Summary of Spoilage and Error Rates with Range Voting versus other Voting Systems
has the lowest observed spoilage rates of any voting system, usually 0.04% to 0.81%.
Range Voting is the next best, usually 0-2% spoilage
(but less if "partially spoiled" ballots counted partially). It is not surprising that range
voting has lower observed ballot-spoilage rates than the (clearly more complicated
with more ways for the voter to "go wrong") rank-order methods like instant runoff
and higher rates than the (clearly simpler) approval voting system.
It is surprising that range voting appears to do better than plurality voting.
The fact that plurality voting
appears to exhibit higher ballot-spoilage rates than range voting (usually 1-4%)
due to the repetition involved in voters giving a score to every candidate
allows them to catch errors; or perhaps due to the fact that any score-assignment
(within the permitted range) is legal, whereas with plurality voting attempts to "overvote"
are illegal and those ballots are discarded.
The worst system, exhibiting the clearly greatest ballot-spoilage and error rates over long
cumulative historical experience (e.g. 80 years use in Australia), is
Instant Runoff Voting (usually 4-9%).
However, there are indications that with user-friendly computerized voting machines it is possible
(via it instantly detecting, warning the voter about, and refusing to accept erroneous votes)
to reduce IRV error rates (and hence all the above error rates)
to very low levels. (The above comparison was based on equal pre-computerized
technology, or anyhow if the computer was used, it was later so that it did not
warn the voter.) In that case error rates would become almost a non-issue. But many
familiar with this area think computers should be forbidden inside voting machines
because fraud and manipulation risks are just too dangerous.
Australia (anyway most of Australia) requires full ranking. Their same high error rates would
presumably be expected from every other rank-order-ballot based voting system
(e.g. Condorcet, Bucklin).
Systems allowing only partial ranking (as is done in Ireland) would be expected to get
(and do get) lower error rates. If rankings with equalities were permitted
(as in some proposed Condorcet systems, but as far as I know this has never been permitted
in any large scale governmental election as of 2010) the effect that would have on error rates is
not clear. The fact that with IRV, local
counting is no longer possible and near-tie "chad-counting nightmares" are possible not just once,
but actually in every "round" of the election process, makes matters even worse.
The actual numbers depend on the country, the part of that country, which technology
is employed and how, and which election it is.
Typical nationwide ballot spoilage rates with plurality voting in presidential elections are
1% to 3%.
With approval voting a large
2002 France study found below 0.39% spoilage –
an enormous improvement.
(Redone in France 2007 finding 0.81% spoilage
and in Germany 2008-2009
finding spoilage rates of 0.07%, 0.14%, and 0.62%, which all are impressively low.)
With range voting, our own study found plurality-spoilage rates were
roughly cut in half (whole ballot) and reduced
by a factor≈14 per entry
(note partly-spoiled range ballots still are usable,
unlike plurality ballots). The larger
2007 French range-voting study conducted with government funding and
cooperation found (whole ballot)
error rates cut by a factor of 2.3 versus the first round, and by a factor of 4 versus the second
round, of the official plurality-voting French president election conducted at the same time.
With Instant Runoff, plurality error rates increased
by a factor of 7 in San Francisco
2004, but the more-experienced election officials and voters in Australia
exhibit IRV spoilage rates of 3% to 7%, which is only about 2-to-6 times worse than
France 2007 study got IRV spoilage rates of 7%,
exceeding plurality's official spoilage rates of 1.44% and 4.20% in the first-round and runoff