Sure distribution can have issues, but do you have any references for simple possession as training and test data?
But if the data/analysis is published, then the data source would need to be disclosed.
See more on the OKCupid case I mentioned above: http://www.vox.com/platform/amp/2016/5/12/11666116/70000-okc...
Sure distribution can have issues, but do you have any references for simple possession as training and test data?