pamela
05-13-2011, 10:16 PM
I have results of bernoulli trials with two "control variables", e.g. age group and type of medication. For each combination the number of trials and number of successes are known. I suspect that correlation exists between my control variables (different medications are effective for different age groups).
My goal is to estimate the p value for a specific age group and medication.
The trouble is, for a specific combination of these two variables, my n is
too low to give an accurate estimate. My idea is to obtain estimates
based on single control variables as well as an estimate based on the combination and combine them.
The number of trials in the two-control-variable estimator is obviously lower
(higher variance???) than in the one-variable estimators.
On the other hand, the two-variable estimator would clearly be my best bet if it only had a reasonable number of trials...
This is where I am lost.
How would you combine these three estimators?
My goal is to estimate the p value for a specific age group and medication.
The trouble is, for a specific combination of these two variables, my n is
too low to give an accurate estimate. My idea is to obtain estimates
based on single control variables as well as an estimate based on the combination and combine them.
The number of trials in the two-control-variable estimator is obviously lower
(higher variance???) than in the one-variable estimators.
On the other hand, the two-variable estimator would clearly be my best bet if it only had a reasonable number of trials...
This is where I am lost.
How would you combine these three estimators?