Research in Library and Information Science

Overview

Research is systematic, planned enquiry aimed at producing trustworthy, shareable knowledge, as distinct from routine enquiry that solves an immediate local problem without generalisable design or documentation. Basic research seeks understanding of phenomena, such as how users search or how subjects grow. Applied research seeks solutions to practical problems, such as improving a catalogue or evaluating a literacy programme. Action research involves practitioners studying and improving their own practice in cycles of planning, acting, observing and reflecting. Evaluative research judges the merit, worth or effectiveness of a service or programme against stated criteria. Descriptive research portrays a situation as it is, while explanatory research tests why relationships occur, and exploratory research investigates a little-known area to generate questions and hypotheses for later study. A good library study states which type it is, because the type sets expectations: an exploratory interview study should not be judged as if it were a controlled experiment, and a descriptive survey should not claim causal effects it did not test.

Research quality rests on validity, reliability and ethics. Validity asks whether the study measures and concludes what it claims: whether instruments capture the intended concept, whether the sample represents the intended population, whether design rules out alternative explanations, and whether conclusions follow from evidence. Reliability asks whether measurement is consistent: whether the same instrument, observer or procedure would give similar results on repetition. Generalisability asks whether findings apply beyond the studied setting, and it depends on sampling, context and replication, not on the researcher confidence alone. Ethics requires informed consent, voluntary participation, confidentiality, data protection, avoidance of harm, and honest reporting, including limitations and negative findings. A study that is technically clever but ethically careless, or statistically impressive but invalid in design, fails as research even if its presentation is polished.

Research: concept, purpose and methods

The research process begins with a problem and a question. A good question is focused, researchable, and significant: it names a population or setting, a phenomenon or intervention, and an outcome or understanding sought, in terms that can be investigated with available time and resources. A literature review then situates the question in prior work, identifies concepts, theories and gaps, and prevents reinventing a study that has already answered the question or repeating its flaws. Concepts must be defined operationally: information literacy, user satisfaction, relevance and service quality mean different things in different studies, and a study must state how each concept is observed or measured, so that readers can judge whether the measure fits the concept. Hypotheses, where appropriate, state expected relationships in testable form, distinguishing the null position of no effect or no difference from the research position, and guiding analysis rather than being assumed true.

Methods are chosen for fit to the question. Quantitative methods measure variables numerically, use structured instruments and larger samples, and test relationships and differences statistically. They suit questions such as how often, how much, or whether an intervention changes an outcome. Qualitative methods explore meaning, experience and process through interviews, focus groups, observation and document analysis, with smaller purposive samples and rich textual data. They suit questions such as how users understand a service, why they behave as they do, or what barriers they experience. Mixed methods combine both, for example using a survey to map patterns and interviews to explain them, with integration at design, analysis or interpretation rather than merely collecting two kinds of data side by side. Historical, bibliometric, experimental, survey, case study, Delphi and content-analysis methods each have distinct logics: experiments manipulate a variable under control, surveys describe populations, case studies examine a bounded system in depth, Delphi builds expert consensus through rounds, and content analysis systematically codes texts. The error to avoid is method-first research, where a favoured tool dictates the question, instead of question-first research, where the question dictates the tool.

Research in library and information science has distinctive topics: user behaviour and information seeking, retrieval effectiveness, collection use, bibliometrics, service evaluation, literacy outcomes, scholarly communication, and management practice. Its evidence often comes from library systems, catalogues, usage logs, surveys and interviews, and its findings must be translated back into practice: a retrieval study should change search design or instruction, a user study should change hours, collections or help, and a bibliometric study should inform collection or evaluation policy with awareness of its limits. Research that ends in a report no practitioner reads has failed its applied purpose, however rigorous its method.

Research design and data collection

Research design is the overall plan linking question, data and conclusions. Experimental designs manipulate an independent variable and compare outcomes, with random assignment where feasible to balance other influences, and with control groups to isolate effect. Quasi-experimental designs compare groups without random assignment, as when two existing classes receive different instruction, and must address pre-existing differences. Survey designs collect standardised data from a sample to describe a population or test relationships, with cross-sectional surveys at one time and longitudinal surveys over time. Case-study designs examine one library, programme or community in depth, using multiple sources to build a rich, contextual account, with generalisation by analytic insight rather than statistical inference. Action-research designs cycle through planning, action, observation and reflection with practitioners as co-researchers, prioritising improvement in the studied setting.

Sampling decides whose data will speak for a larger population. Probability sampling gives every member a known chance of selection, through simple random, stratified, systematic or cluster methods, supporting statistical generalisation when samples are adequate and response is good. Non-probability sampling selects by purpose, convenience, quota or snowball referral, suiting qualitative depth or hard-to-reach groups, but limiting statistical generalisation. Sample size, response rate and non-response bias matter: a survey with a small or self-selected response may describe respondents well while misrepresenting the population, especially if non-respondents differ systematically, such as non-users who never receive a library survey. Variables must be distinguished: independent variables are manipulated or studied as influences, dependent variables are outcomes, and confounding variables are alternative influences that must be controlled, measured or acknowledged. Validity threats include selection bias, history, maturation, testing effects, instrument change, and attrition, and good design anticipates them rather than explaining them away after data are collected.

