Media Data Collection and Fieldwork

Introduction

A media management dissertation is only as credible as the evidence on which it rests. Claims about audience behaviour, newsroom performance, platform reach, revenue, costs, organisational culture or digital engagement require systematically collected evidence rather than convenient impressions. Media data collection and fieldwork therefore involve much more than distributing questionnaires or downloading analytics. They require a defensible research design, negotiated access, appropriate sampling, reliable instruments, secure data handling and critical interpretation.

In mass media organisations, evidence is distributed across several environments. Audience data may be held in ratings systems, subscription databases, surveys and focus groups. Organisational evidence may include editorial policies, schedules, workflow records, staff rosters and meeting documents. Financial information may be contained in budgets, management accounts, invoices and advertising reports. Digital evidence may come from websites, social platforms, mobile applications, customer-relationship systems and content-management systems. These sources differ in purpose, completeness, ownership, measurement logic and risk. A competent researcher must understand these differences before combining them.

Fieldwork is the planned process of obtaining primary or contextual evidence from people, organisations, documents, systems and environments. It includes preparation, access negotiation, collection, quality control, documentation, storage and closure. This chapter presents a practical framework for conducting fieldwork in media management research while protecting participants, organisations and data subjects.

Learning Objectives

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

1. distinguish major forms of media-management data and assess their strengths and limitations;
2. design a fieldwork plan linked to research questions and constructs;
3. select appropriate sampling, recruitment and data-collection methods;
4. evaluate reliability, validity, provenance, completeness and comparability;
5. manage organisational access, confidentiality, consent and data security;
6. collect and document audience, financial, organisational, digital and platform data responsibly;
7. identify bias, missingness, measurement error and ethical risks; and
8. prepare an auditable dataset and fieldwork record for analysis and dissertation reporting.

1. The Evidence Architecture of Media Organisations

A useful starting point is to classify evidence by origin and function.

The distinction between primary and secondary data is important but not absolute. An interview conducted by the researcher is primary evidence. A platform dashboard created for an organisation may be secondary evidence for a dissertation, even though it records first-party activity. The more important questions are: Who generated the data? For what purpose? Under what definitions? With what controls? Who can alter it? And does it measure the construct required by the research question?

1.1 Constructs, indicators and operationalisation

A construct is an abstract concept such as audience loyalty, editorial autonomy, employee engagement or platform dependence. An indicator is an observable measure used to represent that construct. Operationalisation converts the research concept into observable variables, categories or questions.

For example, “audience loyalty” might be operationalised through subscription duration, repeat visits, renewal intention and qualitative accounts of attachment. None is identical to loyalty. Each captures a different aspect. A dissertation should explain why particular indicators are appropriate and should avoid treating a convenient metric as a complete representation of a complex concept.

A data map can connect the research question to evidence:

This mapping prevents “data availability” from silently replacing research relevance.

2. Designing a Fieldwork Plan

A fieldwork plan translates a research design into executable actions. It should specify the research question, units of analysis, variables or themes, sources, instruments, sampling strategy, access requirements, timetable, responsibilities, risks and storage arrangements.

The unit of analysis is the entity about which a conclusion is made. It may be an individual audience member, article, programme episode, social post, newsroom, organisation, campaign or month. Confusing units creates invalid conclusions. For example, thousands of posts from one media outlet do not automatically provide evidence about all media organisations. Similarly, a survey of employees measures reported perceptions, not necessarily actual organisational performance.

A fieldwork protocol should include:

• a concise purpose statement;
• inclusion and exclusion criteria;
• definitions of variables and coding categories;
• recruitment and access procedures;
• interview, observation or survey instruments;
• procedures for recording consent and refusals;
• quality checks and version control;
• data-security arrangements;
• escalation procedures for unexpected risks; and
• a fieldwork completion and disposal plan.

