People get excited about conclusions drawn from statistical analyses, but what if the underlying (raw) data is bad...

What if it doesn't accurately represent the phenomenon we are trying to model?

What then?