Re: st: Fraud methods in Stata

[Nick Cox <n.j.cox@stata.com>]

I would search for publications of longstanding Stata user Stephen Evans 
in this area. He has done very serious work on (possibly fraudulent) 
medical data.

That said, I remain puzzled by the implication that outliers are prima 
facie evidence of fraud. My own impression is that fraudulent people 
wish to create datsets that look genuine and that they are thus unlikely 
to add or manufacture outliers, unless those outliers serve their 
purpose somehow, but that's just a guess. The main ways in which I can 
think of that fraudulent data can sometimes be identified is that often 
agreement is "too good to be true" and through looking at the patterns 
of first and last digits in data. Another obviously related issue is 
plagiarism of published data.

Nick
n.j.cox@durham.ac.uk

Williams, Rachael wrote:


> I am considering methods of detecting fraud in a hypothetical clinical
> trial with a large number of centres, but only a few patients per
> centre.
> In addition, many variables will be binary.
>
> Would Cook's D be appropriate here?
> Is it possible to calculate Mahalanobis' distance in Stata in order to
> detect (possibly fraudulent) inliers, outliers and near duplicates in a
> dataset?
>
> If anyone has any ideas of other ways to detect possible fraud I would
> love to hear from you too!
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/