Study Design
Study design is the architecture that determines what a piece of research can and cannot claim: randomized controlled trials (RCTs) support causal claims through random assignment and controlled comparison; cohort studies follow exposed and unexposed groups forward in time; case-control studies work backward from outcome to exposure; cross-sectional studies capture a single point in time; and quasi-experimental and factorial designs sit between fully controlled experiments and pure observation. This sub-cluster explains the mechanics, strengths, and threats to validity of each design, and how to choose the right one for a given research question — a decision that shapes everything downstream, from IRB review to statistical analysis to how reviewers will judge the resulting manuscript.
Guides
Interim Analyses in Clinical Trials: Why Pre-Specification Comes First
What counts as an interim analysis, why an unplanned look inflates Type I error even without intent to influence the trial, and what has to be pre-specified before the first look — the conceptual foundation beneath alpha-spending and group sequential design.
Superiority, Non-Inferiority and Equivalence Trials
Superiority, non-inferiority and equivalence trials ask three different questions and use three different margin structures. This guide distinguishes the three by hypothesis and decision rule, then walks through a fully computed, reproducible worked example of setting a defensible non-inferiority margin using the fixed-margin (preservation-of-effect) method.
Group Sequential Designs: Alpha-Spending Functions and Stopping Boundaries
How alpha-spending functions let a trial look at interim data multiple times without inflating overall Type I error, with a reproducible O’Brien-Fleming vs Pocock boundary comparison.
Test-Negative Design: The Logic Behind Vaccine-Effectiveness Studies
The test-negative design compares vaccination status between test-positive cases and test-negative controls who share the same symptoms and care-seeking path, controlling for healthcare-seeking-behavior confounding that a standard case-control design does not.
Futility Analysis and Conditional Power: When to Stop a Trial for Futility
How conditional power is calculated at an interim look, why the assumed treatment effect changes the answer, and why non-binding futility rules cannot be used to “save” alpha for the efficacy boundaries.
Ecological Study Design and the Ecological Fallacy
Ecological study design measures variables at the group level, which makes it fast and cheap for hypothesis generation — but a group-level association is not the same statistic as an individual-level one. A worked numeric example shows the correlation sign reverse between the two levels.
Factorial Designs: Testing Multiple Factors and Their Interactions at Once
A full factorial design tests every combination of two or more factors at once instead of one factor at a time — more efficient, and the only way to detect whether one factor’s effect depends on another. Worked through with a computed 2×2 layout separating a main effect from an interaction effect.
Crossover Trial Design and Carryover
How crossover trials work, why the carryover effect threatens the comparison, how washout periods are justified, and when an irreversible outcome or long-acting treatment requires a parallel design instead.
ROBINS-I for Non-Randomised Studies: The Seven Domains, Applied to One Study
ROBINS-I’s seven bias domains, walked through practically for a single non-randomized study of an intervention: the target-trial framing the tool assumes, what evidence each domain actually requires, and a worked E-value example for quantifying confounding sensitivity.
Case-Crossover and Self-Controlled Case Series: Designs That Control Confounding by Design
How case-crossover and self-controlled case series (SCCS) use within-person comparison to eliminate time-invariant confounding by design, and how to tell which one fits a transient exposure versus a recurrent event.
Case-Cohort Studies: Design, Subcohort Reuse, and Efficiency
A case-cohort study draws one random subcohort once, at baseline, and reuses it as the comparison group for every outcome studied from that cohort — a different sampling mechanic from nested case-control’s per-case risk-set sampling, with a specific efficiency payoff for multi-endpoint cohorts.
Split-Plot Designs: Whole-Plot and Sub-Plot Error Terms
A split-plot design nests an easy-to-change sub-plot factor inside a hard-to-change whole-plot factor, which creates two different randomization units and therefore two different error terms. This guide works through the layout, the correct two-error-term ANOVA, and exactly what happens to the significance tests when the two error terms get pooled into one.
Interrupted Time Series Analysis: Segmented Regression, Autocorrelation, and Lag
Interrupted time series analysis uses segmented regression to separate an immediate level change from a slope change at the intervention point. This guide covers the model equation, why time-ordered residuals violate the OLS independence assumption, Newey-West/Prais-Winsten/ARIMA corrections, and how to choose the autoregressive lag order.
Adaptive Trial Designs: Types, Alpha-Spending & Pre-Specification
How group sequential design, sample-size re-estimation, and response-adaptive randomization work, and the alpha-spending and pre-specification requirements that keep each statistically valid rather than exploitable.
Stratified Randomisation and Minimisation
Compares stratified randomisation (permuted blocks within strata) and minimisation (a dynamic algorithm balancing multiple prognostic factors) for baseline balance in trials, including the stratum-count limit, when minimisation scales better, and the requirement to adjust the analysis for whichever factors were used.
Latin Square Designs: Controlling Two Blocking Factors at Once
A Latin square design blocks on two nuisance variables at once (rows and columns) instead of one, using an n×n grid where every treatment appears exactly once per row and once per column.
Nested Case-Control Studies: Risk-Set Sampling and Density Sampling
How risk-set (incidence-density) sampling draws matched controls from a cohort’s risk set at each case’s failure time, and why the resulting odds ratio estimates the incidence rate ratio — with a worked matched-set example.
Natural Experiments: How to Argue a Shock Is Genuinely Exogenous
What makes a natural experiment’s shock genuinely exogenous rather than correlated with unobserved confounders, the pre-trend and placebo checks that establish credibility, and named real-world examples.
Regression Discontinuity Design: Bandwidth, Running Variable, and the McCrary Test
A practical guide to regression discontinuity design: sharp vs. fuzzy RD, what makes a running variable valid, the bias-variance trade-off in bandwidth choice, the McCrary density test for manipulation, and the placebo/robustness checks reviewers expect.
Cohort Study vs Case-Control Study: Research Guide
How cohort and case-control designs differ for epidemiology and clinical research: which one starts from exposure and which from outcome, when relative risk or an odds ratio is the right measure, and how each handles rare disease, confounding and STROBE reporting.
What Is a Control Group? Types, Ethics, and How to Choose One
A control group supplies the counterfactual that lets researchers attribute an observed change to an intervention rather than to time, expectation, or chance. This guide covers the main types of control, the ethics of choosing one under the Declaration of Helsinki, randomization and blinding, control groups outside clinical trials, and what to do when a true control group isn’t possible.