2.1 Pilot work

Pilot work is a small-scale test conducted before full collection. It can reveal ambiguous questions, excessive interview length, unavailable fields, inappropriate response categories, broken links, duplicate records and unrealistic access assumptions. A pilot is not merely a rehearsal; it may lead to a justified revision of the instrument or sampling plan.

For a questionnaire, cognitive piloting asks participants how they interpret each question. For an interview schedule, piloting tests sequencing, neutrality and whether prompts elicit relevant detail. For a platform-data extract, piloting checks field definitions, time zones, pagination, export limits and whether historical data are retained.

3. Sampling and Recruitment

Sampling determines whose or what evidence enters the study. A probability sample gives units a known, non-zero chance of selection and supports statistical generalisation when assumptions are met. Examples include simple random, systematic, stratified and cluster sampling. A non-probability sample does not provide known selection probabilities. Examples include purposive, quota, convenience, snowball and volunteer samples. Non-probability sampling can be appropriate for exploratory qualitative research or hard-to-reach professional groups, but claims must match the design.

3.1 Sampling decisions

The researcher should identify:

1. the target population;
2. the accessible population;
3. the sampling frame, if one exists;
4. the unit of selection;
5. the selection method;
6. the intended sample size or stopping rule; and
7. likely sources of undercoverage and non-response.

Stratification may improve coverage of relevant subgroups, such as age bands, geographic markets, job roles or organisation size. However, it does not automatically remove bias. A quota sample that matches population proportions on age may still be biased by recruitment channel, media use or willingness to participate.

For a simple random sample estimating a proportion, a commonly used initial approximation is:

Here, n₀ is the initial required sample size, z is the standard-normal critical value for the chosen confidence level, p is the anticipated population proportion, and e is the desired margin of error expressed as a proportion. The approximation assumes a probability sample, independent observations and a sufficiently large population. When p is unknown, 0.5 gives the largest variance and therefore the conservative initial size. The calculation does not correct for non-response, design effects or a finite population.

For a finite population of size N, a commonly used correction is:

This correction is relevant when the sample is a substantial fraction of the population and the probability-sampling assumptions remain appropriate. It does not cure a biased sampling frame.

Worked example. Suppose a researcher wants a 95% estimate of a proportion with margin of error 5%, uses z ≈ 1.96 and adopts p = 0.5. Then n₀ = (1.96² × 0.5 × 0.5) / 0.05² ≈ 384.16, so the initial target is about 385 completed responses. If the accessible population is 2,000, the finite-population correction gives approximately n = 384.16 / (1 + 383.16/2000) ≈ 322.3, or 323 completed responses. This is a planning estimate, not a guarantee of representativeness.

3.2 Recruitment in organisational settings

Access through a senior manager can create pressure on employees to participate. Invitations should therefore explain that participation is voluntary, non-participation has no employment consequence and managers will not see individual responses. Recruitment lists should not disclose unnecessary personal information to the researcher. Where possible, an independent contact or survey link should be used.

4. Collecting Different Forms of Media Data

4.1 Audience research

Audience data may be behavioural, attitudinal, demographic or contextual. Behavioural data record actions, such as listening, viewing, clicking, subscribing or renewing. Attitudinal data record perceptions, motivations and intentions. Demographic data describe characteristics, but categories should be justified and collected only when analytically necessary.

Questionnaires require clear wording, mutually intelligible response options and a logical sequence. Avoid double-barrelled questions, leading language, undefined terms and unnecessary sensitive questions. A five- or seven-point agreement scale can measure an attitude, but the researcher must state whether the scale is treated as ordinal or used as an approximate interval measure in analysis. “Engagement” should not be assumed to mean the same thing as “attention,” “satisfaction” or “loyalty.”

Interviews are valuable when the research concerns interpretation, decision-making or institutional experience. A semi-structured schedule combines consistency with the opportunity to probe. Questions should be open enough to elicit evidence but focused enough to remain connected to the research question. Interviewers should distinguish participants’ direct observations from speculation and retrospective reconstruction.

