Assessment of the extent of corroboration of an elaborate theory of a causal hypothesis using partial conjunctions of evidence factors (Q1996769)

This paper deals with the corroboration of causal hypotheses in observational studies. The running example is a classical problem in the epidemiological literature: ``Does the exposure to lead of a parent at the workplace causes high level of lead in the blood of a child at home?''. To corroborate the hypothesis one can define a set of statistical tests, the so-called elaborate theory, and combine all the \( p \)-values. However, when several tests are defined on the same experiment, the \( p \)-values are dependent. Moreover, unmeasured cofounding is a concern in observational studies. The authors provide here a quantitative analysis of the extent of corroboration of an elaborate theory, by decomposing the various tests into independent factors with different sources of potential biases. In addition, partial conjuctions of the tests are considered in order to quantify the evidence supporting some subsets of the collection of predictions. A simulation study is presented to highlight the main features of the proposed method.