Quantitative Data Cleaning for Large Databases [pdf]

Have there been any black hat/white hat discussions of these sorts of techniques? I.e., how to detect intentional biases introduced into a dataset, faked lab data?

Most of what I remember about this is from the book "Bad Science" by Ben Goldacre. Comparing different studies against one another is the gold standard for determining a study's reliability, but of course there isn't always the luxury of having an independent dataset for comparison.

Assuming the data-faking party hasn't been completely stupid and has produced fake data following the expected distribution around each data point, the only thing that comes to mind off hand is Benford's Law: http://en.wikipedia.org/wiki/Benford's_law

Why does HN create a Scribd link for every PDF posted?