Quantitative research designs and trials: a glossary

Active control

Also called: active comparator, active control group, active comparator arm, active-controlled

An active control is a control group that receives an established, effective treatment instead of a placebo or nothing, so that the new intervention is compared with what patients would otherwise be given. It answers the question practitioners actually face, which option to choose, and it is the usual comparator in non-inferiority and equivalence trials.

Adaptive design

Also called: adaptive trial, adaptive trial design, adaptive clinical trial, adaptive randomised trial, adaptive randomized trial

An adaptive design is a trial design that plans in advance how results gathered during the trial may be used to change it while it runs, without undermining the validity of its conclusions. Permitted changes include stopping early for benefit or futility, re-estimating the sample size, dropping arms and shifting the allocation ratio, and each must be set out before interim data are seen.

Allocation concealment

Also called: concealment of allocation, allocation sequence concealment, concealed allocation

Allocation concealment is the set of safeguards that stops the people enrolling participants in a randomised trial from knowing which group the next person will join, so they cannot steer particular people into a group. It differs from blinding, which hides the group after assignment. Concealment can always be achieved, while blinding sometimes cannot.

Allocation ratio

Also called: randomisation ratio, randomization ratio, treatment allocation ratio

The allocation ratio is the planned proportion of participants assigned to each group in a trial, such as 1:1 for equal groups or 2:1 for twice as many in the new treatment group. Unequal ratios are sometimes chosen to gain more experience with a new treatment or to limit costs, and reporting guidelines ask trials to state the ratio used.

Allocation sequence

Also called: randomisation sequence, randomization sequence, random allocation sequence, randomisation schedule, randomization schedule, randomisation list, randomization list

An allocation sequence is the ordered list, usually generated by computer, that decides which group each successive participant in a trial will join. A good sequence is unpredictable, and it must be concealed from the people enrolling participants until each assignment is made. How it was generated and how it was concealed are both examined when a trial's risk of bias is assessed.

Alternating treatments design

Also called: alternating-treatments design, alternating treatment design, ATD

An alternating treatments design is a single-case experimental design in which two or more treatments are switched rapidly on a regular schedule, such as one on alternate days or one in the morning and another in the afternoon, and the behaviour under each is compared. It compares treatments quickly, but only when their effects appear quickly.

Analytical study

Also called: analytic study, analytical design, analytic epidemiological study

An analytical study is a study designed to test whether an exposure is linked to an outcome, which it does by comparing groups, such as exposed with unexposed people or cases with controls. Cohort, case-control and experimental studies are analytical. A descriptive study, by contrast, describes who is affected, where and when, and suggests hypotheses for analytical studies to test.

As-treated analysis

Also called: as treated analysis, analysis as treated

An as-treated analysis is an analysis of a randomised trial in which participants are grouped by the intervention they actually received, not the one they were randomised to. Because the reasons people switch or stop treatment are often linked to how well they are, it gives up the protection randomisation provides, and Cochrane describes it as liable to serious bias.

Basket trial

Also called: basket study, basket design, basket trial design

A basket trial is a trial, run under one master protocol, that tests a single targeted treatment in patients with several different diseases who share the same biological feature, such as a particular genetic mutation found in cancers of different organs. A 2019 review found basket, umbrella and platform trials described in many inconsistent ways, so definitions vary between papers.

Between-subjects design

Also called: between-subject design, between subjects design, between-groups design, between-participants design, independent groups design, independent measures design

A between-subjects design is an experimental design in which each participant takes part in only one condition, so the comparison is between different groups of people. It avoids the practice and fatigue effects of repeated testing, but it needs more participants than a within-subjects design, and random assignment is what makes the groups comparable.

Birth cohort study

Also called: birth cohort, birth-cohort study

A birth cohort study is a cohort study that follows people born in the same period, sometimes the same week, from birth or childhood through their lives, collecting information on health, education, work and circumstances at intervals. The 1958 National Child Development Study and the 1970 British Cohort Study are British examples. STROBE gives birth cohorts as an example of closed cohorts.

Blinded outcome assessment

Also called: outcome assessor blinding, blinding of outcome assessors, assessor blinding, blinded assessment, blinded outcome assessor, masked outcome assessment

Blinded outcome assessment is the practice of keeping the people who measure or judge a trial's outcomes unaware of which group each participant was in. It is often possible even when participants and the people treating them cannot be blinded, and it matters most when judging the outcome involves opinion, such as a clinical examination, rather than a hard measure such as death.

Blinding

Also called: masking, blind, blinded, blinding in research

Blinding is the practice of keeping participants, the people giving care, outcome assessors or analysts unaware of which group each participant is in, so that the knowledge cannot change behaviour or judgement. Masking means the same and is preferred in some fields. Because labels such as double-blind are read in different ways, reporting guidelines ask authors to say exactly who was blinded.

Block randomisation

Also called: block randomization, blocked randomisation, blocked randomization, permuted block randomisation, permuted block randomization, random permuted blocks

Block randomisation is a restricted form of randomisation in which participants are allocated in small blocks, each holding every group in the planned ratio in a random order, so that group sizes stay close throughout recruitment. If the block size is fixed and known, the last allocation in a block can be guessed, so trials often vary the size.

Carryover effect

Also called: carry-over effect, carryover, carry-over, residual effect

A carryover effect is an effect of an earlier condition or treatment that persists into a later one in a within-subjects experiment or crossover trial, so that results in the later period partly reflect what came before. In psychology, practice and fatigue effects are kinds of carryover. In trials a washout period between treatments is used to reduce it.

Case report

Also called: clinical case report, single case report, single-patient case report

A case report is a detailed account of a single patient or participant, usually written because the condition, the response to treatment or a side effect was unusual. With no comparison group it cannot show cause, but it can raise a question for a stronger design to test. It is not the same as a case study in qualitative research.

Case series

Also called: case-series, clinical series, case series study

A case series is a description of several people who share a condition or a treatment, reporting what happened to them without a comparison group. It is often the first published signal of a new disease or harm, as with the early reports of AIDS, but without controls it cannot show that an exposure or treatment caused the outcome.

Case-cohort design

Also called: case-cohort study, case cohort study, case-cohort sampling

A case-cohort design is a design that studies, within a cohort, every member who develops the outcome together with a random subcohort chosen at the start regardless of what later happens to them. Unlike a nested case-control study, which picks controls when each case occurs, one subcohort can serve as the comparison group for several outcomes, and costly measurements are needed only for the sample.

