SD2 vs. Condorcet: Issues

denis bider yahoo at
Mon, 11 Apr 2005 12:30:01 +0200

> -M: Condorcet is still in-degree based, therefore it is still "Joe 
> Sixpack" based.

This is the core argument about which I am not yet quite sure.

See, suppose you have a problem with possible solutions A, B, C, D, and you
have the two approaches I stated in my previous email to this list:

1. Everyone ranks the four solutions in their order of preference, and the
solution that will yield the highest level of satisfaction is chosen
according to Condorcet / CSSD.

2. Everyone enumerates a number of people that they think are competent to
solve the problem, or know someone who is, and the results are ranked
according to SD-2 to produce a number of "experts". The selected few then
decide for either one of A, B, C or D.

Now, suppose that method 1 would show that the solution yielding the highest
level of people-satisfaction is C.

Suppose now that the experts also choose C. In this case, methods 1 and 2
are equivalent.

But suppose the experts choose D.

Question 1:

Is it acceptable for experts to choose a solution which is NOT a solution
which would yield the highest level of satisfaction among people; in this
case, method C?

Question 2:

But, then again: would method 1 have led to the choice of method C, if the
people had known that the experts would have chosen method D? Perhaps people
would then go with the opinion of the experts, and choose method D?

Question 3:

But even if the people had chosen method D if they knew that's what the
experts were going to choose: would that have been the correct choice? Is it
actually true, at all, that experts make decisions which are more correct
than the decisions made by masses when an appropriate aggregation algorithm
is chosen?

Research leads to believe the masses will be consistently correct much more
reliably than individual experts.

Research further leads to believe that a large, diverse group of experts
(not 5 directors, but 100 or more), whose choice is aggregated with a
suitable algorithm on an in-degree basis, might be more consistently correct
than a smaller number of top experts. This is because perception of
expertise in one's own and others' eyes tends to be unrealistically
exaggerated compared to the actual substance of the expertise. There is
further no reason to believe that, in additional iterations of SD-2, this
bias would be self-correcting. Quite the opposite is likely.

The source of this viewpoint is James Surowiecki, The Wisdom of Crowds.