I would like to compare two experimental quantities in two conditions.

I have two replicated experiments A and B in control condition, where total number of replications is $N$. Number of successes in the experiment A is $m_{control}$, number of successes in experiment B is $k_{control}$. The frequency of successes for the first experiment A is $q_{control}=m_{control}/N$ and for the experiment B is $p_{control}=k_{control}/N.$

I have another set of experiments in different condition, where the number of replication is preserved:N. The only things that change is number of successes, therefore we have $q_{cond1}=m_{cond1}/N$ and $p_{cond1}=k_{cond1}/N$.

I would like to test if the following qualities statistically differ between control and "condition" group of experiments: $q_{control}/p_{control}$ vs $q_{cond1}/p_{cond1}$. Which test you would suggest in this case?

PS. The experimental observations are independent and $N>100$.


1 Answer 1


Given a flat prior, the posterior probability distribution for $q_{control}$ is $Beta(1+m_{control}, 1+(N-m_{control})$ (conjugate distribution, cf https://en.wikipedia.org/wiki/Conjugate_prior#Discrete_distributions ). Thus it is easy to draw a large number of candidates for $q_{control}$, $q_{cond1}$, $p_{control}$ and $p_{cond1}$ and simply count, how often one ratio is larger then the other. This is one of the examples where Bayesian flexibility makes things actually easy.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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