Adaptive Laboratory Experiment
Also known as: adaptive lab experiment, sequential adaptive laboratory study, response-adaptive laboratory design, adaptive experimental laboratory design
An adaptive laboratory experiment is a controlled experimental design conducted in a laboratory setting where pre-specified decision rules allow modifications to the study — such as sample size, treatment allocation, or stopping criteria — based on accumulating data. Unlike fixed designs, adaptive designs incorporate planned interim analyses that permit the experiment to respond to emerging evidence while maintaining statistical validity and Type I error control.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use an adaptive laboratory experiment when you are studying multiple experimental conditions in a controlled laboratory environment and anticipate uncertainty about the optimal sample size, the number of viable arms, or the effect size. It is particularly well suited to dose-finding studies, multi-arm screening experiments in pharmacology or cognitive neuroscience, and pilot phases where early evidence should guide resource allocation. Prefer it when ethical or practical constraints make running all participants under clearly inferior conditions undesirable. Do NOT use adaptive designs when the laboratory turnaround is so fast that interim results are unavailable before most participants have already been run; when the team lacks statistical expertise to pre-specify valid adaptive rules and perform corrected analyses; when the regulatory or publication context requires a strict pre-registered fixed design; or when carryover and learning effects in the laboratory make interim-based arm dropping logistically unmanageable.
Strengths & limitations
- Increases efficiency: promising arms receive more participants and inferior arms are dropped, reducing wasted resources.
- Enables early stopping for overwhelming efficacy or clear futility, shortening study duration when evidence is conclusive.
- Sample-size re-estimation corrects for initial variance misestimation without inflating Type I error if done correctly.
- Ethically superior to fixed designs when one arm is clearly harmful or clearly superior mid-study.
- Retains full internal validity of laboratory control — randomization, blinding, and environmental control are unchanged.
- Requires complex pre-specification; poor planning of adaptive rules can introduce bias or inflate false-positive rates.
- Interim analyses add operational burden and may require independent data monitoring infrastructure.
- Response-adaptive randomization can reduce statistical power compared to equal allocation when effect sizes are small.
- Results can be harder to interpret and communicate than those from simple fixed designs, especially for non-statistician audiences.
Frequently asked
Is an adaptive laboratory experiment the same as a sequential experiment?
Sequential analysis is a broader statistical framework — any design that analyzes data as it accumulates qualifies. An adaptive laboratory experiment is a specific application of sequential principles in a controlled laboratory setting that may include not only sequential stopping but also arm dropping, sample-size re-estimation, and response-adaptive randomization. All adaptive laboratory experiments are sequential, but not all sequential experiments incorporate the full range of adaptive modifications.
Does adaptation compromise internal validity?
Not if the adaptive rules are pre-specified and the laboratory conditions themselves are held constant. Adaptation changes who gets which condition and for how long, but it does not alter the control over extraneous variables that defines laboratory research. The main threat is blinding failure: if experimenters learn interim results and unconsciously change how they run sessions, bias can enter. Maintaining blinding of lab staff during interim analyses is therefore essential.
Can I use a staircase procedure as an adaptive laboratory experiment?
Yes. Staircase (or PEST, QUEST) psychophysical procedures are among the oldest and most widely used adaptive laboratory methods. They adjust stimulus intensity trial-by-trial based on participant responses to estimate perceptual thresholds efficiently. They represent a special case of response-adaptive allocation at the single-participant level rather than at the between-participants arm level.
What software supports adaptive laboratory experiment analysis?
R packages such as 'rpact', 'gsDesign', and 'adaptTest' support frequentist adaptive design planning and analysis. EAST (Cytel) is widely used in pharmaceutical settings. For Bayesian adaptive designs, 'RBesT' and custom Stan/JAGS models are common. The key requirement is that the software used at the analysis stage matches the adaptive rules pre-specified at the design stage.
How do I pre-register an adaptive laboratory experiment?
Pre-registration should include the complete adaptive decision rules — interim analysis timing, stopping boundaries, arm-dropping criteria, and re-estimation formulas — as well as the corrected final analysis method. Platforms such as AsPredicted, OSF, and ClinicalTrials.gov (for clinical-adjacent studies) support pre-registration. The pre-registration must be filed and time-stamped before any data are collected.
Sources
- Berry, D. A. (2006). Bayesian clinical trials. Nature Reviews Drug Discovery, 5(1), 27–36. DOI: 10.1038/nrd1927 ↗
- Adaptive design (medicine). Wikipedia. link ↗
How to cite this page
ScholarGate. (2026, June 3). Adaptive Laboratory Experiment. ScholarGate. https://scholargate.app/en/experimental-design/adaptive-laboratory-experiment
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Adaptive ExperimentExperimental design↔ compare
- Adaptive Randomized Controlled TrialExperimental design↔ compare
- Laboratory ExperimentExperimental design↔ compare
- Multi-arm experimentExperimental design↔ compare
- Sequential AnalysisStatistics↔ compare