How to select a control / comparison group? Lets say I'm creating a study that looked at treatments for metastasis in African American colon cancer patients. Would my control group or comparison group be Caucasian colon cancer patients with metastasis? I'm a bit confused about how to create control groups in this context. 
 A: What your control group should be is largely a substantive / theoretical question, not a statistical question.  That is, your control group should be designed to help you answer the theoretical question that you are conducting your study to answer.  
I can imagine several different control groups in your case, but which you should choose is something only you can decide.  Here are some examples:  


*

*Caucasian colon cancer patients with metastasis who are given the treatment.  This design would allow you to isolate the effect of being an African American (vs. Caucasian) with metastasis who is getting the treatment.  

*African American colon cancer patients with metastasis who get placebo treatment.  This design would allow you to isolate the effect of getting the treatment (vs. not getting the treatment) for African Americans with metastasis.  (Since wondering if treatments are effective is a very common question in biomedical science, this setup is probably more typical, but that would only be because that question is more typical, not because this setup is inherently correct irregardless of your question.)  

*You could also include both control groups.  Which would allow you to answer both questions, although at the expense of requiring more data, and the statistical analyses to answer the questions would not be independent (the theoretical questions can be conceived of as independent, though).  

*A final possibility is to have four groups: race (African American vs. Caucasian) X treatment (active vs. placebo).  This would allow you to answer the above two questions independently (from a statistical perspective), and would also allow you to answer if the effect of the treatment depends on race (i.e., if an interaction exists).  


In fact there can be many more possibilities available, depending on the question you want to ask.  For instance, you could compare African Americans with metastasis on the treatment to African Americans (also with colon cancer but) without metastasis on the treatment, etc.  But in all cases, your substantive question comes first, and you design your experiment so as to best answer that question.  
A: Control has to be African American colon cancer patients who get placebo treatment. That is control should be as similar to the other group as possible, except that it's given a fake treatment.
