What Voters Want – score voting seems their favorite

By Warren D. Smith, May 2013.

Collected here – for the first time ever! – is convincing evidence, based on high quality polls, about what voting systems real voters do and do not want. By "convincing," we mean "of comparable or greater clarity than (what are normally regarded as) the greatest-ever popular-vote landslides in USA presidential election history," which (we remind the reader) were about 61% super-majorities or, if instead reckoned by the difference Δ between the winner's and 2nd-placer's vote counts, Δ≈25% of the number of voters.

Here are the answers. Let "A>B" mean "there is convincing evidence that voters prefer voting system A over voting system B."

1.     Plain plurality voting > Instant Runoff Voting (IRV).

2.     Average-based score voting with 3-point numerical scale (don't care whether it is {-1,0,+1} or {0,1,2}) > Approval Voting > Average-based score voting with 21-point numerical scale {0,1,2,...,19,20}.

3.     Average-based score voting (unspecified numerical scale) > Approval Voting > 2-round plurality > "Majority Judgment" with 7-level verbal scale.


The evidence

1. There is convincing poll evidence from UK, Australia, San Francisco, and BC Canada that voters do not want IRV (instant runoff, several flavors) or the related but more complicated STV system, if choice is between IRV & plain plurality voting.

AUSTRALIA October 2010: A nationwide professional telephone poll by "Newspoll" (see question #9) of 1202 random Australian adults found that they prefer plain-plurality voting versus the preferential (instant runoff) system Australians presently use to elect their House. If forced to choose one, they'd choose to abandon IRV – the poll's result was 57% to 37% (with 5% don't know/refuse; this is 60:40 if the "don't knows" are ignored). IRV flavor: In most of Australia, voting is compulsory (nonvoters pay fine) and ranking all candidates also is compulsory. However, in some of Australia, voters are allowed to rank fewer candidates if desired.

A closer look: Plurality enjoyed greater support than IRV in every geographic area, every income level, both genders, and all marital and parenthood statuses. However, there were some demographic subgroups which preferred IRV, namely: young, university educated, and members of Green Party and a few more-minor and obscure political groups.

Australian poll question wording: Currently, elections for the Federal House of Representatives, or lower house, use a preferential voting system. This is where voters indicate an order of preferences for all candidates, and these preferences are taken into account when deciding which candidate wins. (PAUSE). An alternative system would be "first past the post", where voters only vote for one candidate and the candidate with the most votes wins. Would you personally prefer...? READ OUT (PROG. NOTE: SINGLE RESPONSE, RANDOMISE 1-2, THEN 3 LAST):
1A preferential system
2A first past the post system
3DO NOT READ: Neither / don't know

Note the Australian poll result even could be regarded as showing voters are against every rank-order-ballot voting system, not merely IRV. That would be absolutely devastating to the entire political science voting system literature before the year 2000 – good job that literature studiously omits mention of this poll.

MORE ON AUSTRALIA. Antony Green (who often does election commentating for ABC, Australian Broadcasting Co.) wanted to dispute my above poll. To do so, he found two earlier polls by Roy Morgan Research in the years 1974 and 1984 (he says information about them was published in "The Bulletin," 12 Feb 1985):

     Roy Morgan Research 1974       Roy Morgan Research 1984
     first past post   53           first past post   54 
     preferential      38           preferential      39 
     undecided         9            undecided         7 %
which seem entirely to support exactly my same conclusion – Australians prefer plurality voting over preferential ballot systems, by large margins! (Incidentally, the top 4 polling companies in Australia are Roy Morgan, Newspoll, Nielsen, and Galaxy Research, probably in descending order.) Undeterred, Green then announced that all three of these polls found the wrong conclusion! How could Green possibly dispute all three (agreeing, large-margin, respected pollster) polls from different pollsters at different times (when Green had no other such polls)?? Well, Green's argument was long and convoluted, but essentially this:
At all 3 poll times, something political was going on in Australia, and therefore voters supposedly interpreted the questions as some kind of referendum, not on which voting system, but rather on their opinion of that one-time political phenomenon. Even though the poll questions never mentioned one word about that political phenonemenon, nor about any Australian political party. And by pure coincidence, all those three times, the Australians who thus misinterpreted the question always managed by pure bad luck to come out with roughly the same deceptive poll result every time, and this bad luck just kept happening, thus preventing the True View, which Green Alone Knows, and no poll supported, from coming to light: namely, really, Australians prefer preferential voting. And, the IPA ("institute for public affairs"), which paid for the 2010 Newspoll, was in Green's view a biased organization, which (he tried to insinuate) must have meant they bribed Newspoll to fake their results, or something. [Unlike Green himself – he's not biased.] Of course, at any time, something political is always happening in Australia, so one might presume that no matter how many such polls there are repeating the findings of the above three, Green would continue to dispute them all for the same reason? Perhaps not, because Green further supported his views by providing political-party-liking breakdowns of the YES and NO pollees in those three polls. (These breakdowns were given by him without even a single calculation of a single statistical significance number.) Green noted that these breakdowns differed, and in ways which, he thought, somehow proved his point. Of course during the 37 years 1974 to 2010 one might naturally have expected these party breakdowns to time-evolve regardless of the correctness of Green's theory (e.g. entire political parties were born during this span...), but he never mentioned that possibility. Finally, in 2013, Green found a fourth poll, which, he contended (for some reason), now was not merely a victim of temporary political circumstances, but actually revealing the truth:

