I would like to calculate the minimal number of participants (sample size) I need for my experimental protocol. I use logistic regression. I compare two groups in a protocol having three independent variables (each one having two modalities) and one dependent binary variable.
For each experimental condition (e.g., group 1, modality 1 of independent variable 1, modality 1 of independent variable 2, modality 1 of independent variable 3), participants are tested in 9 items (each item is an observation, I guess).
It is worth noticing that the probability of having a 1 (success) is much higher than having 0 in both groups but, according to my hypothesis, I should have more 0s in a group than in the other (even though less 0s than 1s in both).
I know that I need at least 10 observations for each factor when I use a logistic regression, but on the one hand I am not sure what an observation is (number of participants, number of items, or number of participants*number of items, etc.). On the other hand, I don't know if the highest probability of having 1 over 0 influences this estimation, since the logistic regression works better with around 50/50 distribution of zeros (which I most likely won't have as I said previously).
Do you have any idea of the amount of participants I need for condition and/or how am I supposed to estimate it?
