to context. Common practices include purposive sampling, prolonged engagement, triangulation, member reflection, and an audit trail. Interpretive research does not mean that “anything goes”; interpretations must be supported by data and logically connected to the research context.

2.3 Critical perspectives

Critical research examines inequality, domination, exclusion, and the distribution of power. It treats knowledge as connected to social interests and often seeks not only to understand the world but also to transform unjust conditions. Feminist, Marxist, postcolonial, critical race, disability, and critical data studies are examples of critical traditions.

In AI research, a critical perspective may ask:

Who benefits from an automated system?
Who bears its risks or costs?
Which populations are absent from the training data?
How do procurement, labour, law, and institutional authority shape the system?
Does technical optimisation conceal a political decision?

A critical study of automated hiring might combine dataset analysis with interviews of applicants and recruiters, examining how historical employment inequalities are reproduced through apparently neutral features. The study may recommend changes to organisational policy, not merely a more accurate model.

Critical research values participation, historical analysis, and social justice. Its challenge is that researchers must support strong normative arguments without treating their preferred political conclusions as substitutes for evidence. Transparency about commitments and careful documentation are therefore essential.

2.4 Pragmatism

Pragmatism evaluates research ideas partly by their practical consequences and usefulness in addressing a problem. It does not require allegiance to one fixed view of reality or one category of method. A pragmatic researcher asks which combination of evidence will best answer the question and support an informed decision.

Pragmatism is particularly compatible with mixed-method research. For example, a study of an educational chatbot may use log data to estimate changes in usage and achievement, interviews to understand student experiences, and participatory workshops to identify improvements. The methods are combined because each contributes different evidence, not simply because methodological variety is desirable.

Pragmatism does not mean using any convenient method. Choices must still be justified by the question, feasibility, ethics, and quality of evidence. A pragmatic design may also expose tensions: numerical improvement may coexist with reduced student autonomy, or average performance may improve while a minority experiences harm.

2.5 Comparative overview

|---|---|---|---|---|

| Positivist/post-positivist | Knowledge is produced through systematic measurement and critical testing | Does X affect Y? How accurately can Y be predicted? | Experiments, surveys, structured datasets | Validity, reliability, replicability |

| Critical | Knowledge is shaped by power and should expose or challenge injustice | Who benefits, who is excluded, and why? | Testimony, historical records, institutional and technical data | Participation, transparency, emancipatory relevance |

These perspectives are not simply competing labels. A single project can contain different components, but the researcher must explain how the components fit together. Combining methods without addressing philosophical compatibility can produce an incoherent design.

3. Theory, Concepts, and Conceptual Models

3.1 What is a theory?

A theory is a systematic set of concepts and propositions that explains relationships, mechanisms, or patterns. Theory can guide the selection of variables, the formulation of hypotheses, the interpretation of findings, and the identification of alternative explanations. Examples relevant to AI include technology acceptance theory, diffusion of innovations, sociotechnical systems theory, and theories of algorithmic fairness.

Theory can be used deductively or inductively. In a deductive study, existing theory generates hypotheses that are tested with data. In an inductive study, patterns in the data contribute to the development of concepts or a new explanation. Some qualitative studies use abductive reasoning, moving iteratively between surprising observations and possible explanations.

3.2 Constructs, variables, and operationalisation

A construct is an abstract concept that cannot always be observed directly, such as trust, fairness, usability, or perceived usefulness. A variable is a measurable representation that can take different values. Operationalisation specifies how a construct will be observed or measured.

For example, “trust in an AI recommendation” could be operationalised through a validated questionnaire, reliance behaviour, interview accounts, or a combination. Each operationalisation captures a different aspect of the construct. A single numerical indicator should not automatically be treated as the complete meaning of trust.

In quantitative research, a conceptual relationship may be represented as:

Outcome = systematic effects of explanatory factors + other influences + measurement error

This is a conceptual relationship rather than a universal equation. It reminds the researcher that an observed outcome may reflect the variables of interest, unmeasured causes, and imperfect measurement. Causal interpretation requires more than observing that two variables are associated; it requires a defensible temporal order, plausible mechanism, and attention to alternative explanations.

3.3 Conceptual frameworks

A conceptual framework is a researcher’s organised representation of the concepts, relationships, assumptions, and context relevant to a study. It may be expressed in a diagram, narrative, or table. A framework is not necessarily a complete theory. It may integrate several theories and identify expected relationships for one specific investigation.