Case-control study

Also called: case control study, case-control design, case-referent study, case-comparison study

A case-control study is an observational study that starts with people who have an outcome, the cases, and people who do not, the controls, and compares how often each group was exposed to possible causes in the past. It is efficient for rare diseases and long delays between exposure and outcome, but recall and the choice of controls can bias it.

Case-crossover design

Also called: case-crossover study, case crossover design, case-crossover

A case-crossover design is a variant of the case-control study in which each case serves as their own control, comparing their exposure just before the outcome with their exposure during another, earlier period. It suits brief exposures thought to raise risk for a short time, such as using a phone while driving, and it removes the influence of characteristics that do not change.

Clinical trial

Also called: medical trial, clinical research trial

A clinical trial is a study in which researchers assign people to receive a treatment, test, device or other health intervention so that its safety and effects can be measured. Not every clinical trial is randomised or has a control group, and trials of new medicines usually pass through phases I to IV, before and after approval.

Closed cohort

Also called: closed cohort study

A closed cohort is a cohort whose membership is fixed at the start, a defined group enrolled at one time and followed forward, often at set intervals up to a fixed end date. A birth cohort is the classic example. Because everyone starts together, both risks over a period and rates can be estimated, while an open cohort yields rates only.

Cluster

A cluster is a naturally occurring group of people, such as a school, a ward, a general practice, a household or a village, that is allocated or sampled as a whole instead of person by person. In a cluster randomised trial the cluster is the unit of randomisation, and because its members tend to resemble one another, the analysis must allow for that similarity.

Cluster crossover trial

Also called: cluster crossover design, cluster-crossover trial, cluster randomised crossover trial, cluster randomized crossover trial

A cluster crossover trial is a trial in which whole clusters, such as hospital wards, are randomised to a sequence of interventions and switch from one to another over successive periods, so that each cluster receives every intervention. It combines the features of cluster and crossover designs, and its analysis must allow both for similarity within clusters and for the pairing of periods.

Cluster randomised trial

Also called: cluster randomized trial, cluster-randomised trial, cluster-randomized trial, cluster randomised controlled trial, cluster randomized controlled trial, cluster RCT, cRCT, group randomised trial, group-randomized trial

A cluster randomised trial is a randomised trial in which whole groups, such as schools, wards, general practices or villages, are allocated to the interventions instead of individual people. It is used when an intervention is delivered to a group or would spread between people in one setting. Because members of a cluster tend to respond alike, the analysis must allow for clustering.

Cohort study

Also called: cohort design, follow-up study, cohort research

A cohort study is an observational study that identifies a group of people, records who is and is not exposed to a factor of interest, and follows them over time to compare how often an outcome occurs in each group. Because exposure is recorded before the outcome, it shows the order of events, but following a large cohort for years is costly.

Community trial

Also called: community intervention trial, community-level trial

A community trial is an experimental study in which whole communities, such as towns, villages or neighbourhoods, are assigned by the investigator to receive an intervention or not and are then followed to measure its effects. It suits interventions delivered to a population, such as a public health campaign. When the communities are assigned at random, it is a kind of cluster randomised trial.

Completely randomised design

Also called: completely randomized design, CRD, completely randomized experiment

A completely randomised design is an experimental design in which the treatments are assigned to the experimental units, or the runs are ordered, entirely at random, with no blocking. It is the simplest design for comparing the levels of one factor when no nuisance factors need controlling, and it is the design-of-experiments counterpart of a simply randomised parallel-group trial.

Contamination

Also called: treatment contamination, intervention contamination, contamination bias, spillover

Contamination is the unintended exposure of people in a trial's control group to the intervention, or part of it, for example when staff trained in a new method use it with every patient on a ward. It shrinks the difference between the groups and so can hide a real effect. Allocating whole clusters instead of individuals is a common way to prevent it.

Context effect

Also called: contrast effect

A context effect is an order effect in a within-subjects experiment in which taking part in one condition changes how participants perceive or interpret the next. For example, a moderately loud tone may seem quieter to someone who has just heard a very loud one. Like other order effects, it is handled by counterbalancing the order of conditions.

Control group

Also called: comparison group, control arm, comparator group, comparator arm, control condition

A control group is the group in a study that does not receive the intervention being tested, and whose results show what would probably have happened without it. It may receive nothing, usual care, an established treatment, a placebo or a sham. The names control group and comparison group are generally used interchangeably, and ideally the control group resembles the intervention group in everything but the intervention.

Controlled before-and-after study

Also called: controlled before-after study, controlled before and after study, CBA study, controlled pre-post study, pre-test post-test non-equivalent groups design, pretest-posttest nonequivalent groups design

A controlled before-and-after study is a study that measures an outcome before and after an intervention, both in a group that receives it and in a comparison group that does not, where the researchers did not decide who received it. Comparing the change in each group, often by difference-in-differences, beats a single group's before and after, but unmeasured differences between the groups remain a risk.

Correlational research

Also called: correlational study, correlational design, correlational method

Correlational research is a non-experimental design in which the researcher measures two or more variables as they naturally occur and assesses how strongly they are related, without manipulating any of them. It suits questions where manipulation is impossible or unethical, but a correlation alone cannot show which variable affects the other or rule out a third variable behind both.

Counterbalancing

Also called: counter-balancing, counterbalanced design, counterbalanced order

Counterbalancing is the technique of presenting the conditions of a within-subjects experiment in different orders to different participants, so that order effects such as practice, fatigue or carryover are spread evenly across the conditions. Complete counterbalancing uses every possible order, while a Latin square uses a smaller set in which each condition appears once in each position.

Cross-sectional study

Also called: cross sectional study, cross-sectional survey, prevalence study, prevalence survey

A cross-sectional study is an observational study that measures exposures and outcomes in a sample of people at a single point in time, like a snapshot. In health research it is the usual way to estimate prevalence, how common a condition or behaviour is, but because both are measured at once it usually cannot show whether the exposure came first.

Crossover trial

Also called: cross-over trial, crossover design, cross-over design, crossover study, cross-over study, crossover randomised trial

A crossover trial is a trial in which each participant receives two or more interventions one after another, in an order decided at random, so that each person acts as their own control. It needs fewer participants than a parallel-group trial but suits only stable conditions and treatments whose effects wear off, often with a washout period between them to limit carryover.

