Why would you choose endpoints that may (or may not) be correlated with the desired outcome, rather than the outcome you are interested in?

To rule out root causes. If the data showed increases and serotonin, decreases in stress hormones and improved sleep quality and the patient still replies "bad" to the question "How do you feel?" Then obviously a serotonin imbalance is not the problem (which is all SSRIs are designed to do).

If the patient still feels bad and the data shows no serotonin change, then the doctor can conclude that maybe serotonin is still the problem, but Zoloft isn't the right drug to fix it (or the dosage was wrong).

In other words, more data = better ability to get a real answer.

I don't think that it is really a 'debate'. There are different species concepts that are used depending on the needs of the situation (http://en.wikipedia.org/wiki/Species#Other_species_concepts). A single definition cannot be found for less divergent groups because speciation is really a continuous process. Also the mechanisms and course of the process differs between taxonomic groups and, in some cases, may never be 'complete'. The case of horizontal gene transfer mentioned below occurs in eukaryotes as well (http://en.wikipedia.org/wiki/Horizontal_gene_transfer#Eukary...).

For some information on the scope of the problem, I found this report useful. http://www.cdc.gov/drugresistance/threat-report-2013/

Frontline did an episode that covers resistance and the lack of new antibiotic drug development. http://www.pbs.org/wgbh/pages/frontline/hunting-the-nightmar...