[EM] Single-winner election data from the OpenSTV database

Steve Eppley SEppley at alumni.caltech.edu
Wed Dec 3 06:34:07 PST 2008


I think it's misleading to compare voting methods pseudo-empirically the 
way Greg did below, because the positions on the issues that the 
candidates will take (assuming they want to win) depend (in part) on the 
voting method, and the decisions by potential candidates whether to run 
depend (in part) on the voting method. 

It's not surprising that poor methods like plurality rule and single 
winner STV (a.k.a. Instant Runoff) appear to usually elect Condorcet 
winners given the overly simplistic analysis Greg cited, because with 
those methods the elites try hard to organize into 2 large coalitions 
that each nominate only one candidate.

Voting methods that make centrists appear relatively unpopular 
(plurality rule, single winner STV a.k.a. Instant Runoff, etc.) not only 
discourage politicians from taking centrist positions, they discourage 
centrist candidates from competing.  For example, recall John McCain's 
decision not to compete as an independent in the 2000 general election 
after the Republican Party nominated George W Bush.  Under plurality 
rule, McCain running as an independent would have cinched the election 
for Al Gore.  Under Instant Runoff, it probably wouldn't have changed 
the outcome, but McCain would likely have been eliminated before Gore 
and Bush and therefore would have appeared relatively unpopular, so why 
bother running, and why would donors bother contributing campaign funds 
to a likely loser?  Under a Condorcetian voting method, McCain would 
have won. (Of course, if the voting method had really been Condorcetian, 
McCain would have faced different competition from candidates trying to 
be the best centrist compromise.)

For another way that the voting method can affect who runs and the 
positions taken by candidates trying to win, consider methods like 
Voting for a Published Ranking (VPR).  VPR can be expected to greatly 
reduce the campaign money needed by good centrist candidates to win, 
which would change who runs, their positions on the issues, and who wins.

Here's a quote from page 89 of Peter Ordeshook's book Game Theory and 
Political Theory:

     "... since it seems reasonable to require that democratic voting 
     elect Condorcet winners if they exist, scholars have proposed a 
variety of
     voting schemes as alternatives to plurality rule.  But the problem 
with most
     analyses of these alternatives is that they take inadequate account 
of the
     possibility that voters might misrepresent their preferences and 
     might change their campaigns under different rules."

Ordeshook could have added that potential candidates might make 
different decisions about whether to run, and that donors might make 
different decisions about how much to donate and to whom.  He could also 
have added that, by the same reasoning that it seems reasonable to elect 
a Condorcet winner if it exists, it also seems reasonable to elect a 
member of the top cycle (a.k.a. Smith set).

Greg wrote:
> There are 65 real single winner elections in OpenSTV Database
> (stv.sourceforge.net/stvdb).While the database doesn't list the
> individual ballots, it does list the Smith Set for each of these
> races. In 63 of these 65 races, there exists a dominant Condorcet
> winner (the Smith set has a size of 1).
> For 61 of the 65 races, the database also includes the results of
> applying other single-winner methods to the ballots, including IRV,
> Coombs, Borda, Supplementary Vote, Plurality, and Bucklin. 60 of these
> 61 have a Condorcet winner. Here is the Smith efficiency (frequency of
> electing a candidate from the Smith Set) for each of the applied
> methods, from best to worst:
> IRV, Coombs: 100% Smith Efficiency (61 of 61 races)
> Borda, Supplementary Vote: 96.7% Smith Efficiency (59 of 61 races)
> Plurality: 95.1% Smith Efficiency (58 of 61 races)
> Bucklin Voting: 91.8% Smith Efficiency (56 out of 61 races)
> - Greg

More information about the Election-Methods mailing list