Communication Research II: Levels of Measurement, Sampling, Validity, Reliability and Scaling; Methods and Tools of Data Collection, including Media-Specific Methods; Data Analysis and Statistical Tests; and Research Ethics, Report Writing and Plagiarism

Validity is whether an instrument measures what it claims to measure. Face validity is whether it appears to; content validity whether it covers the full range of the concept; criterion validity whether it agrees with an external criterion, measured at the same time (concurrent) or later (predictive); and construct validity whether it behaves as theory predicts, correlating with related measures (convergent) and not with unrelated ones (discriminant). Reliability is consistency. Test-retest reliability compares scores over time; parallel-forms reliability compares two versions; split-half reliability compares two halves of a test, corrected with the Spearman–Brown formula; internal consistency is measured by Cronbach's alpha (1951), with about 0.7 or above usually taken as acceptable. In content analysis, inter-coder reliability measures agreement between coders: Holsti's formula is 2M ÷ (N1 + N2), where M is the number of agreements and N1 and N2 the decisions of each coder, so two coders who each code 50 items and agree on 45 reach 2 × 45 ÷ 100 = 0.9. Scott's pi, Cohen's kappa and Krippendorff's alpha correct for chance agreement. A measure can be reliable without being valid, but not valid without being reliable.

Interviews may be structured (fixed questions in a fixed order), semi-structured (a guide with room to probe) or unstructured and in-depth; focus group discussions bring together about six to twelve participants under a moderator. Surveys collect standardised data from a sample face to face, by post, by telephone (computer-assisted telephone interviewing), or online. The questionnaire is filled in by the respondent; the schedule is a set of questions asked and filled in by an enumerator, which suits respondents who cannot read and ensures that questions are understood. Case studies examine one unit in depth. Ethnography studies a group in its natural setting through participant observation over time, and its online form, netnography, studies digital communities. A diary asks respondents to record their media use or activities as they happen, reducing reliance on memory. Internet-based tools include online questionnaires, web and social-media analytics, and the collection of posts for analysis.

Methods are obtrusive when the people studied know they are being studied, which can change their behaviour (the reactive effect), and unobtrusive when they do not; Eugene Webb and colleagues' Unobtrusive Measures (1966) collected such methods. Content analysis, defined by Bernard Berelson (1952) as a research technique for the objective, systematic and quantitative description of the manifest content of communication, is unobtrusive, since it studies texts already produced. Media research has its own methods. An exit poll surveys voters as they leave polling stations; under section 126A of the Representation of the People Act 1951, inserted in 2009, conducting and publishing exit polls is barred during the period the Election Commission notifies, from the start of polling to half an hour after the close of the last phase. Opinion polls survey intentions before voting, usually with quota or stratified samples, and their accuracy depends on sampling and on converting vote shares into seats. Telephone and SMS surveys are fast and cheap but reach only those with phones who choose to respond. Television audiences are measured by BARC's people meters, which detect audio watermarks embedded in channels' broadcasts. Audience "voting" by SMS or online on general entertainment content, such as reality shows, is a self-selected, repeatable vote, useful for engagement but not a representative measure of opinion.

Descriptive statistics summarise data: measures of central tendency (the mean, median and mode) and of dispersion (the range, variance and standard deviation). For the viewing hours 2, 4, 4, 4, 5, 5, 7 and 9, the mean is 40 ÷ 8 = 5; the squared deviations from the mean are 9, 1, 1, 1, 0, 0, 4 and 16, which sum to 32; the variance of this set is 32 ÷ 8 = 4 and the standard deviation √4 = 2. Inferential statistics generalise from a sample to a population. Analysis is univariate when it examines one variable (a frequency table or a mean), bivariate when it examines the relation between two (a cross-tabulation, a correlation or a t-test), and multivariate when it examines three or more at once (multiple regression, factor analysis, cluster analysis, MANOVA).

Parametric tests assume interval or ratio data, an approximately normal distribution and similar variances; non-parametric tests make fewer assumptions and suit nominal or ordinal data or small, skewed samples. For a chi-square test on a table with r rows and c columns, the degrees of freedom are (r − 1)(c − 1), so a 3 × 4 table has 2 × 3 = 6. Spearman's rho is 1 − 6Σd² ÷ n(n² − 1), where d is the difference between paired ranks: for 5 items with Σd² = 4, rho = 1 − 24 ÷ 120 = 0.8. A test of significance compares the probability of the observed result under the null hypothesis, the p-value, with a chosen significance level α, conventionally 0.05. A Type I error rejects a true null hypothesis, and its probability is α; a Type II error fails to reject a false one; lowering α reduces Type I errors but increases Type II errors unless the sample grows. Computer-mediated research relies on software such as SPSS (first released in 1968) and R for statistics, NVivo and ATLAS.ti for qualitative coding, and computational methods for analysing large volumes of digital text.

Ethical research rests on principles that the Belmont Report (1979) summarised as respect for persons, beneficence and justice. In practice this means informed and voluntary consent; confidentiality and, where possible, anonymity; avoiding harm and unnecessary deception, with debriefing where deception is used; special protection for children and vulnerable groups; honesty in collecting, analysing and reporting data, without fabrication or falsification; and approval by an institutional ethics committee. Media research adds its own questions: whether public social-media posts may be used without consent, how to protect sources and participants in studies of journalists, and how to handle data that could identify people.

A research report usually follows the IMRaD structure, introduction, methods, results and discussion, preceded by a title and abstract and followed by conclusions, references and appendices; a thesis adds a review of literature and chapters. Citation follows a style manual, most often APA (seventh edition, 2019) in communication research, or MLA or Chicago. Plagiarism is presenting another's words, ideas or data as one's own, including self-plagiarism of one's earlier work. The University Grants Commission (Promotion of Academic Integrity and Prevention of Plagiarism in Higher Educational Institutions) Regulations 2018 grade similarity in research work: up to 10 per cent is Level 0 (minor, no penalty); above 10 and up to 40 per cent is Level 1; above 40 and up to 60 per cent Level 2; and above 60 per cent Level 3, with penalties rising by level. Quoted material with citation, references and common knowledge are excluded. Similarity-checking tools include Turnitin, iThenticate and, for Indian universities, ShodhShuddhi, provided through INFLIBNET.