POLL QUESTION ("Essential Media" 11 July 2011 through 14 January 2013):
Which of the following voting systems would you prefer when voting for the Federal House of Representatives?

  1.   A preferential voting system where voters rank all candidates in order of preference.
  2.   An optional preferential system where voters can rank one, some, or all candidates in order of preference.
  3.   A "first past the post" system, where voters only vote for one candidate and the candidate with the most votes wins.

Note that this poll question is worded quite differently from the preceding three. It offers three response-options, not just two. Green claimed this was the first poll ever, to offer these three options. Note that this poll question is badly designed – for precisely the same reason that plurality voting is a poor voting system for 3-candidate elections! – namely, we have 3 options, 2 of which might (at least naively) seem near-clones, leading to the risk of unclear, hard-to-interpret, "vote-splitting" phenomena. It would have been far better to, e.g, design the poll as a rating question, e.g. "for each option, give a number from 0-100 saying how much you like it." (Or, if you insist on a plurality-style poll, then at least make it symmetric by having two first-past-post options, one with mandatory and one with optional voting.) But because perhaps the pollster in this case was not one of the most-respected polling companies in Australia (listed above) but rather the little known firm Essential Media (via subsidiary Essential research) which as of year 2014 does not even mention "polling" as one of its corporate activities, it did not. Also, it should be noted that this poll was conducted online, unlike the previous polls by Newspoll (telephone) and Roy Morgan (face to face interviews), and over a 1½-year span, not a short period. Online polls are much harder to do well (due to sample bias self-selection effects – the most obvious being that those without internet access, are uncounted) than telephone & in person polls, hence are generally regarded with suspicion or as experimental at this time – especially if done by a non-well-established polling firm that hasn't had many years to refine its online techniques via comparisons with other polling methods. In any case, the results (1878 respondents) were

    A: 22%,       B: 26%,        C: 44%,     Don't know: 7%.
This, Green contended, finally showed the Truth that Australians prefer a preferential system, since A+B combined outweigh "first past the post" by 48-44. Unfortunately, Green's interpretation is dubious. First, this 4% margin is far smaller than all three of the margins (20%, 15%, 15%) going the other way in the preceding polls conducted using more-reliable techniques by more experienced pollsters with non-stupidly-designed poll questions. Second, if, say, a mere 1/5 of the "B" respondents preferred C over A as their (unstated) 2nd and 3rd choice options, then adding 22+26 to get 48 would be unjustified (it's only justified if 100% of the B's prefer A>C, and 100% of the A's prefer B>C), and then instead of the so-called 48-44 margin, the real margin would be in the other direction, i.e. now indicating Australians prefer Plurality (first past post) voting! If, further, 1/5 of the A's preferred C>B, then the margin would be at least 49-43 in favor of first past post! Are these "1/5"s realistic possibilities? Certainly, in view of the 3 previous polls. (Also, just in view of the fact 4:1 or larger margins in poll questions, are rare.) Also, we should point out that in real life, Australians do not get to use "system A or B, whichever you prefer." They are forced to use just one preferential voting system (whichever the law demands in their location), often not the version they like best. If asked to choose between just it and first-past-post, this poll would seem again to indicate Australians would select the latter.

UNITED KINGDOM (Britain & N.Ireland): A 5 May 2011 binding nationwide referendum asked voters to decide whether the UK should switch from plurality to IRV voting (called in both Australia and Britain "the alternative vote"). The flavor proposed for the UK (oh sorry, make that "flavour") would have allowed voters to rank as many or as few candidates as they wanted. The result was a massive landslide victory (68% to 32% of the 19.3 million votes) for "stay with plurality."