Focus groups can reveal shared language and disagreement, but participants may conform to dominant views or withhold commercially sensitive information. They are unsuitable when confidentiality between participants cannot reasonably be maintained.

4.2 Organisational records and documents

Records should be assessed for provenance, purpose, completeness, version, date and authorisation. A published annual report and an internal budget may use different accounting bases or reporting periods. A newsroom workflow document may describe formal procedure rather than actual practice. Document analysis should therefore compare documents with interviews, observation or other records where feasible.

A document log should record an identifier, title, date, source, access conditions, version, relevance and any restrictions. Researchers should preserve the original file separately from a working copy and record transformations such as redaction, optical character recognition or conversion to text.

4.3 Financial information

Financial data require particular care because definitions can vary. Revenue may be gross or net of commissions, costs may be direct or allocated, and “profit” may refer to operating profit, contribution, earnings before interest and tax or another measure. The researcher should identify the accounting period, currency, recognition basis, tax treatment and whether figures are audited.

Comparisons require consistent denominators and periods. A rise in nominal revenue may reflect inflation, exchange-rate movement or an acquisition rather than improved underlying performance. Confidential figures should be reported in aggregated, indexed or banded form where exact disclosure would create risk.

4.4 Digital and platform metrics

Digital metrics are measurements produced by systems designed for operational, advertising or product purposes. Common metrics include impressions, reach, views, watch time, unique users, sessions, clicks, engagement actions and conversions. Their definitions are not universal. A “view” may require a minimum duration on one platform and a different condition on another. A “user” may be a browser, device, account or statistically inferred person.

Researchers should record:

• metric name and platform definition;
• collection date and time zone;
• time period and attribution window;
• filters, segments and exclusions;
• whether values are estimated, sampled or deduplicated;
• changes to the platform interface or algorithm;
• export method and file format; and
• permissions governing reuse.

A useful distinction is between exposure, interaction, retention and outcome. Impressions indicate delivered opportunities for exposure, not necessarily attention. Clicks indicate an interaction, not satisfaction. Conversions may be influenced by several channels and attribution rules. Platform dashboards should therefore be interpreted as operational measures, not neutral mirrors of audience reality.

4.5 Observation and field notes

Observation may be overt, in which participants know they are being observed, or covert, which raises substantial ethical concerns and is normally inappropriate in organisational research without exceptional justification. Field notes should separate descriptive notes from interpretation. A robust template records date, setting, participants or roles, observed action, relevant quotation, researcher reaction and emerging analytical questions.

5. Data Quality and Measurement Integrity

Data quality is multidimensional. Accuracy concerns closeness to the intended value; completeness concerns missing required information; consistency concerns agreement across records; timeliness concerns currency; validity concerns whether the measure represents the intended construct; and uniqueness concerns absence of unintended duplicates.

Quality control should occur during collection, not only after it. Examples include range checks, required-field checks, duplicate detection, timestamp validation, interview-audio checks and reconciliation of totals with source reports. Every correction should be logged rather than silently overwriting the original.

5.1 Reliability and validity

Reliability concerns consistency of measurement. A questionnaire may produce stable responses but still fail to measure the intended construct. Validity includes content validity, construct validity and, where relevant, criterion-related validity. In qualitative research, credibility, dependability, confirmability and transferability are commonly used quality principles.

Triangulation compares different sources, methods, investigators or theoretical interpretations. It is not a requirement that all sources agree. Contradiction can reveal different perspectives, changing conditions or measurement problems. A researcher should explain why evidence converges or diverges.

For a binary field, a simple completeness rate can be described as:

The numerator and denominator must refer to the same defined field set and records. A high completeness rate does not establish accuracy or validity.

5.2 Missing data and non-response

Missingness may be accidental, structurally unavailable, refused, not applicable or suppressed for confidentiality. These categories should not be collapsed automatically. Non-response can distort audience surveys if respondents differ systematically from non-respondents. Researchers should report response procedures, refusals where known, incomplete cases and any justified treatment of missing values.

