Adaptive Field Experiment — Sequential Adaptation in Natural Settings
Adaptive Field Experiment · Also known as: adaptive field trial, sequentially adaptive field experiment, responsive field experiment, adaptive randomized field study
An adaptive field experiment is a randomized study conducted in a real-world environment in which pre-specified decision rules allow the researcher to modify the trial as interim data accumulate — for example, by reallocating participants toward more effective arms, adjusting sample size, or stopping early for efficacy or futility — all while maintaining statistical integrity.
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 field experiment when you have multiple treatment arms competing for scarce resources, when ethical concerns make it undesirable to continue assigning participants to an arm that is clearly inferior, or when logistical timelines require interim decision-making before full data collection is complete. It is particularly valuable in development economics, public health, and education policy where scale-up must follow quickly. Do not use it when the outcome takes as long to observe as the entire study duration (no useful interim data), when the field logistics cannot support protocol changes mid-study without contamination, when sample sizes are too small for reliable interim analyses, or when the regulatory or publication context demands a fixed-design approach.
Strengths & limitations
- Allocates more participants to superior arms as evidence accumulates, increasing the expected benefit delivered within the study.
- Can stop early for efficacy or futility, reducing sample size and cost when the answer becomes clear.
- Retains the external-validity advantages of field experimentation — results generalize to real populations and contexts.
- Pre-specified adaptive rules are transparent and reproducible, distinguishing legitimate adaptation from post-hoc data dredging.
- Compatible with Bayesian frameworks that provide richer probability statements about treatment differences than classical tests.
- Substantially more complex to design and execute than a fixed field experiment; requires specialized statistical expertise and pre-registration infrastructure.
- Operational disruptions in the field (missing data, attrition, delayed outcomes) can undermine interim analyses and trigger spurious adaptations.
- Response-adaptive randomization can introduce temporal confounding if background conditions change between early and late enrollment waves.
- Regulators and many journals remain cautious about adaptive designs; additional methodological transparency is required for peer review.
Frequently asked
How is an adaptive field experiment different from a standard field experiment?
A standard field experiment fixes the protocol — sample size, randomization weights, and arms — before data collection begins and does not change them. An adaptive field experiment pre-specifies decision rules that allow modifications (arm dropping, re-weighting, sample-size adjustment) based on interim results. Both are conducted in naturalistic settings, but the adaptive version adds sequential evaluation checkpoints governed by pre-registered statistical rules.
Does adaptation inflate Type I error?
It can, if not handled correctly. Classical p-values computed at the end of an adaptive study are invalid unless an appropriate alpha-spending function (e.g., O'Brien-Fleming boundaries) is applied across all interim and final analyses. Bayesian adaptive designs sidestep this issue by updating posterior probabilities rather than performing hypothesis tests. Either approach is valid, but the choice must be pre-specified.
What is response-adaptive randomization and when should I avoid it?
Response-adaptive randomization (RAR) shifts assignment probabilities toward the arm showing the best interim outcomes. It maximizes benefit within the trial but introduces temporal confounding when background conditions change over the enrollment period — for example, if economic shocks or seasonal effects alter outcomes for later participants. RAR is most defensible when enrollment is rapid and background conditions are stable.
Do I need an independent monitoring board?
For any adaptation that could halt or substantially modify the study, an independent data safety and monitoring board (DSMB or DMC) is strongly recommended. The board reviews unblinded interim data according to the pre-specified rules while the primary research team remains blinded, protecting against bias that could arise if investigators who know the interim results continue to manage participant enrollment.
Can I use this design with a small sample?
Adaptive designs require sufficient interim data to produce reliable estimates at each checkpoint. With very small total samples the interim analyses lack statistical power, and adaptive decisions become driven by noise rather than signal. As a rough guide, each arm should have at least 20–30 outcome observations before an interim analysis is meaningful; below this threshold, a fixed design is safer.
Sources
- Berry, D. A. (2004). Bayesian statistics and the efficiency and ethics of clinical trials. Statistical Science, 19(1), 175–187. DOI: 10.1214/088342304000000044 ↗
- Duflo, E., Glennerster, R., & Kremer, M. (2007). Using randomization in development economics research: A toolkit. Handbook of Development Economics, 4, 3895–3962. DOI: 10.1016/S1573-4471(07)04061-2 ↗
How to cite this page
ScholarGate. (2026, June 3). Adaptive Field Experiment. ScholarGate. https://scholargate.app/en/experimental-design/adaptive-field-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
- Factorial Field ExperimentExperimental design↔ compare
- Field ExperimentExperimental design↔ compare
- Multi-arm experimentExperimental design↔ compare
- Randomized Controlled TrialExperimental design↔ compare