A useful framework for studying adoption of an AI tool might include:

perceived usefulness and ease of use;
organisational support and training;
trust and perceived risk;
professional norms;
actual adoption behaviour; and
contextual moderators such as workload and regulation.

The framework helps determine what data are needed. It also prevents a study from collecting variables merely because they are available.

3.4 Hypotheses and research questions

A hypothesis is a testable proposition, such as: “Greater perceived usefulness is associated with stronger intention to use the AI system.” A qualitative question might instead ask: “How do employees define usefulness when deciding whether to rely on the system?” The first prioritises measurable association; the second explores meaning and context. Neither is inherently superior. Their suitability depends on the purpose of the research.

4. Research Designs and Their Uses

4.1 Quantitative designs

Quantitative research uses structured data and numerical analysis to describe, compare, explain, or predict phenomena. Common designs include surveys, experiments, quasi-experiments, correlational studies, and secondary-data analysis.

An experiment manipulates an independent variable and observes its effect on an outcome, ideally using random assignment and control conditions. In AI, an experiment might compare human decisions with and without decision support. A quasi-experiment evaluates an intervention without full random assignment, for example by comparing departments that adopt a system at different times.

Strengths include comparability, estimation, model evaluation, and the ability to examine patterns across larger samples. Limitations include dependence on measurement quality, risk of omitted variables, artificiality of controlled settings, and the possibility that averages hide important subgroup experiences.

4.2 Qualitative designs

Qualitative research examines meanings, processes, experiences, interactions, and context. Methods include semi-structured interviews, focus groups, ethnography, observation, document analysis, and qualitative analysis of online or system-generated material.

Sampling is usually purposive rather than statistically representative. Participants are selected because they can illuminate the research question. Theoretical saturation refers to the point at which additional data no longer contribute substantially to developing relevant categories; it is not a mechanical rule and must be justified in relation to the study.

Qualitative design is valuable for understanding why a system is used or resisted, how categories are interpreted, and how institutional practices shape technical outcomes. Its limitations include time demands, interpretive disagreement, access constraints, and limited statistical generalisation. Researchers can nevertheless support analytic generalisation by showing how findings illuminate concepts or mechanisms that may be relevant in comparable settings.

4.3 Mixed-method designs

Mixed-method research intentionally integrates quantitative and qualitative evidence within one study. Common structures include:

Convergent design: quantitative and qualitative data are collected during a similar period and compared or integrated.
Explanatory sequential design: quantitative findings are collected first, followed by qualitative work that explains surprising or important results.
Exploratory sequential design: qualitative findings are used first to develop measures, hypotheses, or an intervention that is subsequently examined quantitatively.
Embedded design: one method provides a supporting component within a larger design.

Integration is the defining feature. A project is not genuinely mixed-method merely because it contains a survey and a few quotations. Researchers should state how findings will be connected—for example, through joint displays, case comparisons, or explanation of convergence and disagreement.

4.4 Cross-sectional design

A cross-sectional study collects data at one point or within one limited period. It is useful for describing current attitudes, estimating prevalence, and comparing groups. For example, a university may survey students’ trust in generative AI during one semester.

Its main limitation is weak evidence about change and temporal order. If trust and use are measured simultaneously, it may be unclear whether trust led to use or use changed trust. Cross-sectional associations should not automatically be described as causal effects.

4.5 Longitudinal design

A longitudinal study collects data across multiple time points. It may follow the same participants (panel design), repeatedly sample a population (repeated cross-sectional design), or examine a system before, during, and after an intervention.

Longitudinal research can examine change, adaptation, persistence, and delayed effects. In AI deployment, it may reveal that initial productivity gains decline when users become dependent on a system, or that confidence increases before accuracy improves. Challenges include participant attrition, changing measurement instruments, cost, and historical events that affect observations.

4.6 Case-study design

A case study is an in-depth investigation of a bounded case—such as an organisation, project, platform, community, policy implementation, or incident—using multiple sources of evidence. The case may be intrinsic, because it is important in itself, or instrumental, because it illuminates a broader issue.

A strong case study defines its boundaries, explains case selection, uses evidence from several sources, and distinguishes description from interpretation. A university’s deployment of an AI plagiarism detector, for example, could be studied through policy documents, system logs, interviews, classroom observation, and appeals records.

Case studies offer contextual richness and can reveal mechanisms that broad surveys miss. Their findings should not be generalised statistically without justification, but they may support theoretical or analytic generalisation.

