Research in Education

Overview

Educational research is systematic inquiry that aims to solve problems, build theory, and predict outcomes in education. Its scope covers learners, teachers, curriculum, institutions, and policy. The scientific method gives it discipline: problems are stated clearly, hypotheses are formulated, data are collected systematically, results are analysed, and conclusions are open to checking by others. Its characteristics are replicability, another researcher following the same procedures should obtain similar results; precision, concepts and measurements defined clearly enough to be checked; falsifiability, claims stated so that evidence could in principle refute them; and parsimony, preferring simpler explanations when they account for the data equally well. A claim that cannot be tested, cannot be replicated, or explains everything after the event is not scientific, however persuasive it sounds.

Scientific method and types of research

Types of scientific method describe the purpose of inquiry. Exploratory research investigates a little-known problem to clarify concepts and generate hypotheses. Descriptive research describes characteristics, attitudes, or conditions as they exist, answering what is the case. Explanatory research tests relationships and causes, answering why something happens. These are not ranks: good explanatory work often depends on careful descriptive and exploratory work before it.

Aims of research are problem-solving, improving a concrete educational situation; theory building, developing or refining explanations that generalise beyond one setting; and prediction, forecasting outcomes under stated conditions. Types of research by purpose include fundamental research, also called basic research, which advances knowledge without immediate application; applied research, which solves a practical problem using existing knowledge; and action research, in which practitioners study and improve their own practice in cycles of planning, acting, observing, and reflecting. Approaches divide broadly into quantitative research, which measures variables numerically and tests hypotheses statistically, and qualitative research, which explores meanings, experiences, and processes in depth, often in natural settings. Designs include descriptive designs, which describe and survey; experimental designs, which manipulate a variable and control others to test cause; and historical designs, which reconstruct past events from sources, judging authenticity and credibility of documents. The choice among them follows the question: describe, explain causally, understand meaning, or improve practice.

Variables, hypotheses, sampling and tools

A concept is an abstract idea, a construct is a concept developed for research with defined indicators, and a variable is a property that can take different values across persons or occasions. The independent variable is manipulated or presumed to cause change, the dependent variable is the outcome measured, extraneous variables are other influences that could affect the outcome and must be controlled, intervening variables lie between cause and effect in the causal chain, and moderator variables change the strength or direction of a relationship, for example when a method works for one age group but not another. Confusing independent and dependent variables is the commonest design error: the treatment is independent, the learning outcome is dependent.

A hypothesis is a tentative, testable statement of an expected relationship between variables. Sources include theory, previous research, observation, and practical experience. Types include research hypotheses, the substantive prediction; directional hypotheses, which predict the direction of difference or relationship; non-directional hypotheses, which predict a difference without direction; and null hypotheses, which state no difference or no relationship and are the hypotheses actually tested statistically. A good hypothesis is clear, specific, testable, stated in variables that can be measured, and consistent with existing knowledge. Research proposal steps include title, introduction and problem statement, objectives, review of related literature, hypotheses or research questions, method including design, population, sample, tools, and analysis plan, and a timeline and budget.

Universe, or population, is the whole group to which findings should generalise; a sample is the subgroup actually studied. A good sample is representative, adequate in size, and free from systematic bias. Probability sampling gives every member a known, non-zero chance of selection: simple random sampling, stratified sampling by subgroups, cluster sampling by groups, and systematic sampling at regular intervals. Non-probability sampling includes convenience sampling, purposive sampling of information-rich cases, and quota sampling. Probability sampling supports generalisation; non-probability sampling suits exploration and qualitative depth but limits generalisation. Tools must be judged by validity, whether they measure what they claim; reliability, whether they give consistent results; and standardisation, uniform procedures for administration and scoring. Common tools include rating scales, attitude scales, questionnaires, aptitude tests, achievement tests, and inventories, supplemented by observation, interviews, and projective techniques, in which ambiguous stimuli reveal attitudes and motives that direct questions may miss.

Measurement and quantitative data analysis

Measurement scales determine which statistics are legitimate. Nominal scales classify into categories without order, such as gender or subject. Ordinal scales rank without equal intervals, such as ranks in a class. Interval scales have equal intervals but no true zero, such as temperature in Celsius or many test scores treated as interval. Ratio scales have equal intervals and a true zero, such as height, weight, or time, allowing ratio statements. Using a mean for nominal data or claiming twice as much from an interval scale are category errors that examiners test directly.

Descriptive analysis summarises data: central tendency through mean, median, and mode; variability through range, quartile deviation, standard deviation, and variance; and graphical presentation through histograms, polygons, bar diagrams, and pie diagrams. Fiduciary limits, more commonly called confidence limits, give the range within which a population value probably lies. Hypothesis testing compares an observed result with what chance alone would produce. The null hypothesis is rejected when the result is unlikely under chance at a chosen significance level, commonly five percent. A Type I error rejects a true null hypothesis, a false positive; a Type II error retains a false null hypothesis, a false negative. Power is the probability of correctly rejecting a false null, and effect size measures the magnitude of a difference or relationship, distinct from its statistical significance: a tiny effect can be significant with a huge sample, and a large effect can be non-significant with a tiny sample.

Parametric techniques assume interval or ratio data, normal distribution, and homogeneity of variance; non-parametric techniques make fewer assumptions and suit ordinal data or violated assumptions. Correlation measures relationship strength and direction, with Pearson product-moment correlation for interval data and rank correlation for ordinal data. The t-test compares two means, the z-test compares a mean with a population value or large-sample means, ANOVA compares three or more means by analysing variance, and chi-square tests association between categorical variables or goodness of fit. Qualitative analysis works differently: data reduction selects and summarises material, classification organises it into categories, analytical induction tests emerging explanations against cases, constant comparison continually compares new data with existing categories, and triangulation checks findings across sources, methods, or investigators to strengthen credibility.

Qualitative and mixed-method designs

Grounded theory designs build theory from data rather than testing a pre-set theory: data are collected, coded in open, axial, and selective stages, categories are saturated when new data add nothing new, and a theory grounded in the data emerges. Its strength is fresh theory for little-understood processes; its weakness is time, skill, and the risk of forcing data into categories. Narrative research designs study lives and experiences through stories: participants tell accounts, researchers retell them with attention to time, place, and social context, and meaning emerges from the narrative structure. Its strength is depth of lived experience; its weakness is limited generalisation and reliance on memory and telling.

Case study designs investigate a bounded case, such as one school, classroom, or learner, in depth and in context, using multiple sources of evidence. Their strength is rich, contextual understanding and the ability to study complexity; their weakness is limited generalisation, though analytical generalisation to theory is possible. Ethnography studies a culture or community through prolonged immersion, participant observation, and interviews, describing shared meanings and practices from the insider perspective. Its strength is cultural depth; its weakness is time, access, and researcher influence on the setting. Mixed-method designs combine quantitative and qualitative approaches. Triangulation designs collect both kinds of data to corroborate findings. Explanatory designs collect quantitative data first, then qualitative data to explain results. Exploratory designs collect qualitative data first, then quantitative data to test or generalise emerging ideas. Each design has steps of framing questions, collecting and analysing each strand, integrating findings, and drawing meta-inferences that neither strand alone could support. The mixed promise is completeness; the mixed risk is a study that does two things poorly instead of one thing well.

Measurement scales and legitimate statistics

Test selection guide