1
$\begingroup$

I'm not sure if this is those appropriate way to phrase this question. I have two populations of measurements. In my control population I have a quantitative measure of a proteins abundance. I expect some measurements that were originally relatively abundant in my control population to be missing in my treatment sample. However, some measurements will be missing in my treatment and control due to the stochastic appearance/disappearance of low abundance measurements in both populations.

Based on what I have been able to read. I want to know which values are missing not at random, or MNAR to use the nomenclature from Rubin 1976. Or is this inappropriate here, and I should be using some other model that can take into account the fact that low abundance measurements will naturally be stochastically measured, like left censoring? Is this just a sensitivity threshold problem?

Ideally I would like a method to calculate the p value that a measurement is missing due to the treatment and not due to the stochastic detection of low abundance measurements. Would this be the P(MNAR | Data) ?

I was wondering if this is a similar case of this question that was asked:

Simple case of MNAR missing data

I am an experimental biologist, but am comfortable with R, and was wondering if anyone could help determine if this is the same / a similar problem or a completely different problem.

Thanks!

Improve this question
$\endgroup$
1
  • $\begingroup$ If you know the missing values are missing because they're below some threshold, you could certainly treat it as censored (as long as you know that threshold for each observation -- it needn't be the same). $\endgroup$
    – Glen_b
    Commented Oct 7, 2014 at 8:53

0

Reset to default

Your Answer

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