现况研究生态学研究.ppt

Example An ecological study that showed a direct (positive) correlation between the percentage of the population that was Protestant and suicide rates in a number of Prussian communities in the late 19th century. Based on this observation, the inference is that being Protestant is a risk factor for suicide. Example The mistake is to use the correlation between the proportion of Protestants (which is an aggregate measure) and suicide rate to infer that the risk of suicide is higher in Protestant than in Catholic individuals. If one was to make an inference at population level, the conclusion that predominantly Protestant communities with Catholic minorities have higher rates of suicide would still be valid. Example A study based on individuals would ‘correctly’ find that the risk of suicide is higher in Catholics than in Protestants. This finding would logically suggest explanations of why the suicide rate differs between these religious groups. Is the higher rate in Catholics due to Catholicism per se? Alternatively, is it because of some ethnic difference between Catholics and Protestants? If so, is it due to some genetic component that distinguishes these ethnic groups? The problem is that these questions, which attempt to characterize risk at the individual level, though important, are insufficient to fully explain the web of causality, for they fail to consider the ecologic dimension of whether minority status explains and determines the increased risk of suicide. This example underscores the concept that both individual and ecological studies are necessary to study the complex causal determination not only of suicide but also of many other health and disease processes. Analytical challenges The combination of indvidual and ecologic levels of analysis poses analytical challenges for which statistical models (hierarchical models等级模型) have been developed. However, difficult conceptual challenges remain, such as the development of causal mo