6. Access, Ethics and Governance

Access is both a practical and ethical relationship. A gatekeeper’s permission does not replace individual consent. An organisation may permit access to records that contain personal data, but the researcher still needs a lawful and ethically defensible basis for processing them and must minimise collection.

A participant information sheet should explain the study purpose, procedures, expected duration, risks, benefits, voluntary nature of participation, withdrawal arrangements, confidentiality limits, data retention and contact details. Consent should be informed, specific and appropriately recorded. For interviews, consent to participate, audio-recording and quotation may need to be separate choices.

Key governance principles include:

• data minimisation: collect only what is necessary;
• purpose limitation: do not repurpose data incompatibly without justification;
• access restriction: provide data only to authorised individuals;
• pseudonymisation: replace direct identifiers where identity is not required;
• retention control: keep data only for the approved period;
• accountability: maintain a record of decisions and responsibilities; and
• secure disposal: delete or destroy data safely at the end of the retention period.

Encryption should be used for devices and transfers where appropriate, with strong authentication and separate storage of identifiers and research data. Email is often unsuitable for unprotected sensitive files. Cloud services should be approved by the institution and configured with least-privilege access.

Legal requirements vary by jurisdiction, but researchers should consider data-protection law, copyright, confidentiality, employment law, intellectual-property restrictions, platform terms and freedom-of-information limitations. Public availability does not automatically make data ethically unrestricted. A publicly visible social-media post may still involve contextual privacy expectations, vulnerable users or re-identification risk.

7. Managing Fieldwork Records

A fieldwork repository should contain a protocol, approvals, consent records, instrument versions, recruitment log, source-data register, cleaned-data version, codebook, transformation log, analysis-ready files and a readme file. File names should be consistent and dates should use an unambiguous format such as YYYY-MM-DD.

A codebook defines every variable, label, unit, allowed value, missing-value code and source. For qualitative data, a coding framework should define themes, inclusion rules, exclusion rules and examples. Coding should remain open to revision, but changes must be documented.

A basic access matrix can distinguish roles such as principal researcher, supervisor, transcriber and analyst. Each role should receive only the permissions necessary for its task. Access reviews are particularly important when team membership changes.

8. Technology and AI in Fieldwork

Digital tools can reduce transcription time, identify duplicates, organise documents and assist preliminary coding. Automated speech recognition, however, may misrecognise accents, names and specialist terminology. Machine-generated summaries can omit disagreement or invent connections. AI outputs must therefore be treated as provisional, checked against source material and clearly documented.

Researchers should not upload confidential interviews, personal data or proprietary records to an AI service unless institutional approval, contractual safeguards and an appropriate legal basis exist. Prompts, model versions, outputs and human corrections should be recorded when AI materially contributes to data processing. AI must not replace informed consent, interpretation or accountability.

Contemporary fieldwork increasingly involves API restrictions, changing platform definitions, privacy-enhancing technologies, server-side measurement, synthetic users and automated content moderation. These developments make longitudinal comparison difficult. A researcher should capture methodological metadata at the time of extraction rather than relying on a dashboard that may later change.

9. Applied Case: Evaluating a Local News Subscription Strategy

Background

The fictional Riverside Media Group operates a local news website, weekly print edition and podcast. Management believes that social-media traffic is increasing brand awareness but producing few paying subscribers. A bachelor’s dissertation investigates how platform dependence affects subscription conversion.

Problem

The organisation offers the researcher analytics, but the dashboard contains inconsistent definitions of “session” and “conversion” across months. Editors are concerned that interviews may expose weaknesses in platform strategy, while employees fear that critical comments could be reported to managers.

