随机抽样：因果视角

牟育彤; 郭阳; 李青; 张淼; 索晨; 李亚欣; 刘海燕; 胡安群; 郑英杰

doi:10.16462/j.cnki.zhjbkz.2023.08.020

随机抽样：因果视角

doi: 10.16462/j.cnki.zhjbkz.2023.08.020

1.
复旦大学公共卫生学院流行病学教研室，国家卫生健康委员会卫生技术评估重点实验室(复旦大学)，上海 200032
2.
北京大学深圳医院，深圳北京大学香港科技大学医学中心，深圳 518036
3.
安徽医科大学附属安庆医院，安庆市立医院，安庆 246003

基金项目:

国家自然科学基金 82173582

国家自然科学基金 81373065

国家自然科学基金 81773490

详细信息

通讯作者:
郑英杰，E-mail: yjzheng@fudan.edu.cn; yjzheng@fudan.edu.cn

中图分类号: R780.1;R543.4
计量
- 文章访问数: 223
- HTML全文浏览量: 106
- PDF下载量: 30
- 被引次数: 0
出版历程
- 收稿日期: 2021-09-08
- 修回日期: 2022-03-12
- 网络出版日期: 2023-09-02
- 刊出日期: 2023-08-10

Random sampling: a causal perspective

1.
Department of Epidemiology, Key Laboratory for Health Technology Assessment, National Health Commission of the people′s Republic of China, School of Public Health, Fudan University, Shanghai 200032, China
2.
Peking University Shenzhen Hospital, Shenzhen Peking University-The Hong Kong University of Science and Technology Medical Center, Shenzhen 518036, China
3.
Affiliated Anqing Hospital of Anhui Medical University, Anqing Municipal Hospital, Anqing 246003, China

Funds:

National Natural Science Foundation of China 82173582

National Natural Science Foundation of China 81373065

National Natural Science Foundation of China 81773490

More Information

Corresponding author: ZHENG Yingjie, E-mail: yjzheng@fudan.edu.cn; yjzheng@fudan.edu.cn

摘要

摘要: 本研究从因果图和因果模型的角度解释了简单随机抽样、整群抽样和分层抽样的结构和原理，清晰呈现了抽样偏倚的产生和调整过程。发现当研究的目标属性与是否抽中有关联时，样本对总体的代表性将产生偏差，即抽样偏倚或抽样的系统误差。整群抽样和分层抽样是两种对抽样偏倚的不同调整策略; 前者找到一个目标无关变量作为分群变量，而后者直接调整了引起抽样偏倚且与研究属性关联的变量。在抽样方法的选择上，整群抽样要求群内异质且群间同质，分层抽样则正好相反。最后，本研究为实际研究工作中对抽样方法的选择提供了路线参考，可帮助读者提高对随机抽样的认识，考量研究方法的选择。
- 抽样 /
- 因果图 /
- 抽样偏倚 /
- 分层抽样 /
- 整群抽样
Abstract: This article explained the structure and principle of simple random sampling, cluster sampling, and stratification sampling from the perspective of causal graphs. The process of generation and adjustment of sampling bias was clearly presented. We found that: when the target attribute of the study was related to other attributes which affected the selection of the samples, the representative of the sample would be biased. In other words, the sampling bias or the systematic error of sampling would appear. When selecting sampling methods, cluster sampling requires heterogeneity within clusters but homogeneity among clusters. In contrast, stratified sampling requires the opposite way. Finally, this paper provides a route for the selection of sampling methods in practical research work, which can help readers to improve the acknowledgment of random sampling and consider the best sampling method.
- Sampling /
- Causal graphs /
- Sampling bias /
- Stratification sampling /
- Cluster sampling

HTML全文

图 1 抽样及抽样偏倚因果结构示意图

Figure 1. Causal graphs of sampling and sampling bias structures

下载: 全尺寸图片幻灯片

图 2 随机抽样路线示意图

Figure 2. Schematic diagram of random sampling route

下载: 全尺寸图片幻灯片

参考文献(9)

[1]	Acharya AS, Prakash A, Saxena P, et al. Sampling: why and how of it?[J]. Indian J Med Spec, 2013, 4(2): 330-333. DOI: 10.7713/ijms.2013.0032.
[2]	Jun SJ, Lee S. Causal inference under outcome-based sampling with monotonicity assumptions[M]. arXiv preprint, 2020: 1-169.
[3]	Gabriel EE, Sachs MC, Sjölander A. Causal bounds for outcome-dependent sampling in observational studies[J]. J Am Stat Assoc, 2020, 117(538): 939-950. DOI: 10.1080/01621459.2020.1832502.
[4]	Hernan MA, Robins JM. Causal Inference: What If[M]. Boca Raton: CRC Press, 2020: 1-389.
[5]	Pearl J. Causality: Models, reasoning, and inference[M]. Cambridge University Press, 2009: 1-477.
[6]	郑英杰, 赵耐青. 有向无环图: 语言、规则及应用[J]. 中华流行病学杂志, 2017, 38(8): 1140-1144. DOI: 10.3760/cma.j.issn.0254-6450.2018.01.019. Zheng YJ, Zhao NQ. Directed acyclic graphs: languages, rules and applications[J]. Chin J Epidemiol, 2017, 38(8): 1140-1144. DOI: 10.3760/cma.j.issn.0254-6450.2018.01.019.
[7]	Cochran WG. Sampling techniques[M]. New York: John Wiley & Sons Inc, 1977: 1-11.
[8]	Kish L. Survey Sampling[M]. New York: John Wiley & Sons, Inc, 1965: 1-29.
[9]	Neyman J. Contribution to the theory of sampling human populations[J]. J Am Stat Assoc, 1938, 33(201): 101-116. DOI: 10.1080/01621459.1938.10503378.