Re: [ALSC-Forum] Draft report - statistics and "fraud"

[Kent Crispin]

On Wed, Aug 29, 2001 at 07:23:54PM +0900, Adam PEAKE wrote:
> 
> 
> 
> Mike Roberts wrote:
> > 
> > At 0:16 +0900 8/29/01, Adam PEAKE wrote:
>  
>  snip 
> 
> > >
> > >As your draft explains, "9,721 pin letters were returned (as of August
> > >19, 2001), 92.2% of which had been sent to countries in the Asia region
> > >... raises questions about the efficacy of postal mail" These 9,721 are
> > >included in the 33,043 registrations you suggest may be fraudulent.
> > >Since when was "return to sender" and indication of fraud?
> > 
> > Adam - this part of your note is incorrect.  No one had a pin letter
> > mailed to them if they were in a suspected fraud status.  Those 9,721
> > individuals, and others who didn't get their letters but ICANN did
> > not get them returned, are part of the "registered but did not
> > activate" group, which totaled
> > 67,230 persons.  (143,806 - 76,183).
> > 
> 
> Mike, 
> 
> The 33,043 registrations removed from the database included 10,334

Clarification: they were not removed from the database -- they were
simply flagged as invalid registrations for the reason given.

> letter bounces, <http://www.atlargestudy.org/stats/summary1.shtml>. What
> does "letter bounces" mean, and why were they placed in the suspected
> fraud status?

It might be better to label them as "possibly fraudulent".  It was
probably a mistake to try to assign a number to the "suspected
fraudulent" registrations, because in fact there is no way to know --
"suspected fraud" is an intrinsically subjective judgement. 

In the case of duplicate registrations, one of the duplicates remained
unflagged, and consequently, to the extent that duplicates were
indicative of fraud, then fraudulent registrations were explicitly
permitted. 

In fact, in my experience, almost all of the cases where fraud was
suspected were simply allowed, because there was no realistic way to
check, and one could always imagine possible scenarios where the odd
behavior in fact might be OK.  For example, if you had 5 different email
addresses going to the same physical address, with slightly different
names, that could be a family, or it could be someone "testing the
system".  There is no way to tell; so in almost all cases, the
permissive view prevailed.

It is also important to bear in mind that the human processes were just
as overwhelmed as the server was.  There was simply no way to examine
every record for possible fraud; the 580 records in administrative hold
status each required someone looking at the case, thinking about it, and
making a decision -- if you spent say 10 minutes each, that is 5800
minutes, or 2 and a half full-time person-weeks, just looking at those
records. 



> I was guessing that the returned "9,721 pin letters" the ALSC report
> referred to were among of the 10,334 "letter bounces." Incorrect guess?
> 
> Other records were removed from the database (for suspected or real
> fraud) for reasons described as email bounces, duplicates entries,
> administrative holds, and bad format. For the community and study
> committee to be able to consider future election mechanisms we really
> must be able to understand if fraud, reasonably suspected or proven,
> occurred, and occurred in what circumstances. To simply brand these
> 33,043 registrations as fraudulent is just plain wrong. 

You are whacking hard at a tree, and ignoring the forest.  The wording
may not have been excellent, but the fact is that the processes were
inadequate to deal with fraud, and some significant amount of fraud was
possible (that isn't as strong as "suspected", but it's close).

But far more important: not only is it the case that the infrastucture
to hold what one would normally consider an election with reasonable
integrity *didn't* exist, it still doesn't exist, and there is no 
realistic way to make it exist.

Kent

-- 
Kent Crispin                               "Be good, and you will be
kent@songbird.com                           lonesome." -- Mark Twain