Communication Research

Research process and designs

The research process begins with a problem: a gap, a puzzle or a practical decision that existing knowledge does not settle. A good problem is narrowed into research questions that are clear, researchable and significant, and into objectives that state what the study will do. Where the study tests a relationship, questions are expressed as hypotheses: a null hypothesis states no relationship or no difference, while an alternative hypothesis states the expected relationship or difference. Hypotheses must be testable and falsifiable; a claim that cannot be disproved by any evidence is not a hypothesis but a belief. Variables give hypotheses their structure. An independent variable is manipulated or observed as the presumed cause, a dependent variable is the outcome measured, and intervening, moderating and extraneous variables complicate the relationship and must be controlled or acknowledged. Operational definitions state how each variable will actually be measured, turning an abstract concept such as media trust into observable indicators such as agreement with specific statements, and the quality of those definitions often decides the quality of the study.

A literature review situates the study in existing knowledge: it maps theories, previous findings, debates and gaps, and it justifies the new study by showing what remains unknown or contested. The review should be critical rather than a list of summaries, grouping studies by argument, noting methodological limits, and identifying the niche the new study fills. Research designs then translate questions into plans for evidence. Exploratory designs investigate little-known topics flexibly, through literature, expert interviews and pilot work, to generate questions and hypotheses. Descriptive designs portray characteristics of a population or phenomenon accurately, through surveys and content analysis, answering who, what, when and how much. Explanatory designs test why relationships occur, through experiments and causal analysis. Experimental designs manipulate an independent variable under controlled conditions with random assignment to experimental and control groups, giving the strongest causal evidence but often at the cost of artificiality. Quasi-experimental designs approximate experiments without full randomisation, as when existing classes or channels are compared. Longitudinal designs follow the same people or content over time, revealing change, while cross-sectional designs take a snapshot at one time, which is cheaper but cannot show change or establish temporal order. Case studies investigate a single instance in depth, ethnography studies a culture through sustained participation and observation, and action research combines inquiry with intervention, as when a newsroom studies its own practice to improve it. The design choice must be defended: an experiment cannot answer what a community means by trust, and a case study cannot establish how common a pattern is across a population, and pretending otherwise is a design error, not a matter of preference.

Sampling decides whom or what the study actually examines. A population is the whole group of interest, a sample is the subset studied, and a sampling frame is the list or procedure from which the sample is drawn. Probability sampling gives every member a known, non-zero chance of selection, which supports generalisation: simple random sampling selects by chance, systematic sampling selects every nth member, stratified sampling divides the population into groups and samples within each, and cluster sampling selects whole groups and studies all or a sample within them. Non-probability sampling lacks that known chance: convenience sampling takes whoever is available, purposive sampling selects cases for their relevance, snowball sampling reaches hidden populations through referrals, and quota sampling fills preset categories without random selection. Non-probability samples can be valuable for exploration and for hard-to-reach groups, but they cannot support statistical generalisation to a population, and a study that claims otherwise overreaches. Sample size depends on population diversity, desired precision, expected response and analysis plans, not on a magic number, and response rate, non-response bias and the match between frame and population must be reported, because a large but biased sample can mislead more than a smaller, well-drawn one.

Research areas, measurement and data collection

Communication research areas map onto the field itself. Media effects research studies influence on knowledge, attitudes and behaviour, from agenda setting to cultivation and persuasion. Audience research studies who uses media, how, with what gratifications and interpretations, through ratings, surveys, diaries and ethnographic viewing studies. Content research studies what media actually contain, through content analysis of news, programmes, advertisements and social media, revealing patterns, biases and changes over time. Production research studies newsrooms, studios and organisations, through observation and interviews, showing how routines, ownership and deadlines shape content. Policy and development research studies regulation, access and social change, evaluating campaigns and interventions. Reception research studies how audiences decode texts in everyday life, through interviews and focus groups that reveal negotiated and oppositional readings. Each area has characteristic methods, but the areas overlap: an effects claim needs content evidence about what was actually broadcast, audience evidence about exposure, and reception evidence about interpretation, and a study that supplies only one of the three is incomplete for strong effects conclusions.

