Bocquier Philippe, Sankoh Osman, Byass Peter
a Centre de recherche en démographie , Université catholique de Louvain , Louvain-la-Neuve , Belgium.
b School of Public Health, Faculty of Health Sciences , University of the Witwatersrand , Johannesburg , South Africa.
Glob Health Action. 2017;10(1):1356621. doi: 10.1080/16549716.2017.1356621.
Sampling rules do not apply in a Health and Demographic Surveillance System (HDSS) that covers exhaustively a district-level population and is not meant to be representative of a national population. We highlight the advantages of HDSS data for causal analysis and identify in the literature the principles of conditional generalisation that best apply to HDSS. A probabilistic view on HDSS data is still justified by the need to model complex causal inference. Accounting for contextual knowledge, reducing omitted-variable bias, detailing order of events, and high statistical power brings credence to HDSS data. Generalisation of causal mechanisms identified in HDSS data is consolidated through systematic comparison and triangulation with national or international data.
抽样规则不适用于全面覆盖一个地区级人口且无意代表全国人口的健康与人口监测系统(HDSS)。我们强调了HDSS数据在因果分析方面的优势,并在文献中确定了最适用于HDSS的条件性概括原则。由于需要对复杂的因果推断进行建模,对HDSS数据采用概率观点仍然是合理的。考虑背景知识、减少遗漏变量偏差、详细说明事件顺序以及具备高统计效力,这些都增强了HDSS数据的可信度。通过与国家或国际数据进行系统比较和三角测量,巩固了在HDSS数据中识别出的因果机制的概括。