Example to demonstrate why IRV cannot be counted in precincts

(Demonstration that IRV is not "additive")

Counting votes in precincts (or counties), then sending the precinct totals to the statewide election office – it has been a way of life in almost all US states for as long as any of us have lived.

But if we were to switch to IRV (instant runoff) voting, that would no longer be possible. All the votes, en masse, would have to be given to statewide central tabulating for them to count; there simply could no longer be such a thing as a "county subtotal."

Is that an insuperable obstacle to adopting IRV voting? No. It'd be a huge change in US election procedures, financing, and organization, and it might make everything much more easily vulnerable to huge centrally-organized election frauds – you won't be able to observe your and your neighbors' own votes being counted anymore – but it won't be fatal to IRV, since central counting is certainly possible.

To make it clearer why there is no such thing as a district "subtotal" anymore if we use IRV voting, consider the following example.

In district I, IRV eliminates C, then B wins 7:6. In district II (same as district I but the roles of A and C are reversed), B also wins 7:6. But in the combined 2-district country, B has 8 top-rank votes, A and C have 9 each, so B is eliminated and either A or C wins. Thus merging two districts both won by Bush under IRV, can produce an IRV victory for Gore.

Wait – Can't IRV be counted in precincts (and on old-style plurality machines) in this other way?

IRV can be done on plurality machines, and counted in precincts, by simply making each permutation (rank ordering) be a "candidate."

Our response: That's true but practically infeasible. In a C-candidate race, each voting machine would have to have C! pseudo-candidates and each precinct would have to pass C! "subtotal" counts on to the central tabulator. If C is large this is infeasible:

(Aside: Actually, it is even worse than this if voters are allowed to "truncate" their preferences – then even more than C! ballot types become possible, namely ⌊C!e⌋ where e=2.71828… and note, if you do not count an all-blank ballot as a "ballot type," then subtract one. See puzzle #91.)

Note, 13! is about equal to the present world population. Heck, you could just pass all the V votes to the central tabulator, and that'd be easier than passing the subtotals (if C!>V) which defeats the purpose of having subtotals. (A typical precinct has V=2000 voters. But 7! = 5040. Also if the IRV rules allow "ballot truncation" then the true number of ballot types actually would be much larger than C!.)

So "counting in precincts" is silly if precincts have to pass an exponentially large amount of information along – larger than just not totalling at all and just sending all the votes in unprocessed form!

Also, more to the point, I want precinct totals to be published. That's not going to happen if a precinct is going to have to publish 6!=720 "totals" in one race. And even if that did happen, then this publishing would defeat ballot secrecy and open the door to vote-selling and coercion.

Wait – Can't IRV be counted in precincts via Two-Way Communication?

Yes. IRV can be counted in precincts if there is two-way communication between precincts and headquarters – headquarters informs all precincts who to eliminate next, then precincts report their top-rank totals (among remaining candidates) and then the cycle repeats.

But this basically is centralized counting. If you have counting in one central location directed by some "director," what is the difference versus having it in, say, ten central locations, directed by that same director using modern two-way communication technology?

So yes, it will still work. And according to Rob Richie (June 2008 web post), "In Ireland there are 43 counting centers in the presidential race. Election administrators count ballots and report their totals to a national office that in turn instructs the administrators at each counting center on what to do next. The entire process takes less than a day even though more than a million ballots are cast."

There are several problems with the rosy picture Richie here tries to paint. First, doing this requires modern communication technology, e.g. computer networks. (E.g. Richie suggests "If ballot images are recorded on optical scan equipment, the data from those images can be collected centrally.") It won't work if that communication technology is disabled, broken, misused, or attacked. That's a step backwards (increased vulnerability, reduced simplicity, reduced transparency) versus voting systems like plurality and approval where precinct subtotals exist and no two-way communication is needed.

Second, this is a centrally-directed scheme. There still is no such thing as a "district subtotal." If a district counts its ballots locally under central direction, then it cannot publish its subtotals because there is no such thing as a district subtotal. That reduces the transparency and checkability of elections.

And doing this could require all workers at every location to keep counting and recounting until every single last vote everywhere gets counted, which'd be a very long workday (or perhaps a better word than "day" is "month"...).

