A colleague has a dataset from a before and after study looking at a continuous outcome that was measured in individuals before and after an intervention. An unknown but not insubstantial proportion of the individuals were only measured before or after the intervention.

However, due to an oversight following changes to the study design due to ethics they have only now realised that there is no way to link the before and after measurements that were made to the same individuals, i.e. there is no person ID.

Consequently, the data consists of a mix of paired and independent samples, but can only be treated as independent samples.

Are there any reasonable ways to mitigate this problem or address it, e.g. via some kind of sensitivity analysis?


marked as duplicate by kjetil b halvorsen, user158565, Peter Flom Jan 17 at 12:01

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • $\begingroup$ I am just now looking into the same problem with a dataset that I have! It is complicated, there are some articles about similar problems, search google scholar for "repairing broken sample" or "file linkage problem". $\endgroup$ – kjetil b halvorsen Jun 1 '16 at 15:17