Difference-in-differences

Also called: difference in differences, difference-in-difference, DiD, diff-in-diff, double difference

Difference-in-differences is a quasi-experimental method that estimates an intervention's effect by comparing the change in an outcome over time in a group exposed to it with the change over the same time in a group that was not. It removes stable differences between the groups and trends they share, but it assumes both groups would have changed alike without the intervention.

Double-blind

Also called: double blind, double-blinded, double-masked, double-blind study, double-blind trial, double-blind randomised controlled trial

A double-blind study is a study in which neither the participants nor the researchers or clinicians dealing with them know who is receiving which intervention. The label is used loosely, and trials described as double-blind have been found to blind many different combinations of people, so a clear report names each group that was kept unaware.

Ecological study

Also called: ecologic study, ecological design, ecological analysis

An ecological study is an observational study in which the unit of analysis is a group, such as a region, a school or a country, comparing group averages or rates instead of data on individuals. It is cheap and useful for comparing places, but a link seen between groups need not hold for the people within them, the error known as the ecological fallacy.

Effectiveness

Also called: clinical effectiveness, real-world effectiveness, treatment effectiveness

Effectiveness is how well an intervention works under the usual conditions of everyday practice, with typical patients, practitioners and settings, compared with doing nothing or with another option. It answers the question does it work, and it is what pragmatic trials measure. It differs from efficacy, which is how well the intervention works under ideal conditions.

Efficacy

Also called: treatment efficacy

Efficacy is how well an intervention works under ideal, tightly controlled conditions, such as in an explanatory trial with carefully chosen and closely supervised participants. It answers the question can it work. It differs from effectiveness, which is how well the intervention works in everyday practice, and a treatment with good efficacy may prove less effective in routine use.

Enrichment design

Also called: adaptive enrichment, population enrichment, enrichment strategy, adaptive enrichment design

An enrichment design is a trial design that concentrates recruitment on the patients most likely to benefit, for example those with a particular biomarker, either from the start or, in an adaptive version, after an interim analysis shows who is responding. It can make a trial smaller and better able to detect an effect, but its results apply only to the kind of patients it enrolled.

Equivalence margin

Also called: margin of equivalence, equivalence limits, equivalence bounds, equivalence range

An equivalence margin is the largest difference between two interventions, in either direction, that would be accepted as clinically unimportant, fixed before an equivalence trial starts. The trial concludes equivalence only if the confidence interval for the difference lies wholly inside the margin on both sides. Setting it is a clinical judgement as much as a statistical one.

Equivalence trial

Also called: equivalence study, equivalence design, equivalence randomised trial

An equivalence trial is a randomised trial designed to show that a new intervention is neither meaningfully better nor meaningfully worse than a comparator, within a margin set in advance. It is used, for example, to compare a generic medicine with the original. It differs from a non-inferiority trial, which asks only whether the new intervention is not worse.

Experimental study

Also called: experiment, experimental research, interventional study, intervention study

An experimental study is a study in which the researcher decides which participants receive which intervention or condition and then measures the effect. Assignment may be random, as in a true experiment or a randomised trial, or not, as in a non-randomised trial. An observational study differs because the researcher only records exposures that people already have.

Explanatory trial

Also called: explanatory randomised trial, explanatory randomized trial, explanatory clinical trial, efficacy trial

An explanatory trial is a randomised trial designed to test whether an intervention can work under ideal conditions, with carefully selected participants, expert delivery, close monitoring and often a placebo comparator. It answers a question about efficacy. Explanatory and pragmatic are best seen as the two ends of a continuum, and most trials sit somewhere between them.

External control

Also called: external control group, external comparator, external control arm

An external control is a comparison group made up of people who were not in the same randomised study as those receiving the intervention, drawn for example from an earlier trial, a disease registry or health records. It is used mainly when randomising is not feasible, as in some rare diseases, but differences between the groups make its results more speculative than a randomised comparison.

Factorial design

Also called: factorial trial, factorial experiment, 2x2 factorial design, two-by-two factorial design, factorial randomised trial, factorial randomized trial

A factorial design is a design that tests two or more interventions or independent variables at once by combining their levels, so that every combination forms a group. In a two-by-two trial of treatments A and B, participants receive A only, B only, both or neither. One sample answers several questions and shows whether the factors interact.

Fatigue effect

Also called: boredom effect

A fatigue effect is an order effect in which participants in a within-subjects experiment perform worse in later conditions because they have become tired or bored, not because of the condition itself. It is the opposite of a practice effect, and researchers spread its influence evenly across conditions by counterbalancing their order.

Field experiment

Also called: field trial, field-based experiment, experiment in the field

A field experiment is an experiment carried out in a real-world setting, such as a school, a workplace, a hotel or a website, in which the researcher still manipulates a variable and usually assigns participants at random. Its results are closer to everyday life than a laboratory experiment's, at the cost of less control over other influences.

Fractional factorial design

Also called: fractional factorial, fractional factorial experiment

A fractional factorial design is a factorial experiment that runs only a carefully chosen fraction, such as a half or a quarter, of all the combinations of factor levels that a full factorial design would require. It makes experiments with many factors affordable, since a full two-level design with six factors already needs 64 runs, and a well-chosen fraction remains balanced. It is common in engineering and quality improvement.

Group sequential design

Also called: group sequential trial, group-sequential design, sequential trial design, GSD

A group sequential design is an adaptive trial design in which the accumulating results are examined at a small number of planned interim analyses, each with stopping rules fixed in advance, so that the trial can end early if there is clear evidence of benefit, of harm or of futility. It lets a trial stop as soon as its question is answered.

Hierarchy of evidence

Also called: evidence hierarchy, levels of evidence, level of evidence, evidence pyramid, hierarchy of study designs

A hierarchy of evidence is a ranking of study designs by how well each protects against bias when answering a particular kind of question, usually with systematic reviews and randomised trials near the top and case series and expert opinion near the bottom. Hierarchies differ by question type and have been criticised when applied rigidly, because the quality of each study matters as well as its design.

