In Instant Runoff Voting, it sounds like it would be a good idea to publish
all ballots, so that the election could be studied, reconstructed,
and re-counted by whoever wanted to.
Unfortunately, that's a bad idea.
The reason is this.
Suppose there were 20 candidates.
Note that 17! far exceeds
world population.
Then IRV ballot-publication opens to door to vote-buying and coercion.
To coerce you to vote for Evil, and/or buy your vote, I proceed thusly:
Then I'll look at the list of published ballots afterwards. If I do not see
the exact ordering I specified as a ballot
on that list, I'm going to have you fired. If I do see it,
then you get a dollar. Capisce?"
Actually, I just said "20 candidates" to be generous/safe. The argument actually works
in most elections with "10 candidates."
The 2002 French Presidential election had 16.
The Congo 2006 presidential election had at least 30.
Is this just a flaw in IRV? Or in every voting system? What?
It is just a flaw in IRV. Many other rival voting systems do not suffer this flaw.
More precisely, this flaw seems to affect every voting system in which both
(a) there are
no such things as "precinct subtotals," and
(b) the number of possible legal votes you could
cast, is large.
With plain plurality voting, it is no problem to publish every vote in a large election.
With other voting systems like Approval and Range, while precincts probably cannot
safely publish all ballots, they can at least publish that precinct's subtotals
with essentially no risk of breaking vote privacy.
But with IRV, there is
no such thing
as a "precinct subtotal" hence no possibility for precincts to publish them.
And publishing all IRV ballots is basically just government assistance to vote-buyers
and coercers – if the buyers had to make a wish for what they wanted the government to
do to help them maximally, this would be that wish.
IRV could avoid this by restricting the number of candidates in a race to be small
(say at most 4 candidates permitted) – but that would mean IRV was not really a voting
system, in my eyes, at all. IRV also could avoid it by having voters rank only
(say) the top three candidates, not all N. That is what was done in San Francisco 2004.
As a result,
no IRV winner got a majority of the votes unless they also were a plain-plurality
majority winner,
completely belying the claim (which was silly anyway)
that IRV "elects majority winners" which is its "advantage" over plain plurality voting.
(In most parts of Australia, IRV voters are required to provide a full ranking.)
Why is this bad?
It means, in operation,
IRV is inherently less transparent and less-checkable than other
voting systems.
Should we really worry about this flaw?
That depends:
Do you think election transparency and verifiability are good and important? (I do.)
Do you think vote-buying and/or vote-coercion are bad and important?
(I haven't encountered anybody who didn't think vote-coercion was bad, but I
have encountered those who dispute the badness of vote-buying.
Vote-buying definitely
has been important throughout much of US history, although it might not be at the present time.
See also this.
And re coercion, the battle between the mining
tycoons William A. Clark and Marcus Daly
in Montana in the 1890s involved lots of vote-coercion of the form "vote for me or
lose your mining job" e.g. see
Michael P. Malone: The Battle for Butte:
Mining and Politics on the Northern Frontier 1864-1906, Univ. of Washington
Press 1981 & 2nd ed. 2006.
In 2000, Jon Corzine spent $60.2 million – $35 million of it his own –
to win a New Jersey senate seat,
which since he got 1.51 million votes, averages out to $40 per vote.
We are not saying this to accuse Corzine of "buying votes"; it is simply to give you an idea
of how much those votes were worth, and to allow you to contrast it with the much
smaller expense
of keeping track of every seller's special vote-ordering on the vote-buyer's computer.
And in some elections like "county highway commissioner" votes are probably worth more than
that to a corrupt candidate.)
Do you think any such vote-buying would not be a problem because the buyers would easily be caught?
(Well, if so, congratulations. You've just proved that Boss Tweed never existed.
Get it: when the corruptors own the cops and the judges and the prosecutors,
the obviousness and immensity of voting fraud is not a problem. Of course you could argue that
Boss Tweed managed to corrupt plurality,
not IRV, voting – true – having
a more-auditable voting system like plurality is not by itself a sure-fire
way to stop election fraud. But why give up ground?)
Do you think it is important for the purpose of assessing how well a voting system like IRV is
working, to allow access to enough data to reconstruct the election? (I do.)