I would like to test whether there is a significant difference in outcome in a pre/post scenario.

I have the following set-up:

Measured level of concentration of A (continuous) in blood pre-death and measured concentration of A post-death for each individual (roughly 100 individuals in total).

Working hypothesis: After death, the concentration of A falls and hence should be on average lower then pre-death concentration.

So, my first impulse was to go for a paired-t-test given all assumptions are fulfilled. However, my problem arises from the fact that the post-death concentration of A is not always measured at the same time after the death has occurred. So maybe 1h after death, sometimes 12h later and so on. I wonder if this causes any problems and whether I can somehow factor in the difference in post-death time-to-test. (There are also other parameters which change on an individual basis: age, concentration, time, sex, cause of death; they are all categorical)

Is there any alternative approach to this or is the "significant difference" approach, given so many loose ends, just not reasonable at all? In the paired t-test I would just ignore all other parameters, though they might affect the concentration of A as well.

Another idea would be to use a one-way MANOVA with the dependent variables "pre-death_A" and "post-death_A" and the independent variables age, concentration, time, sex, cause of death.

I have never done this before so I am a bit hesitant. Any other suggestions?


I would analyze the data a bit more before forming the hypothesis; run a correlation and a multiple logistic regression on time after death of sample vs. concentration (including age, etc.). If time after death matters little, make your hypothesis simple: pre vs. post. If time after death does matter, then you'll want to form your hypothesis differently. You may need to set aside a training set & re-validate using the other set to avoid "torturing the data until it confesses" so to speak.

  • $\begingroup$ Yeah, maybe that's really a good idea to first figure out whether time is a relevant factor or can be dropped. I'll definitely look into it... $\endgroup$
    – user190080
    Oct 19 '20 at 20:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.