The problem is if central doesn't know whom to eliminate next because that depends on a very close vote which depends on every last absentee ballot, missing chad, debate over who intended what on some unclearly-marked ballot, etc. And even if this particular elimination-decision might between two "no hopers" who have only 1% of the vote each, each elimination decision can amplify arbitrarily much and thus alter the ultimate winner, so Central can't just "ignore the issue and move on." And then everybody everywhere has to remain at work waiting a month for Central's next announcement. And in principle such a chad-counting nightmare could happen in every IRV round. So the worst case scenario is really very bad.

If indeed, every precinct in the USA (I think there are over 100,000 of them) were to try to count ballots under central direction as Richie has in mind, that could require every precinct-worker in the entire USA to keep counting for 1 month. That would raise the cost of elections astronomically. So in fact (to avoid that) really the counting would need to happen at a much higher level than "precinct," such as "statewide." Which is fine – but then it is not local anymore and it is a big change from today's organization of elections with local, not central, counting.

The way many IRV countries actually do it is, they do a single pass making a probabilistic model of who is likely to be eliminated and in what order. If the model turns out to be correct, then everything goes smoothly. If it turns out to be wrong, then they get into trouble and have to start recalling the election workers and doing more counts... If it gets it wrong in more than one way we can really get into huge difficulties... but fortunately this in practice seems rare, probably because all IRV countries so far have been 2-party dominated. In Australia's 24 November 2007 elections, the Election Commission was unable to determine the composition of parliament until over 1 month later because of numerous races which were difficult for them to count.

Here's what happened when North Carolina tried to do IRV with precinct counting...

The enormously complicated official rules (pdf) boggled everybody's mind.

• This "opti-scan" machine IRV counting method requires (to accomplish just this one simple counting round for only one election contest) feeding each ballot one at a time by hand through the precinct opti-scanners up to four times with the optical scanner needing a differently programmed PCMCIA card in it for each of these four "counts," during which poll workers must pay close attention to whether or not each ballot is "rejected" or "accepted" and put each ballot in a correct pile depending on which stack it comes from and whether it is rejected or accepted by the M100. Note this procedure is used to only count one IRV counting round since the first round is assumed to be previously counted at the central county office and announced to all the polling locations first only after all the candidates except for the top two 1st choice vote-getters have been "eliminated" by the central county office counts.
• The local scanners, the M100s, or the PCMCIA cards (or both) must be reprogrammed by poll workers between each of the four times that ballots are fed into the optical-scanners. The PCMCIA cards seem to need to be "burned" in three separate configurations (to count each of the three ballot rankings) at the poll locations only after the two winning/continuing candidates are announced. The instructions left out the exact details of how to "burn" the PCMCIA cards.
• The ballots must be resorted and re-fed one at a time into the optical scanner four (4) times the number of IRV election contests. E.g. If there are three IRV contests on the ballots, then many individual ballots could be scanned and resorted up to twelve (12) times each!
• I.e. the ballots would have to be accurately sorted into four piles during the first feed (one ballot at a time) into the precinct scanners...
• and then sorted into five piles (sorted to four piles coming out of the optiscanner plus one pile not run through it) during the second feed (one ballot at a time) into the precinct scanners...
• and then sorted into six total piles (sorted to three piles coming out of the optical scanner plus three piles not run through it this time) during the third feed (one ballot at a time) into the precinct scanners...
• and then sorted into seven total piles (sorted to two piles coming out of the optiscanner plus five piles not run through it this time (one ballot at a time) into the precinct scanners.

We thank Kathy Dopp for contributing this info about North Carolina. (She also wrote a paper summarizing 18 flaws in IRV.) She concludes: "It will be nothing short of a praise the Lord miracle if this process is performed accurately in all polling locations." Note that even after adopting this hugely-complex procedure, North Carolina still failed in the sense that if some IRV decision made before its final round (for example, its first round) were a close one, then even this procedure quite likely would not work – the IRV counts would have to be restarted from the beginning and counted centrally. Even early-round subcontests between candidates who might naively seem to have no hope of election, can have ballooning effects which alter the ultimate winner.

Analogous problem for Condorcet voting systems

Return to main page