A closer look: Plurality won in every one of the 12 geographic regions, the smallest victory being by 56.32% in N.Ireland, with all others exceeding 60.47%. (Probably related to: N.Ireland is adjacent to Ireland, which has long used IRV to elect president.) Unofficial professional polls indicate it also won in both genders, in every age group except 18-24, and in every political subgroup except LibDem. Voting in the UK is not compulsory (nor would it have become so), and indeed the turnout for this referendum was 42.2%. It also should be noted that, although voter and proponent understanding on both sides could easily be criticized, this referendum was undoubtably the greatest education-and-debate effort ever conducted about a choice between voting systems.

UK referendum's wording: At present, the UK uses the "first past the post" system to elect MPs to the House of Commons. Should the "alternative vote" system be used instead?

BRITISH COLUMBIA, CANADA 12 May 2009: In a binding province-wide referendum, "switch to BC-STV" (from plain plurality) lost with only 39.09% of the 1.65 million votes cast. (55% turnout.) The proposed BC-STV system involved "votes" which were rank-orderings of the candidates; each voter being allowed to rank as many or as few of the candidates as desired. The algorithm for computing the winners from the votes is more complicated than IRV, involving "reweightings" as well as "transfers."

BC referendum's wording: Which electoral system should British Columbia use to elect members to the provincial Legislative Assembly?

SAN FRANCISCO, USA: In a 5 March 2002 referendum (31% turnout), voters decided (by 55-45 margin) to switch from a 2-round genuine runoff system to a 1-round instant runoff voting system. It was first used in SF's Nov. 2004 elections. As a result, at the time of this writing (2013) San Francisco has more IRV experience than any other US city. Why did SF's voters go for IRV? It may have had something to do with the fact that the official referendum ballot question contained a lie.

Official question wording: Shall the City use Instant run-off voting to elect City officers with a majority of votes without separate run-off elections? (YES / NO).

Note the words we've italicized. They are a lie. IRV, especially the IRV3 rules SF adopted, does not ensure that those elected have a majority. (Indeed: after passage, IRV elections were duly held in SF, and indeed elected many people who did not have majorities.) Further, it can happen that a candidate who really is majority-preferred versus every rival (such as Montroll in Burlington Vermont's IRV mayoral election of 2009) loses with IRV. The wording also is deceptive in the sense that with the old genuine-runoff system, if anybody had a majority (50%+1 vote), then the "separate run-off" 2nd round automatically would not happen. This wording created the erroneous impression that was not the case, and that IRV would "repair" this "flaw." It really was an outrage that this question-wording was even permitted.

But anyhow, either because of this official-wording lie or for some other reason, SF's voters went for it. But here's what happened in February 2011, when SF Chamber of Commerce hired David Binder Research to poll 500 SF voters to see what they now thought (with the benefit of 9 years of hindsight) – and then again for a re-poll on the same question in February 2012 (after 10 years hindsight):

San Francisco CityBeat poll results (Each poll 500. Standard errors also shown with ±.)
Preference CityBeat 2011 CityBeat 2012 Both combined
Prefer (old style) genuine-runoff elections 52% ±2.2% 58% ±2.2% 55% ±1.6%
Prefer Instant Runoff (Ranked Choice) voting 42% ±2.2% 31% ±2.1% 36.5% ±1.5%
Don't know/Refuse 6% ±1.1% 11% ±1.4% 8.5% ±0.9%

Note that these poll results are in the opposite direction and both have a greater margin than the original enactment vote, and each is statistically fairly significant (confidence≥98% that IRV loses) with the combined confidence (i.e. based on both polls together) exceeding 99.999995%. The margin if "don't know"s are ignored is 60-40 in favor of "go back to the old genuine-runoff system."

NEW ZEALAND 26 Nov. 2011: Voters were asked (in a nationwide non-binding referendum) whether to get rid of their present MMP system and replace it with something else. Whether to keep MMP was question #1, and "keep MMP" won with 57.77% of the vote. About 2.2 million voted in this referendum (about 74% turnout).

More interesting for our purposes, was question #2, which was "assuming 'abolish MMP' wins in question #1, then what?" The voters were offered 4 choices:

  1. Plain plurality voting (called in NZ "first past the post")
  2. instant runoff voting (IRV, but in NZ called the "preferential" system)
  3. An STV multiwinner system (similar to IRV but more complicated due to reweightings as well as transfers...)
  4. "Supplementary member" (SM) system rather related to MMP.
The results were: "Plain plurality" won by landslide with 46.66% of the vote, versus IRV (12.47%), STV (16.73%), and SM (24.14%). Admittedly since this was a 4-choice vote, we, technically speaking, do not have here a direct comparison between plain plurality versus IRV. However, with plain plurality way in first place and IRV in last place it nevertheless looks fairly clear New Zealand voters prefer the former. And they know something of what they speak because IRV has actually been used to elect, e.g, the mayor of Wellington (NZ's second largest city, and capital).