Data collection tools must be designed, piloted and revised. Questionnaires need clear, neutral, single-idea questions, appropriate response scales, logical order, and piloting with people like the intended respondents to catch ambiguity, leading wording, double-barrelled questions and missing options. Interviews range from structured, with fixed questions, through semi-structured, with guides and probes, to unstructured conversation, with choice depending on whether comparability or depth matters more. Observation may be structured with checklists or unstructured with field notes, and must address observer effects and ethical visibility. Document and log analysis uses existing records, such as circulation, search logs, citations and policies, with attention to what records omit and how they were created. Bibliometric data collection uses publication and citation databases with careful definition of scope, time, document types and counting rules, because different choices produce different rankings. Every tool needs a data management plan: how data are stored securely, anonymised where promised, backed up, and retained or shared under ethical and legal constraints.

Data analysis and statistical packages

Data analysis begins with cleaning and description. Quantitative data are checked for missing values, impossible values, duplicates and coding errors, then summarised with measures of central tendency and dispersion, frequency distributions, tables and graphs. The mean is the arithmetic average, sensitive to extreme values. The median is the middle value, robust when distributions are skewed, as with citation counts or loan counts where a few high values dominate. The mode is the most frequent value. Dispersion is shown by range, variance and standard deviation, which describe spread around the mean, and by quartiles and percentiles, which describe position in a distribution. Graphs must be honest: axes, scales and categories should not exaggerate differences or hide missing data, and a chart that looks dramatic because its axis is truncated misleads as surely as a wrong number.

Inferential analysis moves from sample to population with quantified uncertainty. It tests hypotheses about differences and relationships, reports test statistics and probability values, and increasingly emphasises effect sizes and confidence intervals, which show magnitude and precision rather than merely whether a result crosses a significance threshold. Correlation measures association between variables, but correlation alone does not prove causation: a third variable, reverse causation, or coincidence may explain the pattern. Common tests compare means, compare proportions, or test association in tables, and each test carries assumptions about data type, distribution and independence that must be checked, not assumed. Qualitative analysis codes textual data into themes, compares cases, seeks negative examples, and builds explanations grounded in evidence, with transparency about coding decisions and researcher reflexivity. Mixed analysis integrates strands, for example by explaining survey patterns with interview themes or by testing qualitative hypotheses quantitatively, and it must show where strands agree, disagree, and why disagreement is informative rather than embarrassing.

Statistical packages support this work by handling data management, computation, tables and graphs reproducibly, but they do not choose analyses or interpret results. A researcher must still decide the question, check assumptions, handle missing data, and report methods in enough detail for replication. Common errors in package-driven work include running many tests until one is significant, reporting only significant results, confusing statistical significance with practical importance, and presenting package output without understanding what was computed. Reproducibility practices — saved syntax or scripts, versioned datasets, clear variable labels, and documented cleaning decisions — protect against these errors and let other researchers verify findings. In library research, where samples are often small and populations are specific, cautious interpretation matters more than impressive software: a modest, well-described finding with stated limits contributes more than a sweeping claim built on a fragile analysis.

Report writing and metric studies

A research report communicates question, method, evidence and conclusions so that readers can judge trust and reuse findings. Its typical structure includes title and abstract, introduction and problem, literature review, objectives and hypotheses or questions, methodology, results, discussion, conclusions and recommendations, references, and appendices with instruments or supplementary data. The abstract must summarise purpose, method, key findings and implications accurately, without overstating. The methodology must be detailed enough for replication: population, sampling, instruments, procedures, analysis and ethical steps. Results present findings plainly, with tables and graphs that match the text and do not duplicate it unnecessarily. Discussion interprets results against prior literature, acknowledges limitations, considers alternative explanations, and states what should and should not be claimed. Conclusions follow from evidence, recommendations follow from conclusions, and references follow a consistent style, giving credit and allowing verification. Plagiarism, fabrication, falsification and duplicate publication violate research integrity, and authorship should reflect genuine contribution, not seniority or courtesy alone.

Metric studies measure scholarly communication through publication and citation data. Bibliometrics studies patterns in publications and citations, such as productivity, collaboration, growth, and distribution of literature across journals and authors. Scientometrics extends measurement to science as a system, including funding, institutions and policy, while informetrics covers information phenomena more broadly, and webometrics applies measurement to web links, downloads and online attention. Common measures include publication counts, citation counts, average citations per paper, journal impact indicators based on citations to recent articles, and author indicators that combine productivity and citation, such as the h-index, which identifies the largest number h such that an author has h papers each cited at least h times. These measures are useful for describing patterns, comparing like with like over defined periods and fields, and identifying highly cited work, but they are not direct measures of quality, importance or individual merit. Citation practices differ by field, document type, language and time, self-citation and citation gaming distort counts, and a single number hides the distribution behind it: an author with a few highly cited papers and many uncited ones differs from an author with consistent moderate citation, even if an indicator coincides.

Critical use of metrics therefore requires context: define the database, period, document types and counting rules; compare within fields and career stages; report distributions as well as averages; complement metrics with peer review and qualitative judgement; and avoid using journal indicators to judge individual articles or authors. Metric studies in libraries inform collection decisions, such as identifying core journals or highly used subjects, research evaluation support, and scholarly communication advice, but they must not become automatic ranking machines that reward gaming and penalise new, interdisciplinary or locally important work. A librarian who can explain what a metric counts, what it omits, and what decision it can and cannot support provides a more valuable service than one who merely reports a number.

Quantitative versus qualitative research compared

Metric measures and their limits