Models of causal inference: critical analysis of the use of statistics in epidemiology

Abstract

The foundations on which the concept of risk has been constructed are discussed. A description of Rubin's model of causal inference, which was first developed in the domain of applied statistics, and later incorporated into a branch of epidemiology, is taken as the starting point. Analysis of the premisses of causal inference brings to light the logical stages in the construction of the concept of risk, allowing it to be understood "from the inside". The abovementioned branch of statistics and epidemiology seeks to demonstrate that statistics can infer causality instead of simply revealing statistical associations; the model gives the basis for estimating that which way be defined as the effect of a cause. Using this procedural distinction between causal inference and association, the model also seeks to differentiate between the epidemiologial dimension of concepts and the merely statistical dimension. This leads to greater complexity when handing the concepts of interation and coofounding. The redective aspects inherent in this methodological construction of risk are here high lighted. Thus, whether applied to individual or populational inferences, this methodological construction imposes limits that need to be taken into account in its theoretical and practical application to epidemiology.