A 16 June 2011 Research New Zealand poll asked "Supposing the choice was between MMP as it is and the old First Past the Post system, which would you prefer?" and MMP won with 49% versus FPP with 33% (the rest were "neither," "don't know," or refused). Same preference for every ethnicity, gender, and income range surveyed.

UNITED STATES OF AMERICA 1900-2013: As far as I know as of 2013 there has never been a nationwide or statewide "switch to IRV" referendum. But there have been many such referenda in cities, towns and occasionally counties (as well as "abolish IRV" referenda in places where they'd enacted IRV earlier). I have not done a precise count, but I believe that IRV has lost more USA referenda than it won. More precisely, I believe about 90% of the times IRV or a closely related system got enacted throughout USA history, it later was repealed in a follow-up referendum; plus there were many cases where "enact IRV" referenda were defeated. Although that was only an educated guess by me, the following is

FACT: If all the "IRV versus plain plurality" referendums ever held in world history (up to now, 2013) are added together, then the "plain plurality" side wins by a large margin.

basically because the British referendum vote-count overwhelms all the others combined.

2. There is landslide poll evidence voters do not want "majority judgment" with 7-point verbal scale, if choosing between it and present French 2-round plurality plus 2nd round runoff system.

FRANCE 6-7 April 2011: A professional nationwide poll of 1000 French was conducted asking the voters to score the projected candidates for the April 2012 presidential election on the 7-point verbal scale

Excellent / Tres bien / Bien / Assez Bien / Passable / Insuffisant / a Rejeter.

and provided them with an oversimplified (but otherwise well-written) 1-page description of the MJ voting method. They also were asked to vote using the normal ("name one candidate") French method. (We report the results elsewhere.) After completing those voting tasks, the pollees were asked which voting method they preferred:

3. Academics conducted several studies of approval voting and several flavors of score voting in three French towns (as "exit polls" with government funding and assistance) during the 2012 presidential election. 2340 voters participated in this voting study, which was 54% of the voters at the official polling places they used. However, some only participated partially, i.e. not answering every question.

All the voting methods in these studies (as well as the official election) produced the same winner: Hollande. (We discuss the results elsewhere.)

I'm mainly relying on this July 2012 academic paper (in French, by A.Baujard, F.Gavrel, H.Igersheim, J-F.Laslier, I.Lebon). For our purpose here, the important thing is that a questionnaire was appended to the ballots used in this voting study. 85.9% of the pollees opted to answer the questionnaire.

QUESTION 1 asked which kinds of voting they wanted. 1958 answered. The 4 choices were:

  1. Les deux regles (meaning, both approval and score voting should be used) 27.53%
  2. Vote par approbation ("approval voting") 29.47%
  3. Vote par note ("score voting") 32.84%   WINNER!!
  4. Aucune des deux ("neither of them") 10.11%

Incidentally, elsewhere in that same paper the official 2-round plurality (with 2-man runoff if needed) system was described as "Vote uninominal a deux tours (officiel)."

Note, there are 4 "binary" possibilities here, namely 00, 01, 10, and 11 where the first bit is "want score voting?" and the second is "want approval?", perhaps best displayed in tabular form as

Want scoreDo not want score
Want approval27.53%29.47%57.00%
Do not want approval32.84%10.11%42.95%
60.37%39.58%

The most generous interpretation would be that an overwhelming 89.89% of the voters want at least one of {score, approval} voting. However, some would argue that their question 1 was poorly worded. So probably more clear and convincing for us were questions 2a and 2b:

QUESTION 2a: asked for which kinds of elections the approval voting system should be used (or not). 4 subquestions:

    Elections presidentielles:  61%.
    Elections legislatives:     57%.
    Elections municipales:      61%.
    Associations:               52%.
  (Average of above numbers)    58%

QUESTION 2b asked for which kinds of elections the score voting system should be used (or not). Same 4 subquestions:

    Elections presidentielles:  62%.
    Elections legislatives:     55%.
    Elections municipales:      66%.
    Associations:               51%.
  (Average of above numbers)    59%

Note that a majority of respondents prefer either – versus France 2011's official systems – for every purpose, and 61% is comparable to the largest ever USA presidential "landslides" while 90% (question 1, generous interpretation) is far greater.