Nursing APA citations and the sources that count as evidence

Historical control

Also called: historical control group, historically controlled study, historically controlled trial, historical comparison group

A historical control is a comparison group made up of people treated or observed in an earlier period, whose results are compared with those of people receiving a new intervention now. It is a kind of external control. Changes over time in diagnosis, care, outcome definitions and the kind of patients treated can make the new intervention look better or worse than it is.

Intention-to-treat analysis

Also called: intention to treat analysis, intent-to-treat analysis, ITT analysis, ITT, intention-to-treat principle, analysis as randomised

Intention-to-treat analysis is the analysis of a randomised trial in which every participant is counted in the group they were randomised to, whether or not they received, completed or switched treatment. It preserves the balance randomisation created and estimates the effect of assigning an intervention. Versions labelled modified intention-to-treat leave some people out, such as those with missing outcomes.

Interim analysis

Also called: interim look, interim review

An interim analysis is an analysis of a trial's accumulating data carried out before recruitment or follow-up is complete, at points planned in advance, to decide whether the trial should continue, stop or change. In an adaptive design, what each look may lead to, such as stopping early or dropping an arm, must be set out in the protocol before any interim data are seen.

Interrupted time series

Also called: interrupted time-series, interrupted time series design, interrupted time series study, ITS design, ITS study

An interrupted time series is a quasi-experimental design that measures an outcome at many regular intervals before and after an intervention starts, and asks whether its level or trend changed beyond what the earlier trend predicted. Cochrane's EPOC group suggests at least three points before and three after. Adding a comparison series that had no intervention makes it stronger.

Intervention group

Also called: experimental group, treatment group, intervention arm, experimental arm, treatment arm, test group

An intervention group is the group in a study that receives the treatment, programme or condition being tested, and whose outcomes are compared with those of a control group. It is also called the experimental or treatment group, and in a trial the experimental arm. A study can have several intervention groups, each receiving a different version or dose.

Laboratory experiment

Also called: lab experiment, laboratory study, lab study

A laboratory experiment is an experiment run in a controlled setting chosen by the researcher, where the independent variable is manipulated and other conditions are held constant. That control makes cause and effect easier to establish, but the artificial setting and often narrow samples, such as undergraduate students, can limit how far the findings apply to everyday life.

Latin square design

Also called: Latin square, Latin-square design

A Latin square design is a design that arranges treatments in a grid with as many rows and columns as there are treatments, so that each treatment appears exactly once in every row and every column. In the design of experiments it blocks two nuisance factors at once, and psychologists use it to counterbalance the order of conditions with fewer orders than complete counterbalancing.

Master protocol

Also called: master protocol trial, master protocol design, master trial protocol

A master protocol is a single overall protocol under which several related sub-studies run, sharing infrastructure, screening and procedures, so that more than one question about treatments or diseases can be answered at once. Basket, umbrella and platform trials are the three main kinds, although the three names are used inconsistently in the literature.

Matched-pairs design

Also called: matched pairs design, matched-groups design, matched groups design, matched-subjects design, matched subjects design

A matched-pairs design is an experimental design in which participants are paired on characteristics that could affect the outcome, such as age or a pre-test score, and the members of each pair are then assigned to different conditions, ideally at random. It makes the groups more alike than chance alone would, at the cost of finding and measuring matches beforehand.

Mendelian randomisation

Also called: Mendelian randomization, MR study, Mendelian randomisation study, Mendelian randomization study

Mendelian randomisation is a method that uses genetic variants, which people inherit largely independently of their lifestyle, as stand-ins for an exposure, such as a gene affecting how alcohol is processed standing in for drinking, to estimate that exposure's effect on an outcome. It is an instrumental variable approach, often counted among natural experiments, and assumes the variant affects the outcome only through the exposure.

Minimisation

Also called: minimization, minimisation allocation, minimization allocation

Minimisation is a method of allocating participants in a trial in which each new person is placed in whichever group would best keep the groups balanced on several chosen characteristics, usually with a random element added. It can balance more factors than stratified randomisation in a small trial, and trials using it are generally treated as equivalent to randomised trials.

Mixed design

Also called: mixed factorial design, between-within design, mixed between-within design, mixed between-within subjects design

A mixed design is a factorial design with at least one between-subjects factor and at least one within-subjects factor. For example, every participant might solve both easy and hard puzzles, while one group works in silence and another with music. It should not be confused with mixed methods research, which combines quantitative and qualitative data.

Modified intention-to-treat analysis

Also called: modified intention to treat analysis, modified ITT, mITT, modified intent-to-treat analysis

A modified intention-to-treat analysis is an analysis that follows intention-to-treat in keeping participants in their randomised groups but leaves some of them out. What is left out varies: Cochrane uses the name for excluding people with missing outcome data, while NICE describes drug trials counting only people who took at least one dose, so readers need to check who was excluded.

Multi-arm multi-stage design

Also called: multi-arm multi-stage trial, MAMS design, MAMS trial

A multi-arm multi-stage design is an adaptive trial design that compares several interventions with a shared control group and, at planned interim stages, drops arms that are not working or are unsafe while continuing those that look promising. It tests many options more efficiently than separate trials. Some authors use the name almost interchangeably with platform trial, which can also add new arms.

Multi-arm trial

Also called: multi-arm study, multiple-arm trial, multi-arm randomised trial, multi-arm randomized trial

A multi-arm trial is a randomised trial with more than two groups, for example two new treatments and a shared control, or several doses of one drug. It answers several comparisons in one study. When its results enter a meta-analysis, care is needed not to count a shared control group twice when each intervention arm is compared with it.

Multicentre trial

Also called: multicenter trial, multi-centre trial, multi-center trial, multicentre study, multicenter study, multisite trial, multi-site study

A multicentre trial is a trial that recruits participants at several sites, such as different hospitals, schools or countries, under one protocol. It lets a trial recruit more people than one site could and tests the intervention in more than one setting, but it needs consistent procedures across sites, and randomisation is often stratified by centre.

Multiple-baseline design

Also called: multiple baseline design, multiple baseline study

A multiple-baseline design is a single-case experimental design in which a baseline is recorded for several participants, behaviours or settings, and the treatment is then introduced to each at a different, staggered time. If each changes only when the treatment reaches it, other explanations become unlikely. It avoids withdrawing a treatment that works, which a reversal design requires.

