GCRI · Labs
研究方案
A research protocol fixes, before data are collected, what a study asks, how it will answer and what would count as an answer: the hypotheses, the design, the controls, the outcomes, the sample and the analysis. Many failures of credibility trace back to a protocol that was vague, unwritten or changed after the data were seen. This page sets out how protocols are built, registered, reviewed and reported, and what ethics committees, editors and funders rely on when they accept a study’s claims.
工作如何开展。
方案与复现看板
Research protocols · Labs · Quarter 3
协议
| PR-378 sensor drift | 田野实验 | 是 | 已批准 | 运行中 |
| PR-705 AI triage | 随机化 | 是 | 条件 | 设置 |
| PR-694 message recall | 调查实验 | 是 | 已批准 | 已完成 |
| PR-360 heat exposure | 学员群组 | 否 | 待处理 | 草稿 |
Replication R-25
- 原始效应
- 0.4 SD
- 复现效应
- 0.2 SD
- 结论
- 较小，符号相同
- 已发布
- 附数据和代码
试验平台可用性
图Protocol and replication board: example names and illustrative values, not a record or a live system.
Research protocols: how the risk develops.
Hypotheses, controls and scientific acceptance. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.
Flexibility becomes bias
When outcomes, exclusions, covariates or stopping rules are chosen after the data are seen, many analyses become possible and the most striking one tends to be reported. Each choice looks reasonable alone; together they inflate false positives and effect sizes.
Weak controls confound results
A missing control, unblinded assessment or non-random allocation lets something other than the intervention explain the difference. Design flaws cannot be repaired by analysis; they travel into every result and review built on the study.
Underpowered designs mislead twice
Small samples miss real effects, and the effects they do detect tend to be overestimated. Underpowered studies that reach significance seed literatures that later, larger replications cannot match.
Deviations go unreported
Protocols change for good reasons, such as slow recruitment or a safety signal. When a change is not dated, explained and reported, readers cannot tell a justified amendment from outcome switching, and reviews pool what should not be pooled.
How research protocols is modelled, measured and tested.
模型、其假设、数据和不确定性，以科学家或工程师会核查的形式呈现。结构遵循已发表的方法。所取数值仅为示意，不是研究发现。
模型、方程与假设
The model structure used for research protocols, the assumptions it makes and how each is tested.
Ri = f(Hi, Ei, Vi, Ci)危害、风险暴露、脆弱性和应对能力EAL = ∑k pk · Lk对年概率为 pkxt+1 = A xt + B ut + εtA carries the couplings, u the shock, ε the error| 假设 | 为何制作 | 如何检验 |
|---|---|---|
| 在一次事件中，危害与暴露相互独立 | 使损失积分保持可计算 | 与基于联合记录拟合的 copula 进行比较 |
| 耦合在工作点附近是线性的 | 让冲击通过 A 传播 | 在线性范围之外施加扰动并比较 |
| 脆弱性曲线可以在不同地点之间迁移使用 | 很少有本地曲线可用 | 在有观测损失的地方据此进行验证 |
- 输入
- 带溯源信息的版本化数据集
- 成果
- 分布，绝不是单一数字
- 代码
- 连同模型卡和测试一起发布
- 复现
- 相同的数据、代码、环境和随机种子
衡量什么，以及衡量得如何。
每个指标都注明其单位、频率、背后的来源类型以及数值将附带的证据标签。此处不显示数值：数值由计划依据其自身权限提供。
| 指标 | 单位 | 节奏 | 来源类型 | 标签 |
|---|---|---|---|---|
| Registration before data collection | percent of studies with a time-stamped registration | 年度 | Trial and study registries | 已发布统计 |
| Primary outcome consistency | percent of reports matching the registered primary outcome | 年度 | Registry and publication comparison | 观测 |
| Sample size justification | percent of protocols with a power or precision rationale | 年度 | Protocol audit | 观测 |
| Reporting guideline adherence | percent of applicable checklist items reported | 每项研究 | Assessment against CONSORT, PRISMA, ARRIVE or STROBE | 估算 |
| Protocol amendments | count per study, each with a stated reason | 每项研究 | Registry history and ethics committee records | 已发布统计 |
| Registered report availability | number of journals offering the format | 年度 | Center for Open Science journal list | 已报告 |
| Results posting timeliness | months from primary completion to posted results | 每项研究 | Trial registries | 已发布统计 |
| Randomisation and blinding reported | percent of eligible studies | 年度 | Risk-of-bias assessments | 估算 |
证据来自何处。
具名公共数据集与机构记录，各自注明许可类别。依协议获得的来源仅按其条款使用，绝不再次发布。
| 来源 | 它提供什么 | 许可类别 |
|---|---|---|
| OSF Registries (Center for Open Science) | Time-stamped preregistrations and registered report protocols across disciplines | Open, licence per registration |
| ClinicalTrials.gov (US National Library of Medicine) | Registered trials, protocol change history and posted results | 公有领域 |
| WHO International Clinical Trials Registry Platform | Search across primary national trial registries | Free, WHO terms of use |
| PROSPERO (NIHR) | Registered protocols for systematic reviews | Free access, terms of use |
| protocols.io | Versioned, citable step-by-step laboratory and field methods | Open, CC BY for public protocols |
| EQUATOR Network library | Reporting guidelines by study design, with extensions | Free access, licence per guideline |
| Crossref metadata | Links between registrations, articles, corrections and funders | Open metadata, no restrictions |
由谁决定，以及证据服务于什么。
实验室 prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.