4.7 Action research

Action research links systematic inquiry with practical change. It commonly follows cycles of planning, action, observation, reflection, and revised planning. Participants are collaborators rather than merely subjects.

An action-research project might work with teachers to introduce an adaptive-learning tool, monitor classroom effects, reflect on unintended consequences, and revise the tool or teaching practice. The design is suited to local improvement and professional learning. It can be difficult to maintain distance, preserve critical independence, and separate the effects of several simultaneous changes. Clear agreements about data ownership, participation, and reporting are essential.

5. Aligning Philosophy, Question, Theory, and Design

Research coherence can be assessed by asking four questions:

1. What is the problem and what kind of answer is needed? A prevalence estimate, causal explanation, lived experience, critique of power, or practical intervention requires different evidence.

2. What assumptions about knowledge are being made? Measurement, interpretation, participation, and problem-solving imply different philosophical commitments.

3. What theory or conceptual model guides the inquiry? The framework should influence the questions and evidence rather than appear only in the literature review.

4. Does the design produce evidence capable of supporting the intended claim? A one-time survey cannot ordinarily establish long-term effects; a small case study cannot automatically estimate population prevalence.

A useful alignment matrix can specify the question, paradigm, theory, design, sample, data source, analysis, and claim. For example:

|---|---|---|---|

6. Worked Design Example

Suppose a researcher asks: “Does an AI coding assistant improve the learning of first-year programming students?”

A narrow quantitative design might randomly assign students to access or no access, measure learning through a common assessment, and compare outcomes. The theory could involve cognitive load or learning-by-doing. The design can estimate differences under the specified conditions, but it may not explain how students used the tool.

An interpretive extension could interview students and observe coding sessions. It might discover that some students use the assistant to request explanations, while others copy solutions without understanding them. A critical component might examine whether students with weaker prior preparation are disproportionately penalised by unreliable suggestions or surveillance features.

A pragmatic mixed-method design could therefore:

1. measure assessment performance and patterns of tool use;

2. interview a purposive sample of students with contrasting outcomes;

3. compare experiences across demographic and educational groups; and

4. use findings to revise guidance and evaluate the intervention over two semesters.

The resulting claim would be more cautious and useful: the assistant’s educational value depends on how it is used, students’ prior knowledge, assessment design, and the support provided by instructors. This claim is stronger than either an unsupported assertion that the tool “works” or a single average score detached from context.

7. Illustrative Case Study: An AI Triage System in a Community Clinic

Background

The following is a fictional illustrative case. A community clinic introduces an AI system that prioritises incoming patient messages according to predicted urgency. The system is intended to reduce response time and help nurses manage workload.

Problem

After six months, average response time has fallen, but several nurses report that messages from older patients and patients with limited English proficiency are more likely to be placed in low-priority categories. Clinic leaders want to know whether the system is effective, why concerns have arisen, and how it should be governed.

Analysis and design choice

A purely positivist evaluation could compare response times, agreement with clinician triage, escalation rates, and adverse events before and after implementation. These indicators are important, but they may not reveal how staff interpret the scores or how patients experience delayed responses.

An interpretivist component could use interviews and observation to examine nurses’ reasoning, work routines, and trust in the system. A critical perspective would examine whether performance differs across language, age, disability, or socioeconomic groups, and whether affected patients have a meaningful route to challenge decisions. A pragmatic mixed-method case study would combine these components.

The study could use a longitudinal design, comparing several months before and after implementation. Quantitative analysis would report overall and subgroup performance, while qualitative analysis would investigate workflow and meaning. Researchers should avoid treating demographic variables as merely technical controls; they should explain why categories are collected, how they are defined, and what privacy risks they create.

Approach and possible response

The clinic might require human review for low-confidence or high-risk cases, provide multilingual message support, audit subgroup outcomes, retain an appeal process, and train nurses to treat the score as decision support rather than an instruction. The system should be evaluated continuously rather than approved permanently on the basis of an initial benchmark.

Learning points

The case demonstrates that effectiveness is multidimensional. Faster processing does not by itself establish safer or fairer care. Philosophical perspective affects what counts as evidence, while design determines whether the evidence can support a particular claim. Technical metrics, user experience, institutional practice, and power relations must be considered together in sociotechnical AI research.

8. Ethics, Governance, and Quality

Research involving AI may process personal, sensitive, or inferred data. Ethical review should address informed consent, privacy, data minimisation, secure storage, secondary use, and the possibility of re-identification. Participants should understand whether data will be used to train models and whether withdrawal is technically possible.