N-of-1 trial

Also called: n of 1 trial, N-of-1 study, n-of-one trial, single-patient trial, single patient trial, individual patient trial

An N-of-1 trial is a randomised crossover trial carried out in a single person, who receives the treatments being compared in several periods in a random order, often blinded, to find out which works best for them. It suits stable long-term conditions and treatments whose effects start and stop quickly, and a series of N-of-1 trials can be combined.

Natural experiment

Also called: natural experimental study, natural experimental evaluation, natural experiments

A natural experiment is a study that uses an event or policy outside the researcher's control, such as a new law, a tax change or a famine, which leaves some people exposed and others not, to estimate the effect of that exposure. The term has no exact definition, and it is used both for the event itself and for the study of it.

Nested case-control study

Also called: nested case control study, nested case-control design, case-control study nested in a cohort

A nested case-control study is a case-control study carried out inside an existing cohort, taking as cases the members who develop the outcome during follow-up and choosing controls from members who have not. Exposure data were often collected before anyone fell ill, which limits recall bias, and it is relatively cheap because the cohort's data already exist.

No-treatment control

Also called: no-treatment control group, no treatment control, no-intervention control, no-intervention group, untreated control group

A no-treatment control is a control group that receives no intervention at all during the study, so its results show what happens with the passage of time alone. It is simple, but its members know they are untreated, so any benefit in the intervention group may partly reflect expectation, which placebo, sham and wait-list controls are designed to address.

Non-equivalent groups design

Also called: nonequivalent groups design, non-equivalent control group design, nonequivalent control group design, non-equivalent comparison group design, NEGD

A non-equivalent groups design is a quasi-experimental design that compares a group receiving an intervention with a group that does not, where the groups were not formed by random assignment, as with two existing classes in different schools. Measuring both groups before and after helps, but differences between the groups may still explain the results.

Non-inferiority margin

Also called: noninferiority margin, non-inferiority limit, NI margin, margin of non-inferiority

A non-inferiority margin is the largest amount by which a new intervention may be worse than the standard one and still be judged acceptable, fixed before a non-inferiority trial begins. Non-inferiority is shown if the confidence interval for the difference does not cross that margin. A margin set too wide can let a genuinely worse treatment pass.

Non-inferiority trial

Also called: noninferiority trial, non-inferiority study, non-inferiority design, NI trial

A non-inferiority trial is a randomised trial designed to show that a new intervention is not worse than an established one by more than a margin set in advance. It is used when the new option has another advantage, such as lower cost, fewer side effects or easier use. Its hypothesis is one-sided, and both intention-to-treat and per-protocol analyses are recommended.

Non-randomised controlled trial

Also called: non-randomized controlled trial, nonrandomised controlled trial, nonrandomized controlled trial, non-randomised trial, non-randomized trial, NRCT, controlled clinical trial, CCT

A non-randomised controlled trial is an experimental study in which the researchers allocate participants to intervention and control groups by a method that is not random, such as clinician choice or place of treatment. It carries a greater risk of bias than a randomised trial, because the groups may differ in ways that affect the outcome.

Non-randomised study of interventions

Also called: non-randomized study of interventions, NRSI, non-randomised study, non-randomized study, non-randomised intervention study

A non-randomised study of interventions is any study that estimates the effect of an intervention without allocating people at random, including non-randomised trials, controlled before-and-after studies, interrupted time series and cohort studies. Cochrane uses the name as an umbrella and advises judging such studies by their actual design features, since the labels for them are applied inconsistently.

Observational study

Also called: observational design, non-experimental study, nonexperimental study, non-interventional study

An observational study is a study in which the researcher records exposures and outcomes as they occur, or occurred, without assigning any intervention. Cohort, case-control, cross-sectional and ecological studies are the main kinds. It can examine exposures that could never be assigned, such as smoking, but confounding makes causal claims harder than in an experiment.

One-group post-test only design

Also called: one-group posttest-only design, one-group posttest only design, one-shot case study

A one-group post-test only design is a design in which a single group receives an intervention and is measured once afterwards, with no pre-test and no comparison group. It is the weakest quasi-experimental design, since nothing shows what the outcome would have been without the intervention, yet its results are often reported and misread.

One-group pre-test post-test design

Also called: one-group pretest-posttest design, pre-test post-test design, pretest-posttest design, pre-post study, pre-post design, before-and-after study, before-after study, uncontrolled before-and-after study

A one-group pre-test post-test design is a quasi-experimental design in which a single group is measured, receives an intervention and is measured again. It is simple, but with no comparison group any change could come from other events, natural improvement, regression to the mean or practice on the test. Adding a comparison group makes it far stronger.

Open cohort

Also called: dynamic cohort, dynamic population, open cohort study

An open cohort is a cohort whose membership changes during the study, as people enter and leave through birth, death or moving, such as the residents of a town. Because people are followed for different lengths of time, it yields incidence rates and rate ratios, not risks. Its make-up by age and sex may still stay roughly stable over a short period.

Open-label trial

Also called: open label trial, open-label study, open trial, open study, unblinded trial, non-blinded trial

An open-label trial is a trial in which both the participants and the researchers know which intervention each person is receiving. It is used when blinding is impossible, as when surgery is compared with medication. Allocation can still be concealed until the moment of assignment, and outcome assessors can sometimes still be kept unaware.

Order effect

Also called: sequence effect

An order effect is a change in participants' responses caused by the order in which they experience the conditions of a within-subjects experiment, not by the conditions themselves. Carryover, practice, fatigue and context effects are all kinds. Researchers deal with order effects by counterbalancing, so that each condition comes first, second and so on equally often.

Panel study

Also called: panel design, panel survey, longitudinal panel study, longitudinal panel survey

A panel study is a longitudinal study that surveys the same individuals or households repeatedly, in waves, often over many years. Because the same people answer each time, it can track change within individuals, which a repeated cross-sectional survey drawing a fresh sample each time cannot. A household panel usually starts from a random sample of the population.

Parallel-group trial

Also called: parallel group trial, parallel-group design, parallel design, parallel-arm trial, parallel trial, parallel group randomised trial, parallel assignment

A parallel-group trial is a randomised trial in which each participant is allocated to one group and stays in it, and the groups are followed side by side and compared at the end. It is the most common design for randomised trials. It differs from a crossover trial, in which each participant receives every intervention in turn.