| 谁来决定 | 问题 | 本页面服务的证据 |
|---|---|---|
| Research ethics committees and institutional review boards | Whether a protocol’s risks, consent and design justify approval, and whether an amendment needs fresh review | Complete protocol, sample size rationale, amendment history with reasons |
| Journal editors | Whether to accept a registered report in principle, or whether a submitted analysis matches its plan | Time-stamped registration compared line by line with the manuscript |
| 研究资助方 | Whether a proposal’s design can answer its question before money is committed | Power or precision rationale, control conditions, data management plan |
| Medicines and device regulators | Whether trial evidence was generated under the protocol submitted to them | Registry history, deviation logs, results posting record |
该主题产生的产出
- Protocol templates stating hypotheses, controls, outcomes and analysis plans before data collection
- Registration records with time stamps and version history
- Deviation logs that date and explain every amendment
- Reporting checklists completed against the relevant EQUATOR guideline
What experts ask of any protocol.
- What exactly is the hypothesis, and what result would count against it?
- What is the comparison: a control group, a baseline period, a counterfactual model or a randomised arm?
- Which outcomes are primary, which are secondary, and how will each be measured?
- What sample, duration or number of runs is needed to detect an effect of the size that matters?
- What would make the team stop early, for safety or for futility, and who decides?
- What will be published if the result is null or the experiment fails?
Why experiments drift without a protocol.
Every experiment involves choices: which observations to exclude, which covariates to adjust for, which time window to use. Each choice may be reasonable alone, but when they are made after seeing the data, the number of possible analyses grows quickly and one of them will usually show something. This is the problem methodologists call analytic flexibility, and it is a central reason why published effects often shrink or vanish when repeated.
Interdisciplinary work adds its own drivers. Disciplines differ in what they treat as a control, how they handle missing data and what level of evidence they expect. A hydrologist, an economist and an epidemiologist may each consider a design rigorous by their own standard while disagreeing about what it shows. A protocol forces those differences into the open before work starts.
Incentives matter too. Positive, novel findings are easier to publish and fund, which rewards flexibility and penalises null results. Pre-registration and registered reports are the main structural responses.
Tools protocols draw on, and where they stop.
- 01
Pre-registration
A time-stamped record of hypotheses and analysis plans. It limits post hoc changes but cannot stop a poorly chosen question.
- 02
Power and design analysis
Estimates how much data is needed to detect a meaningful effect, using 概率与统计. It depends on assumed effect sizes that are often uncertain.
- 03
Causal designs
Randomisation, difference-in-differences, regression discontinuity and instrumental variables from 因果推断. Each rests on assumptions that must be stated and, where possible, tested.
- 04
Acceptance criteria
Pre-agreed thresholds for when a method, model or prototype is considered to work. They make success testable but can be gamed if set too loosely.
A protocol cannot make a bad measurement good or fix a biased sample. Deviations from protocol are permitted when justified, but are reported with their reasons.
Where this page’s references lead.
阅读本页时所对照的机构和文件。状态和版本会变化；每个链接都指向发布机构。
问题与解答。
What is the difference between a preregistration and a registered report?
Does preregistration forbid exploratory analysis?
Which reporting guideline applies to my study?
Can a protocol be changed after registration?
Is pre-registration required for every Labs study?
Can a protocol change during a study?
Who writes the protocol?
从本页贡献证据。
Reports is the Consortium’s quarterly publication. What you know about this subject can become evidence, a case or a correction in the next edition, credited to you and published under CC BY 4.0 where it is your own work.
- 电子邮件网关 GW-GCRI从本页发送证据、更正、案例、方法、审查、翻译或概况投稿。它会带上本页标签送达，由负责的编辑接收。
- Zenodo开放存缴存缴并提交到社区用您的 ORCID iD 在 Zenodo 上发布您自己的记录，然后提交到那里的 Reports 社区。由策展人审核；DOI 仍归您所有。
- 报告GCRI查看所有贡献方式贡献类型、各自的评审方式、您获得的署名、许可协议，以及绝不接受的内容。
切勿发送个人数据、敏感基础设施细节或您无权分享的任何内容。贡献在使用前会经过审查；发送贡献不赋予署名权或编辑权，赞助也绝不能换取研究结论。
法律文件。
本网站上的地图、数据和图表仅作为背景和一般信息展示。它们不是建议、预测或预警；信息的局限见免责声明和数据说明。