Analysis and approach

The researcher first defines the unit of analysis as the website visit for quantitative analysis and the employee or manager for qualitative analysis. The data map includes monthly referral source, landing page, subscription event and attribution window. The researcher requests a data dictionary, records the extraction date and limits the period to twelve comparable months.

The study uses three evidence streams: a de-identified analytics extract, a document review of platform and subscription policies, and semi-structured interviews with editorial, commercial and product staff. Purposive sampling ensures representation of different functions, but the dissertation avoids claiming that the interviewees represent all media workers. Invitations are sent by the researcher rather than through line managers, and interview transcripts use pseudonyms.

The researcher pilots the interview schedule with one former employee and discovers that “platform dependence” is interpreted differently by editorial and commercial staff. The final schedule therefore asks about specific referral practices, algorithm changes, traffic decisions and perceived bargaining power. Analytics data are checked against monthly subscription records, but differences remain because the systems use different attribution rules.

Outcome

The evidence shows that social referrals generate substantial exposure but low directly attributed conversion. Interviews indicate that editors nevertheless value social distribution for discovery and public-service reach. The conclusion is not that social media is ineffective; rather, its value depends on the outcome being measured. The dissertation recommends separating exposure, engaged reading and subscription conversion, and reporting attribution limitations explicitly.

Learning points

The case demonstrates that access does not guarantee usable data, platform metrics require definition, triangulation can explain apparent contradictions, and employee protection is essential when research occurs inside a workplace. It also illustrates why a researcher should not infer organisational strategy from a single dashboard measure.

10. Common Challenges and Best Practices

Fieldwork frequently encounters gatekeeper delays, changing personnel, incomplete records, low response rates, inconsistent definitions, technical export failures and participant withdrawal. A resilient plan includes alternative sources, a documented change-control process and realistic time for cleaning and verification.

Best practice is to:

1. begin with research questions rather than available dashboards;
2. define constructs, units and metrics before collecting data;
3. pilot every instrument and technical extraction;
4. preserve source files and document every transformation;
5. separate identifiers from research data;
6. obtain consent and explain confidentiality honestly;
7. triangulate without forcing agreement;
8. report missingness, exclusions and access limitations;
9. distinguish descriptive evidence from causal claims; and
10. make the final dataset and fieldwork decisions auditable within ethical limits.

Chapter Summary

Rigorous media fieldwork connects research questions to suitable evidence through operationalisation, sampling and a documented collection protocol. Media-management data include audience research, organisational records, financial information, digital metrics, platform data, documents and primary research. Each source has distinctive definitions, purposes, biases and access conditions.

Reliable fieldwork requires piloting, careful recruitment, quality checks, provenance records, codebooks and version control. Researchers must distinguish exposure from engagement and outcomes, and must not treat proprietary platform metrics as universally comparable. Ethical practice requires voluntary consent, confidentiality, data minimisation, secure storage, restricted access and responsible retention. Triangulation strengthens interpretation by comparing perspectives and sources, while transparent reporting of missingness and limitations protects the credibility of conclusions. Technology and AI can support fieldwork, but human verification, institutional approval and accountability remain essential.

Key Terms

Operationalisation: Converting an abstract research construct into observable indicators or questions.

Unit of analysis: The entity about which a research conclusion is made.

Sampling frame: The list or system from which a sample is selected.

Probability sample: A sample in which units have known, non-zero selection probabilities.

Purposive sampling: Selecting cases because they possess characteristics relevant to the study.

Non-response: Failure of a selected or invited unit to provide usable data.

Construct: An abstract concept studied through indicators.

Indicator: An observable measure representing part of a construct.

Provenance: The documented origin, history and ownership of data.

Validity: The extent to which a measure represents the intended concept.

Reliability: The consistency of a measurement process or instrument.

Triangulation: Comparing multiple sources, methods or perspectives.

Pseudonymisation: Replacing identifying information with codes while retaining controlled re-identification capacity.

Data minimisation: Collecting no more personal or sensitive data than necessary.

