Data from: A problem of bias and response heterogeneity, in Standing With Giants: A Collection of Public Health Essays in Memoriam to Dr. Elizabeth M. Whelan

There is extensive literature on the question, “Does air quality have health effects?” For example, Google Scholar gives 199,000 hits for (“mortality” and “air pollution”). See Health Effects Institute (2010) and editorial by Brauer and Mancini (2014), for example. One paper that appeared in 1993 has over 6,000 citations. The vast majority of these papers find a positive association between air quality health effects (death). A few papers make the case that if potential bias is carefully taken into account then there is no association between air quality and deaths, e.g. Chay et al. (2003), Enstrom (2005), Janes et al. (2007), Greven et al. (2011), Cox et al. (2013). Clearly the weight of evidence is for a positive association, but for any particular type of claim, logically it takes only one true negative to negate all the positives associations with respect to causation for that claim. A real, causative claim should always be detected in a well-designed and properly run experiment. What are some of the factors that lead to these discordant literature results?

学界围绕"空气质量是否会对健康产生影响"这一问题已积累了大量研究文献。例如，以"死亡率"与"空气污染"为关键词在谷歌学术（Google Scholar）中可检索到199000条检索结果。例如可参考健康效应研究所（Health Effects Institute）2010年的相关报告，以及Brauer与Mancini于2014年发表的社论。1993年发表的一篇相关论文的被引次数超过6000次。绝大多数此类研究均证实，空气质量与健康效应（死亡风险）之间存在正向关联。但也有部分研究提出，若审慎考量潜在偏倚因素，则空气质量与死亡风险之间并无关联，例如Chay等人（2003）、Enstrom（2005）、Janes等人（2007）、Greven等人（2011）以及Cox等人（2013）的研究成果。显然，现有证据的整体倾向支持正向关联的结论，但从逻辑层面而言，针对某一特定因果主张，仅需一项确凿的阴性研究即可否定所有支持该因果关联的阳性研究结论。真正具备因果效力的主张，理应在设计严谨、实施规范的实验中得到验证。那么，究竟是哪些因素导致了这类研究结论的分歧呢？