返回提示词目录第 1444 页

核查知名研究结论的复现证据

帮你梳理原始研究、复现实验与元分析,形成证据对照表及符合当前研究支持程度的表述。

完整提示词

<context>
Many famous findings, especially in social psychology, nutrition and some areas of biomedicine, failed when tested again with larger samples and pre-registered methods. The strongest evidence comes from direct, pre-registered replications with large samples and from multi-lab projects (such as the Many Labs projects and Registered Replication Reports), because they remove the flexibility that inflates original effects. Meta-analyses of the original literature can mislead if they pool small, selectively published studies; look for ones that test for publication bias and report a corrected estimate. A finding can be partly true: the effect may exist but be far smaller than first claimed, hold only under narrow conditions, or survive for one outcome (feeling more powerful) but not another (hormone changes). Popular retellings often describe the original, larger claim long after researchers revised it.
</context>

<task>
Check the replication record of this finding.
<finding>
{{finding}}
</finding>
{{#use}}Planned use: {{use}}{{/use}}

1. Identify the original study or studies: authors, year, design, sample size and the exact effect reported. Separate the original claim from the popular version.
2. Search for direct and conceptual replications, multi-lab or registered replication projects, and meta-analyses, noting for each the sample size, pre-registration, result and effect size.
3. Weigh them: large pre-registered direct replications count most; small conceptual replications count least; meta-analyses count for more when they correct for publication bias.
4. Note any later debate: original authors' responses, moderators proposed, and whether the field's view has settled.
5. Give a verdict and wording that matches the current evidence for the planned use.
</task>

<constraints>
- Cite only studies and projects you opened or found records of in this session, with links. Never invent a replication, sample size or effect size.
- If you have no web access, say so at the top, give only a provisional view marked as such for well-known cases, name no specific figures, and list what to search.
- Distinguish "failed to replicate" from "disproved": a null replication with a wide interval may still be compatible with a small effect. Say which applies.
- Do not judge the original researchers' honesty. Describe methods, not motives.
- If the finding has replicated well, say so plainly.
</constraints>

<output_format>
## Verdict
One of: replicated well, replicated with a smaller effect, mixed or contested, did not replicate, or not yet tested. Then two sentences.
## The original finding
What was claimed, by whom, with what design and sample, and how the popular version differs.
## Replication record
Table: study or project (linked) | year | type (direct, conceptual, multi-lab, meta-analysis) | sample | pre-registered | result.
## Why it may not have held up
Short bullets, or "Not applicable" if it held up.
## How to describe it now
One or two sentences ready to use.
</output_format>

可替换内容

  • {{finding}}
  • {{#use}}
  • {{use}}
  • {{/use}}

分类与标签

学习教育、科研复现、文献核查、证据评价、科学传播

前往 AI 图片创作