Per-protocol analysis

Also called: per protocol analysis, PP analysis, per-protocol population analysis

A per-protocol analysis is an analysis of a trial that includes only the participants who kept to the protocol, for example by taking their assigned treatment as planned. It estimates the effect of actually following the intervention, but leaving people out can undo the balance randomisation created, so on its own it can be biased. An as-treated analysis, grouping people by what they received, is different again.

Period effect

A period effect, in a crossover trial, is a systematic difference between outcomes in the first and later treatment periods that is not caused by the treatments, for example because the disease progresses or care improves over time. Analyses of crossover trials allow for it. Cochrane names it, with carryover, as a key concern when judging a crossover trial.

Phase 0 trial

Also called: phase 0 study, phase zero trial, microdosing study, microdose study, exploratory IND study

A phase 0 trial is an exploratory first study of a new medicine in a small number of people, given at very low doses for a short time, up to about a week, with no aim of treating or diagnosing anyone. It checks how the body handles the drug, often using a microdose, and it followed European guidance in 2003 and US guidance in 2006.

Phase I trial

Also called: phase 1 trial, phase I clinical trial, phase 1 clinical trial, phase I study, phase 1 study

A phase I trial is the first stage of testing a new medicine in people, usually 20 to 100 healthy volunteers or patients, to learn how the body handles it, which doses are tolerated and what side effects appear. It typically lasts several months. Cancer medicines are usually tested at this stage in patients who have the disease.

Phase II trial

Also called: phase 2 trial, phase II clinical trial, phase 2 clinical trial, phase II study, phase 2 study

A phase II trial is the second stage of testing a new medicine, in which up to several hundred people with the condition receive it to gather early evidence of whether it works and more information on side effects. It lasts from several months to about two years and is too small to settle benefit, but it shapes the design of phase III.

Phase III trial

Also called: phase 3 trial, phase III clinical trial, phase 3 clinical trial, phase III study, phase 3 study, pivotal trial, pivotal study

A phase III trial is a large trial, involving roughly 300 to 3,000 people with the condition, that tests whether a new medicine offers a real benefit to a defined population and monitors less common side effects, typically over one to four years. It provides most of the safety data gathered before approval and is sometimes called a pivotal study.

Phase IV trial

Also called: phase 4 trial, phase IV study, phase 4 study, post-marketing study, postmarketing study, post-approval study

A phase IV trial is a study carried out after a medicine or device has been approved and is in general use, often involving several thousand people, to monitor its safety and effects over a longer time and in a wider population. It is part of post-market safety monitoring and can reveal rare harms that earlier, smaller trials could not detect.

Placebo control

Also called: placebo control group, placebo group, placebo arm, placebo comparator, placebo-controlled

A placebo control is a control group that receives a placebo, an inactive treatment made to look and be taken exactly like the one being tested, such as a sugar pill. It allows participants and staff to be blinded and separates the treatment's own effect from the effect of expecting to be treated. For procedures, the equivalent is a sham control.

Platform trial

Also called: platform study, platform design, platform randomised trial, platform randomized trial

A platform trial is a trial run under one master protocol that compares several interventions, often against a shared control group, and lets arms be dropped or new ones added over time according to rules set in advance. It can continue for years as treatments come and go. Some papers call it a multi-arm multi-stage design, and definitions vary between authors.

Post-test only control group design

Also called: posttest-only control group design, posttest-only randomized experiment, two-group posttest-only randomized experiment, post-test only randomised design

A post-test only control group design is a true experimental design in which participants are randomly assigned to an intervention or a control group and measured only once, afterwards. Because random assignment makes the groups equivalent in expectation, no pre-test is needed, which also avoids the risk that a pre-test alters later responses. It is the simplest randomised experiment.

Practice effect

Also called: learning effect

A practice effect is an order effect in which participants in a within-subjects experiment perform better in later conditions because they have had practice at the task, not because of the condition itself. It is a kind of carryover effect and the opposite of a fatigue effect, and it is controlled by counterbalancing the order of conditions.

Pragmatic trial

Also called: pragmatic randomised trial, pragmatic randomized trial, pragmatic clinical trial, pragmatic randomised controlled trial, effectiveness trial

A pragmatic trial is a randomised trial designed to show whether an intervention works in ordinary practice, with typical participants, usual settings and staff, flexible delivery and a comparison with usual care, so that its results can guide real decisions. It measures effectiveness rather than efficacy, and it lies towards the opposite end of a continuum from the explanatory trial.

Pre-test post-test control group design

Also called: pretest-posttest control group design, pretest-posttest randomized experiment, pre-test post-test randomised design

A pre-test post-test control group design is a true experimental design in which participants are randomly assigned to an intervention or a control group and measured both before and after the intervention period. The pre-test lets researchers check that the groups started alike and measure change, but it may itself affect later answers, a risk the Solomon four-group design was built to test.

Prospective cohort study

Also called: prospective cohort, prospective study, concurrent cohort study, prospective follow-up study

A prospective cohort study is a cohort study that recruits people, records their exposures and then follows them into the future to see who develops the outcome. Data can be collected as planned, but results may take years. Because authors use prospective and retrospective in different ways, STROBE asks them to say how and when data were collected.

Quasi-experimental design

Also called: quasi-experiment, quasi experiment, quasi-experimental study, quasi experimental design, quasi-experimental research

A quasi-experimental design is a study that compares conditions or measures the effect of an intervention, as an experiment does, but without assigning participants to groups at random. It is used when randomising is impractical or unethical, as with a new school policy. Common forms include non-equivalent groups, interrupted time series and regression discontinuity designs.

Quasi-randomised trial

Also called: quasi-randomized trial, quasi-randomised controlled trial, quasi-randomized controlled trial, quasi-random allocation

A quasi-randomised trial is a trial that allocates participants by a rule that looks random but is predictable, such as alternation, date of birth or day of admission. Because the next allocation can be foreseen, it is open to selection bias. The term is used in different senses by different authors, and Cochrane's EPOC group prefers to call such studies non-randomised trials.

Randomisation

Also called: randomization, random allocation, random assignment, randomised allocation, randomized allocation

Randomisation is the allocation of participants to groups by chance, using a method such as a computer-generated sequence, so that each person has a known, usually equal, chance of each group. It balances known and unknown characteristics on average across the groups. It is different from random sampling, which decides who enters a study, not which group they join.

