Aggregating observations with missing into group counts to compute treatment effects I have observations from $n$ groups where half of the groups were assigned to a treatment and the rest to a control condition. 
As observations within groups are dependent, I aggregate data into group counts and want to see to test treatment effects across groups. 
For variables with complete cases, this seems to work just fine. Problems arise when I have missing data within a group and I want to aggregate them into counts. Do I still get an estimate of treatment effects if I simply ignore the missing values? Or rather, under what conditions identification will happen?
 A: There are many methods for doing this, it depends a lot on the kind of variables you have.
First of all I suggest to explore the reasons why the cases are incomplete, if it is for example due to equipment malfunction you might want to interpolate or estimate them from the complete cases e.g. a sensor that was disconnected from the network. On the other hand the fact that the value is missing might be meaningful for and you might want to use it for the analysis e.g. It was not possible to measure some variable on a patient due to and underlying condition. For the second case, you might find a confounding variable that makes the case to be incomplete.
If you are interested on completing each case I suggest you reading on missing value imputation methods here are a few examples:


*

*mean

*regression models

*K nearest neighborhood imputation
As I said, there are many methods and it depends a lot on the phenomena you are studying, can you give us more information so we can make some suggestions?
To answer your question, yes, if you omit them you would get an estimate but it will be statistically less significant than if you were able to use all the cases. I think it is a good idea to try imputing the missing values or alternatively, create a metric that takes into account the number of incomplete cases.