Researchers must consider fairness and representation. Removing protected attributes does not necessarily remove bias, because other variables may act as proxies. Reporting only aggregate accuracy can conceal unequal error rates. Where automated decisions affect people, accountability should include human oversight, explanation appropriate to the audience, documentation of limitations, and mechanisms for contesting outcomes.

Quality criteria depend partly on the paradigm. Quantitative studies should address construct validity, internal validity, external validity, reliability, statistical power, missing data, and robustness. Qualitative studies should address credibility, dependability, confirmability, transferability, reflexivity, and adequate evidence for interpretations. Mixed-method studies should justify integration and explain contradictory findings rather than selectively reporting agreement.

Open science practices—such as preregistration where appropriate, version-controlled analysis, data dictionaries, reproducible code, model cards, and transparent reporting—can improve trust. However, openness must be balanced with confidentiality, intellectual property, security, and the rights of communities whose data are represented.

9. Contemporary Developments and Best Practices

Generative AI is changing both the subject and practice of research. Researchers may use language models for transcription, coding assistance, literature discovery, or drafting, but outputs require verification. Fabricated citations, hidden training data, leakage of confidential information, and automation bias are significant risks. Institutions should specify acceptable use, disclosure expectations, and responsibility for errors.

Participatory and community-based approaches are increasingly important when research affects marginalised groups. Co-design can improve relevance and identify harms that technical teams overlook. Researchers should avoid extracting data from communities without reciprocal benefit or meaningful influence over decisions.

Good practice includes:

beginning with a precise research problem rather than a fashionable method;
making philosophical and positional assumptions explicit;
using theory to guide data collection and interpretation;
defining constructs carefully and checking whether measures represent them;
aligning the design with the intended claim;
planning for missingness, attrition, uncertainty, and alternative explanations;
integrating quantitative and qualitative evidence deliberately;
reporting negative, null, and contradictory findings;
involving affected stakeholders where appropriate; and
treating ethics and governance as design requirements, not administrative afterthoughts.

Challenges and Limitations

No research perspective provides a complete view of reality. Positivist designs may reduce complex experiences to indicators. Interpretive studies may be difficult to compare and may reflect researcher interpretation. Critical research may face disputes about normative commitments and researcher authority. Pragmatic research may become methodologically eclectic without a clear rationale.

All designs face practical constraints involving time, access, cost, expertise, sample quality, data security, and organisational cooperation. AI systems also change during research, creating problems of versioning and reproducibility. A model’s performance in a benchmark environment may not transfer to a real setting. Researchers should therefore state the boundaries of their evidence and avoid claims that exceed the design.

Chapter Summary

Research philosophy provides assumptions about reality, knowledge, values, and evidence. Positivist and post-positivist approaches emphasise measurement and systematic testing; interpretivist approaches examine meaning and context; critical approaches investigate power, inequality, and transformation; and pragmatism selects and integrates methods according to the problem and consequences of action.

Theory explains relationships and mechanisms, while conceptual frameworks organise the constructs and context of a particular study. Constructs must be operationalised carefully, and hypotheses or qualitative questions should follow from the research purpose. Quantitative, qualitative, and mixed-method approaches offer different but potentially complementary forms of evidence.

Cross-sectional designs describe a point in time, whereas longitudinal designs examine change. Case studies provide bounded, contextual analysis, and action research connects inquiry with collaborative improvement. A coherent study aligns its philosophical perspective, research question, theory, design, methods, evidence, and intended claims. In AI and ML research, this alignment must also address fairness, privacy, accountability, stakeholder participation, and the changing nature of technical systems.

Key Terms

Ontology: Assumptions about the nature of reality.

Epistemology: Assumptions about how knowledge is produced and justified.

Methodology: The logic connecting research philosophy, design, and methods.

Positivism: A perspective emphasising systematic observation, measurement, and testing.

Post-positivism: A fallibilist view that treats knowledge as provisional and affected by uncertainty.

Interpretivism: An approach focused on understanding meanings and situated human action.

Constructivism: The view that knowledge and social realities are constructed through interaction and interpretation.

Critical research: Inquiry that examines power, inequality, exclusion, and possibilities for transformation.

Pragmatism: A perspective that evaluates methods by their suitability for addressing practical problems.

Theory: An organised set of concepts and propositions explaining relationships or mechanisms.

Conceptual framework: A structured representation of concepts, assumptions, and expected relationships in a study.