Randomised block design

Also called: randomized block design, randomised complete block design, randomized complete block design, RCBD

A randomised block design is an experimental design that sorts experimental units into blocks that are alike on a nuisance factor, such as batch, day or operator, and then assigns every treatment at random within each block. It removes that factor's variation from the comparison. It is a design-of-experiments idea, distinct from block randomisation, which keeps trial groups equal in size.

Randomised controlled trial

Also called: randomized controlled trial, RCT, randomised trial, randomized trial, randomised clinical trial, randomized clinical trial, randomised control trial, randomized control trial

A randomised controlled trial is an experiment in which participants are allocated at random to two or more groups, one usually receiving the intervention being tested and another a comparison such as placebo or usual care, and the groups are followed to compare outcomes. Randomisation makes it the most reliable single design for judging whether an intervention causes an effect.

Regression discontinuity design

Also called: regression-discontinuity design, regression discontinuity, RD design, RDD

A regression discontinuity design is a quasi-experimental design in which people are assigned to an intervention solely by whether they fall above or below a cut-off on a score measured beforehand, such as a test mark or a poverty rate, and the outcomes on either side of the cut-off are compared. Done well, it rivals a randomised experiment in credibility but needs more participants.

Repeated cross-sectional study

Also called: repeated cross-sectional survey, repeated cross-sectional design, serial cross-sectional survey, trend study, trend studies

A repeated cross-sectional study is a study that asks the same or similar questions of a new sample of people at regular intervals, so that changes in the population can be tracked over time. Because different people answer each time, it shows change in groups, such as smoking rates among men and women, but it cannot show how any individual changes.

Repeated measures design

Also called: repeated-measures design, repeated measures study, repeated-measures study

A repeated measures design is a design in which the same participants are measured more than once, under different conditions or at several points in time. Many textbooks treat the name as a synonym for a within-subjects design, others use it for measuring one group repeatedly over time, and Cochrane's EPOC group applies it to an interrupted time series measured in the same people.

Response-adaptive randomisation

Also called: response-adaptive randomization, outcome-adaptive randomisation, outcome-adaptive randomization, RAR

Response-adaptive randomisation is a form of randomisation in which the allocation ratio changes during a trial according to the outcomes seen so far, so that more participants are assigned to treatments that appear to be doing better. It is used in some adaptive and platform trials, and it needs outcomes that become known quickly, since allocation can only respond to results already observed.

Restricted randomisation

Also called: restricted randomization

Restricted randomisation is any method of randomisation that places limits on the allocation sequence to keep the groups balanced, instead of leaving every allocation to pure chance. Blocking is the most common form, and stratified randomisation depends on it. The balance comes at some cost in predictability, which is why block sizes are often varied and kept from recruiters.

Retrospective cohort study

Also called: retrospective cohort, historical cohort study, non-concurrent cohort study, nonconcurrent cohort study

A retrospective cohort study is a cohort study that begins after both the exposure and the outcome have occurred, using existing records, such as employment or medical files, to reconstruct who was exposed and what happened to them. It is quicker and cheaper than a prospective cohort study, but it depends on records that were not made for research.

Reversal design

Also called: ABA design, A-B-A design, ABAB design

A reversal design is a single-case experimental design in which a baseline is recorded (A), a treatment is introduced (B) and then withdrawn to return to baseline (A), often with the treatment brought back again (ABAB). If the behaviour changes each time the treatment starts and stops, the treatment is the likely cause. It cannot be used when removing a treatment would be unethical.

Run-in period

Also called: run-in phase, run in period, lead-in period, placebo run-in period

A run-in period is a stage between enrolment and randomisation in a trial during which all participants receive the same thing, an active treatment, a placebo or nothing, and those who do not take it reliably or cannot tolerate it are excluded before randomisation. It can raise adherence during the trial, but the people finally randomised may be less typical of patients in general.

Sample size re-estimation

Also called: sample size recalculation, sample size re-calculation, sample size reassessment, sample size adjustment

Sample size re-estimation is an adaptation in which a trial's planned number of participants is recalculated partway through, using interim data, because assumptions made at the design stage, such as how variable the outcome is, may have been wrong. It can be carried out with or without knowledge of which participants are in which group.

Seamless phase II/III design

Also called: seamless phase 2/3 design, seamless phase II/III trial, seamless design, phase II/III trial, phase 2/3 trial

A seamless phase II/III design is an adaptive trial design that joins the dose-finding or treatment-selection stage and the confirmatory stage in one trial under a single protocol, instead of running two separate trials. It removes the gap between phases. In an inferentially seamless version, data from both stages contribute to the final analysis, while an operationally seamless one analyses them separately.

Sham control

Also called: sham procedure, sham intervention, sham comparator, sham surgery, sham treatment, sham-controlled

A sham control is a control group that receives an imitation of a procedure, made to resemble the real one in every respect except its active element, such as an operation under anaesthesia that leaves out the key surgical step. It plays the part a placebo pill plays in drug trials, allowing blinding and separating the specific effect from the context of treatment.

Simple randomisation

Also called: simple randomization, unrestricted randomisation, unrestricted randomization, complete randomisation, complete randomization

Simple randomisation is randomisation with no restrictions, in which each allocation is made by chance alone, like tossing a coin for every participant. It is the least predictable method and so the hardest to subvert, but in small trials it can leave groups unequal in size or make-up, which is why block and stratified methods are used.

Single-arm trial

Also called: single-arm study, single arm trial, single-group trial, one-arm trial, uncontrolled trial

A single-arm trial is a trial in which every participant receives the intervention being tested and there is no concurrent control group. Its results are compared with an external benchmark, such as historical data, which makes the effect harder to judge than in a randomised trial. It is used mainly when randomising is not feasible, as in some rare diseases.

Single-blind

Also called: single blind, single-blinded, single-masked, single-blind study, single-blind trial

A single-blind study is a study in which one party, usually the participants, does not know which intervention each person is receiving, while the researchers do. Because the label is interpreted in different ways, reporting guidelines ask authors to name exactly who was blinded instead of relying on it.

