I've got 2 identical experiments, each are however under-powered to detect the effects of interests.
I would like to combine the two datasets and re-run an analysis on the combined dataset.
Not surprisingly I got clearer results on the combined dataset.
However since this latter analysis consists in a second peek into my data, doing so increases the risk of false positives (or rate of type I error).
I would like to correct for this inflated risk but I could not find any correction method specific to the analysis of combined experimemnts/datasets.
In my design, I test two hypotheses, which consists in 2 planned t-test contrasts for each of 4 DVs.
What I am looking for is a (not too complicated) way to correct the p-values obtained from the contrasts or the critical p-value threshold.
Do you any suggestions?
False discovery rate, for instance. It is vey likely that your typical software has it already implemented.
A direct approach to your question would be to set up a random-simulation experiment in which your multi-stage inference procedure is adequately represented.
- Badges
- Science topic
- Similar topics
- Correction