4 Principles of Research Design

Research questions and design

A research design is the plan linking a research problem to the evidence needed to answer it. It identifies what will be studied, which people or cases will be included, how information will be collected and measured, and how findings will be interpreted. A coherent design starts with a focused research question, rather than a preferred statistical test or data-collection tool.

The kind of answer sought shapes the design:

• Descriptive questions ask what is happening or how common something is.

• Associational questions ask whether two or more characteristics vary together.

• Causal questions ask whether a change or exposure produces an outcome. Appropriately designed experiments can support causal conclusions, while observational associations alone do not establish causation.

• Exploratory or interpretive questions ask how people understand, experience, or explain a phenomenon.

A design may also be cross-sectional, collecting information at one time, or longitudinal, collecting information at multiple times. These choices shape which conclusions the evidence can support.

Concepts, constructs, and variables

A concept is a broad idea of interest, such as stress, belonging, or academic achievement. A construct is a concept made more precise for a study. Researchers first conceptualize a construct by stating what it means in their study, then operationalize it by specifying how it will be observed or measured.

For example, a study might define “sleep” as nightly duration and measure it using a participant diary or a wearable device. The selected measure affects what the findings mean.

A variable is a characteristic that can differ across people, groups, events, or time. Common roles include:

• Independent variable: a proposed explanatory factor or, in an experiment, the factor deliberately varied by the researcher.

• Dependent variable: the outcome the study examines.

• Control variable: a factor accounted for to clarify the relationship of interest.

• Confounder: a factor associated with both the explanatory factor and the outcome that may distort their apparent relationship.

For a study of whether a tutoring program improves test scores, program participation might be the independent variable and scores the dependent variable. Prior achievement could also matter, so researchers might measure it and account for it in the design or analysis. In a nonexperimental study, researchers observe naturally occurring differences rather than assigning participants to conditions; causal claims therefore require particular caution.

Forms of variables

Variables can be represented in different forms. Categorical measures identify groups, such as school type. Ordinal measures have an order, such as a rating from “low” to “high.” Numerical measures represent quantities, such as hours studied or number of absences. A measure should preserve the distinctions that matter to the research question.

Sampling and the people studied

The target population is the larger group to which a study hopes its conclusions will apply. The sample is the subset actually studied, and the sampling frame is the practical list or source from which that sample is selected. A mismatch between the target population and the people who can be reached can limit how broadly results apply.

In probability sampling, selection uses a random process, giving members of a defined population a known chance of inclusion. Simple random, stratified, and cluster sampling are examples. When well executed, probability sampling can support estimates about the population, although nonresponse and other sources of bias still matter.

In nonprobability sampling, selection does not use random selection. Convenience sampling recruits accessible participants; purposive sampling selects people with experience relevant to the question. These approaches can be useful for exploratory or in-depth studies, but they do not by themselves justify statistical generalization to a larger population. Qualitative studies often select participants purposefully to obtain detailed, relevant accounts rather than to estimate population percentages.

Sample size does not substitute for sound selection: a very large, biased sample may still misrepresent the population. Quantitative studies should justify sample size in relation to the precision or effect they seek to estimate. Qualitative studies should explain how participant selection and depth of information serve the inquiry.

Measurement and evidence quality

Measurement is the systematic process of representing an attribute through observations, questions, records, or instruments. A measure should be appropriate to the construct and applied consistently.

Reliability concerns consistency. For example, if two trained observers code the same behavior, similar ratings indicate stronger inter-rater reliability. Validity concerns whether the measure or procedure supports the intended interpretation. A measure can be consistent yet measure the wrong thing, so reliability alone does not establish validity.

Measurement error is the difference between the recorded value and the attribute the researcher intends to capture. Unclear questions, inconsistent procedures, faulty instruments, or observer expectations can introduce error or bias.

In quantitative research, researchers often specify indicators and scoring rules before collecting data. In qualitative research, concepts may be refined as interviews, observations, or documents are examined. Rather than forcing experiences into numerical variables, qualitative rigor depends on transparent procedures, careful interpretation, and a clear account of how evidence supports conclusions.

Aligning methods and claims

A study is strongest when its question, sample, measures, data collection, and analysis fit together. For the question, “How do first-year students describe the challenges of adjusting to college?”, semi-structured interviews with purposefully selected first-year students are well aligned because they invite detailed accounts. The findings can describe participants’ experiences, but a small purposive sample cannot establish how common each experience is among all students.

For the question, “What proportion of first-year students report difficulty accessing academic support?”, a clearly defined student population and a survey with an appropriate sampling strategy better fit the goal of estimating prevalence. If the question asks whether a particular support program causes improved outcomes, the design must address comparison groups, timing, and plausible alternative explanations. A simple one-time survey would not establish causation.

Before collecting data, researchers can check alignment by asking:

1. Does the question specify the phenomenon, population, and relevant timeframe?

2. Do the concepts have clear definitions and suitable indicators?

3. Can the sampling approach reach the people or cases needed to answer the question?

4. Will the chosen methods produce the type of evidence the question requires?

5. Do the planned conclusions stay within what the design can support?

Ethical feasibility is part of good design. Participant burden, privacy, informed consent, and potential harms should be considered before data collection. A method that could answer a question in theory may not be appropriate if it cannot be conducted responsibly.

What makes a design defensible

Research design links a question to a defensible plan for gathering and interpreting evidence. Researchers define concepts, operationalize constructs, identify variables, choose a sample, and select measures and methods suited to the conclusions they intend to draw.

Probability sampling can support population estimates when the sample and implementation are appropriate. Purposive and other nonprobability approaches may better serve in-depth inquiry but limit statistical generalization. Reliable measurement is consistent, and valid measurement supports the intended interpretation; neither removes the need to consider bias. Ultimately, the strength of a conclusion depends on the fit among the question, sample, methods, measures, and claims.