Measurement turns concepts into numbers or categories that can be compared. Scales of measurement form a hierarchy. Nominal scales classify into categories without order, such as medium type or gender as recorded. Ordinal scales rank order without equal intervals, such as agreement from strongly disagree to strongly agree, or programme preference rankings. Interval scales have equal intervals but no true zero, such as temperature in Celsius or many attitude scores. Ratio scales have equal intervals and a true zero, such as viewing minutes, age, or number of articles, permitting all arithmetic operations. The hierarchy matters because it limits analysis: means are inappropriate for nominal data, and medians suit ordinal data better than means when distributions are skewed. Reliability concerns consistency: a measure is reliable when repeated applications give similar results, assessed through test-retest stability, internal consistency across items, and inter-coder agreement in content analysis. Validity concerns truth: a measure is valid when it measures what it claims, assessed as face validity, content coverage, criterion comparison with an external standard, and construct validity, whether the measure behaves as theory predicts. A measure can be reliable without being valid, consistently measuring the wrong thing, but it cannot be valid without being reasonably reliable, and pilot testing is the practical route to both, revealing confusing items, ceiling effects and coding disagreements before the main study.

Data collection methods each trade depth, breadth and reactivity. Questionnaires collect standardised responses from many people efficiently, but they depend on clear wording, avoid leading and double-barrelled questions, and suffer when respondents misunderstand items or give socially desirable answers. Interviews range from structured, with identical questions, through semi-structured, with a guide and probes, to unstructured, with open conversation, trading comparability for depth. Focus groups use group interaction to reveal norms, disagreements and language, but they risk dominance by vocal participants and conformity, so moderation and group composition matter. Observation ranges from non-participant watching to participant immersion, revealing behaviour that self-reports miss, but raising reactivity, where people change behaviour because they are observed, and ethical questions about consent. Content analysis systematically codes media content into categories, with a coding scheme, unit of analysis, coder training and reliability reporting, turning large volumes of text and image into countable patterns. Experiments and surveys can be combined with digital methods: web analytics, social media data and trace data offer scale and behavioural detail, but they raise consent, privacy and representativeness problems, because platform users are not a population sample and platform data reflect platform design as much as human behaviour. Triangulation, using more than one method, source or analyst, strengthens confidence when findings converge, and it is most valuable when methods have different weaknesses, so that their agreement is not an artefact of a shared flaw.

Data analysis, reporting and ethics

Data analysis begins with preparation: cleaning data, checking for missing values and outliers, coding open responses, and describing the sample honestly, including who did not respond. Descriptive statistics summarise: frequency distributions show how cases spread across categories, central tendency is captured by mean, median and mode, and dispersion by range, variance and standard deviation, which show how far cases spread around the centre. Cross-tabulation shows relationships between categorical variables, and correlation measures the strength and direction of association between numerical variables, from strong negative through zero to strong positive, without by itself proving causation. Inferential statistics generalise from sample to population with stated uncertainty: confidence intervals estimate a population value within a range, significance tests assess whether an observed difference is unlikely under the null hypothesis, and tests such as chi-square for categorical associations and t-tests for mean differences are chosen by variable type and design. The common errors are overclaiming: treating a significant result as important, treating correlation as causation, generalising from a convenience sample, and hiding non-significant results. Effect size, confidence intervals and honest reporting of all planned analyses guard against these errors better than significance stars alone.

Qualitative analysis follows a different logic. Transcripts, field notes and texts are coded into themes, patterns and exceptions, through open coding that names concepts, axial coding that connects them, and selective coding that builds an explanatory account, with constant comparison checking each new case against emerging themes. Discourse and textual analysis examine language, narrative and visual choices to show how meaning and power are constructed, while thematic analysis summarises patterns across cases without claiming statistical prevalence. Quality in qualitative work is judged by credibility, through prolonged engagement, triangulation and member checking; by transferability, through thick description that lets readers judge relevance to other settings; by dependability, through an audit trail of decisions; and by confirmability, through reflexivity about researcher influence. Mixed methods analysis integrates the strands: quantitative results may be explained by qualitative findings, qualitative cases may be selected from survey extremes, or both may address different facets of one question, and the integration must be reported, not left as two separate studies stapled together.

Reporting communicates findings, limits and implications in a structure readers can check: introduction and questions, literature and theory, methods in enough detail to replicate, results without selective omission, discussion that answers the questions and states what the study cannot show, and references that allow verification. Tables and charts must be honest: axes must not exaggerate, bases must be stated, and percentages must carry their counts where small numbers could mislead. Research ethics governs the whole process. Informed consent requires that participants understand purpose, procedures, risks and their right to withdraw, in language they understand, with special care for children, vulnerable groups and those with limited literacy. Confidentiality and anonymity protect identities in data storage and reporting, and deception is permitted only when justified, minimal and followed by debriefing. Plagiarism, fabrication and falsification are misconduct, not shortcuts, and they destroy the trust on which research depends. Institutional review, data protection and secure storage are procedural expressions of these duties, and the researcher duty to participants outranks the desire for a striking finding. A study that harms participants, misleads readers about its limits, or hides its methods has failed as research even if its conclusions are interesting, and examinations in this unit reward the candidate who can state that judgement plainly.

Scales of measurement

Designs and their strength