I'm a social science person who used to work with hard-core data nerds (demographers), so this rings all too true. Nothing may be what it seems, including data: The New Yorker recently had a great piece about the "decline effect," which means that the strong results you see in an early trial or experiment become less strong over time (documented in a number of disciplines). Here's the abstract:
http://www.newyorker.com/reporting/2010/12/13/101213fa_fact_...