Single-case experimental design

Also called: single-case design, single-subject design, single-subject experimental design, single-subject research design, SCED

A single-case experimental design is an experimental design in which one participant, or a few, is measured repeatedly while a condition is introduced and withdrawn, as in the ABA reversal design, or introduced at staggered times, as in the multiple-baseline design. Each participant acts as their own control, and results are usually judged from graphs. It is central to applied behaviour analysis.

Solomon four-group design

Also called: Solomon four group design, Solomon 4-group design, Solomon design

A Solomon four-group design is a true experimental design with four randomly assigned groups, in which two receive the intervention and two do not, and within each pair one group takes a pre-test and the other does not. Comparing the four shows whether taking the pre-test itself changed the results, as well as the intervention's effect, at the cost of needing twice as many groups.

Stepped-wedge design

Also called: stepped wedge design, stepped-wedge trial, stepped wedge cluster randomised trial, stepped wedge cluster randomized trial, stepped-wedge cluster trial, SW-CRT

A stepped-wedge design is a cluster randomised trial in which all clusters begin without the intervention and then switch to it one after another, at times decided at random, until every cluster has it. It appeals when every cluster is meant to receive the intervention in the end. Because later periods contain more clusters with the intervention, the analysis must adjust for trends over time.

Stratified randomisation

Also called: stratified randomization, stratified random allocation, stratified block randomisation, stratified block randomization

Stratified randomisation is randomisation carried out separately within subgroups, or strata, defined by characteristics that strongly affect the outcome, such as study centre or disease severity, so that the groups end up balanced on them. It needs a restriction such as blocking within each stratum to work. It is not the same as stratified sampling, which chooses who takes part.

Superiority trial

Also called: superiority study, superiority design

A superiority trial is a randomised trial designed to show that one intervention is better than another, or than placebo, on the main outcome. Most trials are of this kind. A trial that fails to show superiority has not shown that the two are equal, which is a separate question for an equivalence or non-inferiority trial.

Survey research

Also called: survey study, survey-based research, survey research design, survey method

Survey research is a research design in which information is gathered by asking a sample of people questions, in questionnaires or interviews, usually to describe a larger population or to relate their answers to one another. It is often cross-sectional but can be repeated or longitudinal, and its worth depends heavily on how the sample was drawn.

Switching replication design

Also called: switching replications design, switching-replication design, pretest-posttest design with switching replication, switching replications

A switching replication design is a design with two groups in which one group receives the intervention while the other waits, and then the roles switch so that the second group receives it too, with measurements at three points. The effect is tested twice, and everyone gets the intervention in the end. It is used both with random assignment and in quasi-experiments.

Synthetic control method

Also called: synthetic control, synthetic controls, synthetic control approach, synthetic control methodology

The synthetic control method estimates the effect of an intervention on a single area or group, such as a country or region, by building a comparison from a weighted combination of similar units that did not receive it. The weights make the synthetic control track the treated unit closely before the intervention, so a gap afterwards is read as the effect. There is no agreed test of what counts as a good fit.

Target trial emulation

Also called: target trial, emulated target trial, target trial framework, trial emulation

Target trial emulation is an approach to analysing observational data in which researchers first write the protocol of the randomised trial they would ideally run, the target trial, and then design the analysis to mimic it, specifying eligibility, the treatment strategies, assignment, the start of follow-up, outcomes and analysis. Doing so helps avoid errors such as immortal time bias.

Trial arm

Also called: arm, study arm, arm of a trial, arm of a clinical study

A trial arm is one of the groups in a trial, defined by the intervention its participants receive, such as the new drug arm, the placebo arm or the usual care arm. A two-arm trial compares two groups and a multi-arm trial compares more. Each arm is set out in the protocol with its own intervention and schedule.

Triple-blind

Also called: triple blind, triple-blinded, triple-masked, triple-blind study, triple-blind trial

A triple-blind study is a study in which the participants, the people giving care or collecting data, and a third group, usually those analysing the results, are all kept unaware of who received which intervention. Its meaning varies between reports, so reporting guidelines recommend stating exactly who was blinded instead of using the label.

True experiment

Also called: true experimental design, true experimental research, randomised experiment, randomized experiment

A true experiment is an experiment in which the researcher manipulates an independent variable and assigns participants to conditions at random, while controlling other variables. Random assignment is what separates it from a quasi-experiment and gives it the strongest grounds for a causal conclusion. In health research the randomised controlled trial is its equivalent.

Umbrella trial

Also called: umbrella study, umbrella design, umbrella protocol

An umbrella trial is a trial, run under one master protocol, that studies a single disease and assigns patients to different targeted treatments according to the biological features of their disease, such as the mutation their tumour carries. It is the mirror image of a basket trial, which tests one treatment across several diseases. Definitions of both vary between papers.

Unit of randomisation

Also called: unit of randomization, unit of allocation, randomisation unit, randomization unit, allocation unit

The unit of randomisation is the thing that is allocated at random in a trial, usually an individual person but sometimes a cluster such as a school, a ward or a general practice. When outcomes are analysed for individuals in a trial that randomised clusters, the analysis must account for the clustering, or the results will look more precise than they are.

Usual care control

Also called: usual care, treatment as usual, TAU, usual care group, standard care control, standard of care control, usual care arm

A usual care control is a control group that continues to receive whatever care it would normally get, so that a new intervention is compared with current practice rather than with nothing or a placebo. It is typical of pragmatic trials. Because usual care differs between places and over time, its content needs describing for results to be interpreted.

Wait-list control

Also called: waitlist control, wait-list control group, waitlist control group, waiting-list control, wait-list control condition

A wait-list control is a control group whose members are told they will receive the intervention after the study period, and who are compared meanwhile with people receiving it now. It is often used in psychotherapy research, since everyone is offered help in the end, though people waiting for treatment may still improve on their own.

Washout period

Also called: wash-out period, washout, wash-out, washout phase

A washout period is an interval in a crossover trial, between one treatment period and the next, during which no study treatment is given, so that the effects of the first treatment can fade before the second begins. Its length depends on how long the treatment acts, and Cochrane notes that whether one is needed at all depends on the treatment.

Within-subjects design

Also called: within-subject design, within subjects design, within-participants design, within-groups design

A within-subjects design is an experimental design in which every participant takes part in every condition, so each person is compared with themselves. It needs fewer participants and removes differences between people from the comparison, but the order of conditions can affect the results, which researchers deal with by counterbalancing. Many textbooks also call it a repeated measures design.