Construct: An abstract idea such as trust, fairness, or usability.

Variable: A measurable characteristic that can take different values.

Operationalisation: The specification of how an abstract construct will be observed or measured.

Cross-sectional design: Research conducted at one point or within a short period.

Longitudinal design: Research that collects evidence across multiple time points.

Case study: In-depth investigation of a bounded case using one or more sources of evidence.

Action research: Cyclical inquiry linking investigation, action, reflection, and improvement.

Reflexivity: Examination of how the researcher’s position and assumptions influence the study.

Triangulation: Deliberate comparison of evidence from different sources, methods, researchers, or perspectives.

Review Questions

Short Answer Questions

1. Distinguish ontology from epistemology.

2. What is the main difference between a theory and a conceptual framework?

3. Why is a cross-sectional design limited for establishing temporal order?

4. Define reflexivity and explain its relevance to AI research.

5. What makes a study genuinely mixed-method rather than merely multi-method?

Long Answer Questions

1. Compare positivist, interpretivist, critical, and pragmatic perspectives in terms of assumptions, research questions, evidence, and quality criteria.

2. Explain how constructs are operationalised in quantitative research and discuss the risks of using inadequate measures in AI studies.

3. Discuss the strengths and limitations of cross-sectional and longitudinal designs for evaluating an AI deployment.

4. Explain the purposes, structures, and integration challenges of mixed-method research.

5. Describe how a case-study design can investigate both technical performance and organisational consequences of an AI system.

Analytical / Application Questions

1. Design a research study to investigate whether a university chatbot improves student learning. Justify the paradigm, theory, design, methods, and intended claims.

2. An automated hiring tool produces higher average accuracy but lower selection rates for one demographic group. Analyse the issue from positivist, interpretivist, critical, and pragmatic perspectives.

3. A hospital wants to introduce an AI triage system. Propose an action-research or longitudinal mixed-method design, including ethical safeguards and measures of success.

Multiple Choice Questions with Answers

Question 1

Which perspective primarily focuses on understanding the meanings people assign to their experiences?

A. Positivism

B. Interpretivism

C. Behaviourism

D. Formalism

Correct Answer: B. Interpretivism

Question 2

What is the main purpose of a conceptual framework?

A. To replace the need for evidence

B. To organise relevant concepts and expected relationships

C. To guarantee causal proof

D. To determine the sample size automatically

Correct Answer: B. To organise relevant concepts and expected relationships

Question 3

Which design is most directly suited to studying change across several time points?

A. Cross-sectional design

B. Longitudinal design

C. One-time case report

D. Post hoc classification

Correct Answer: B. Longitudinal design

Question 4

A study that combines surveys with interviews and deliberately integrates their findings is best described as:

A. Mixed-method research

B. Purely experimental research

C. Archival research only

D. Unstructured observation

Correct Answer: A. Mixed-method research

Question 5

Which perspective most explicitly asks who benefits from and who is harmed by an AI system?

A. Critical perspective

B. Positivist perspective

C. Descriptive perspective

D. Instrumental measurement

Correct Answer: A. Critical perspective

Question 6

In research, operationalisation means:

A. Choosing a journal for publication

B. Converting a construct into observable or measurable indicators

C. Removing all variables from a model

D. Repeating a study without changing its context

Correct Answer: B. Converting a construct into observable or measurable indicators

Question 7

Which feature is central to action research?

A. A single measurement with no intervention

B. Cycles of planning, action, observation, and reflection

C. Exclusive use of randomised experiments

D. Avoidance of participant involvement

Correct Answer: B. Cycles of planning, action, observation, and reflection

Question 8

A major limitation of a cross-sectional study is that it usually:

A. Cannot collect numerical data

B. Cannot describe current conditions

C. Provides weak evidence about temporal order

D. Requires repeated measurement over years

Correct Answer: C. Provides weak evidence about temporal order

Question 9

A case study is best characterised as:

A. An in-depth investigation of a bounded case

B. A probability sampling formula

C. A method restricted to laboratory experiments

D. A design that always permits statistical generalisation

Correct Answer: A. An in-depth investigation of a bounded case

Question 10

Pragmatism generally selects methods according to:

A. A requirement to use only numerical data

B. The practical problem and the evidence needed to address it

C. A rejection of all theory

D. The researcher’s preferred software

Correct Answer: B. The practical problem and the evidence needed to address it

Research Methodology & Technical Writing
Chapter 4 of 14

Chapter 4 Research Philosophy, Theory, and Design