PROJECT AIMS
Aim | 01
Aim | 01 Mediation with sparse data
Among other developments in the field of causal inference, mediation analysis became a major focus of scientific literature in the last decade. The ultimate aim of mediation analysis is to decompose a marginal effect into direct and indirect components.
While much of recent literature is devoted to structural problems many applied researchers are still faced with classical statistical problems surrounding these new methods. Bias arising from sparse data (e.g. from analyses with very unbalanced categorical variables) is one such problem, that is also the focus of our first research aim.
Aim | 02
AIM | 02 Cox and Aalen models In practice
The most commonly used method to analyise time-to-event data is undoubtedly Cox`s proportional hazard model. It`s popularity is rooted in its compact outcome measure - supplying hazard ratios as effect sizes - that can seemingly be interpreted like relative risks.
However with the shift towards more causal thinking in statistics also concerns regarding the causal interpretation of the hazard ratio have been raised. An alternative - Aalen`s additive hazards model - has proven particularly advantageous in the context of causal inference. However results obtained from an additive model could be more difficult to interpret for practitioners.