9.6 Conclusion

Finally some thoughts about causality, and the techniques in use for “causal inference”. As mentioned above, in order to claim causality, one must show (or assume) the there are “no unmeasured confounders”. Unfortunately, this is impossible to prove or show from data alone, but even worse is the fact that in practice, at least in demographic and epidemiological applications, there are always unmeasured confounders present. However, with this in mind, note that

• Causal thinking is important,
• Counterfactual reasoning and marginal models yield little insight into “how it works”, but it is a way of reasoning around research problems that helps sorting out thoughts.
  • Joint modeling is the alternative.
• Creation of pseudo-populations through weighting and matching may limit the understanding of how things work.
  • Analyze the process as it presents itself, so that it is easier to generalize findings.

Read more about this in Aalen, Borgan, and Gjessing (2008).

References

Aalen, O. O., Ø. Borgan, and H. K. Gjessing. 2008. Survival and Event History Analysis: A Process Point of View. New York: Springer.