Attribution window: The defined period in which an outcome is credited to an interaction or source.

Engagement: A specified form of audience interaction, such as a click, comment, share or viewing duration.

Codebook: A document defining variables, values, units and missing-data rules.

Gatekeeper: A person or institution controlling access to participants, records or settings.

Field notes: Systematic records of observations, context, reflections and emerging interpretations.

Review Questions

Short Answer Questions

1. Why is operationalisation necessary in media-management research?
2. Distinguish primary data from secondary data using a platform dashboard as an example.
3. What is the difference between exposure, engagement and conversion?
4. State two purposes of piloting a fieldwork instrument.
5. Why should a researcher maintain a codebook?

Long Answer Questions

1. Explain how a researcher could design a fieldwork plan for studying audience loyalty at a digital news organisation.
2. Compare probability and non-probability sampling, including their implications for generalisation.
3. Discuss the ethical and governance requirements for interviewing employees about organisational performance.
4. Explain why platform metrics may not be comparable across platforms or across time.
5. Evaluate the role of triangulation in improving the credibility of a media-management dissertation.

Analytical / Application Questions

1. A media organisation reports rising social-media views but falling subscriptions. Design a mixed-source fieldwork strategy to investigate the apparent contradiction.
2. An employee survey has high completion but many respondents use the same extreme response category. Identify possible explanations and propose quality checks.
3. You receive a confidential analytics file containing direct identifiers, unclear metric definitions and missing months. Explain the steps you would take before analysis.

Multiple Choice Questions with Answers

Question 1

Which concept identifies the entity about which a research conclusion is made?

A. Sampling frame

B. Unit of analysis

C. Indicator

D. Attribution window

Correct Answer: B. Unit of analysis

Question 2

Which sampling method selects participants because they possess characteristics relevant to the research question?

A. Convenience sampling

B. Simple random sampling

C. Purposive sampling

D. Systematic sampling

Correct Answer: C. Purposive sampling

Question 3

What is the primary purpose of a pilot study?

A. To guarantee statistical significance

B. To test and improve the research procedures before full collection

C. To replace the final sample

D. To remove the need for ethical approval

Correct Answer: B. To test and improve the research procedures before full collection

Question 4

A platform’s reported ‘view’ should be interpreted cautiously mainly because

A. all views are necessarily fraudulent

B. the term may have a platform-specific definition

C. views cannot ever be counted

D. views are always identical to subscriptions

Correct Answer: B. the term may have a platform-specific definition

Question 5

Which principle requires collecting only data necessary for the research purpose?

A. Data minimisation

B. Triangulation

C. Stratification

D. Attribution

Correct Answer: A. Data minimisation

Question 6

If z = 1.96, p = 0.5 and e = 0.05, the approximate initial sample size from n₀ = z²p(1−p)/e² is

A. 96

B. 196

C. 323

D. 385

Correct Answer: D. 385

Question 7

Which practice best protects employee participants in a workplace study?

A. Allowing managers to see named responses

B. Making participation compulsory

C. Separating identifiers and reporting findings in a way that limits identification

D. Removing the consent information sheet

Correct Answer: C. Separating identifiers and reporting findings in a way that limits identification

Question 8

What does a codebook primarily provide?

A. A list of platform passwords

B. Definitions of variables, values and missing-data rules

C. A replacement for source data

D. A legal contract with participants

Correct Answer: B. Definitions of variables, values and missing-data rules

Question 9

Triangulation involves

A. deleting contradictory evidence

B. comparing multiple sources, methods or perspectives

C. using only quantitative data

D. selecting the largest possible sample

Correct Answer: B. comparing multiple sources, methods or perspectives

Question 10

A high completeness rate demonstrates that a dataset is

A. necessarily valid

B. necessarily accurate

C. complete for the defined fields, but not necessarily accurate or valid

D. free from sampling bias

Correct Answer: C. complete for the defined fields, but not necessarily accurate or valid