What about possible biases in these exit-poll studies? In case you are worried that those who opted to answer the questionnaire were a biased subsample of the poll-respondents, keep in mind that even if every no-questionnaire pollee were against score voting, that still would not have been enough to swing any ≥59% result for score voting. If you are worried the 54% of the official voters who chose to participate in the study were a biased subsample of "all voters," then keep in mind the exit polls are regarded as the most accurate kind of poll, 54% is a very high participation rate for an exit poll, and 2340 pollees is regarded (by pollsters in the year 2012) as an unusually large poll; and finally we can check how biased these 54% were by comparing their votes with the known official vote counts for the full 100% of the voters at the studied locations. Unfortunately only 1345 study participants out of 2340 (57%) agreed to reveal their official vote, leading to additional selection effects which hurt our checking ability! Anyhow, to perform that check, just compare the last two rows of Table 3.1 in the paper, which we re-give below:

                  Hollnd  Sarkozy  LePen  Mlnchon  Bayrou  Joly  Dpnt-Agnn  Poutou  Arthaud  Chemnd
Official results   33.16   22.31   12.57   13.54   11.60   3.61   1.56       0.97     0.57    0.12 
Study results (%)  41.11   14.37    5.87   16.62   13.37   5.95   1.16       1.00     0.15    0.39

So it would appear the study-participants (at least the ones who revealed their official votes!) were more pro-Hollande and more anti-Sarkozy and anti-LePen than the full voter set at those locations, and these three discrepancies are statistically significant (although the discrepancies for the other 7 candidates tabulated may not be). These three discrepancies were 41.11-33.16=7.95%, 22.31-14.37=7.94%, and 12.57-5.87=6.70% of the voters, i.e. 7.95+7.94+6.70=22.59% in all. If we model these as 22.6% of the study voters were 'strange' and the remaining 77.4% 'normal' and if we (to play it safe) assume the strange voters were voting 70-30 in favor of approval/score voting, then even if we were to remove all those strange voters, then still, score-voting would be winning by a comfortable margin – no longer 59%, but over 54%. (Indeed, you'd need to assume the strange voters were 90-10 in order to overturn Score's win. That seems very implausible.)

So in short: even after taking into account possible flaws in the poll sample, and even doing so in a pretty pessimistic way, score voting still clearly wins.

A closer look: actually this academic study trialed three different score voting systems, one in each of the 3 towns:

     TOWN...........#POLLED.........TYPE OF SCORE VOTING TRIALED
     Louvigny.........930..............{-1, 0, +1}  check one of three boxes for each candidate
     St.Etienne.......387...............{0, 1, 2}   check one of three boxes for each candidate
     Strasbourg......1023..........{0, 1, 2, ..., 19, 20}   write two digits for each candidate

And the results of question 2b (when examined town by town) show there was clearly more support for {-1,0,+1} and {0,1,2} than for {0,1,2,...,19,20} but in contrast question 2a got about the same support rates in all 3 towns. Indeed, in Strasbourg, the town using {0,1,2,...,19,20} score voting, alone, the 802 responses to QUESTION 1 were

  1. Les deux regles 25.81%
  2. Vote par approbation 35.29%
  3. Vote par note 28.43%
  4. Aucune des deux 10.47%

indicating approval voting was actually preferred over {0,1,2,...,19,20} score, even though with the full distribution of score-sets used, overall score was preferred versus approval despite the handicap that 44% of the pollees were in Strasbourg.

The lesson of that is that score voting with the most-popular score-set will be even more supported. Unfortunately we do not currently know what that set is, although the authors of the study conjecture {-2, -1, 0, +1, +2} ought to be more popular than the three sets they tried. (We elsewhere have some data about what scales are liked/not, suggesting a 10-level scale is most wanted.)

A cautionary tale: Compulsory voting

Question #10 of the same Oct. 2010 Australian poll asked whether Australian voting should continue to be compulsory, or should become voluntary. Result: compulsory=69%, voluntary=29%, neither/don't know=2%. Compulsory won in every economic, age, educational, and geographic demographic subgroup, and also in every political subgroup except for "non-voters" where it only got 49% support.

That seems pretty clear. There is just one problem:

This result was virtually the exact opposite of that found in a 2-6 June 2004 ABC News poll of 1001 USA adults: 72% opposed a law that would require all eligible citizens to vote in national elections and levy a small fine on non-voters who do not have a good excuse for skipping the polls. Only 21% said the law would be a good idea. "Non-compulsory" won in every USA demographic subgroup investigated. [And this 2004 result was almost identical to the results from Gallup polls in 1965, when 69% opposed such a law.]

The lesson of that contradiction is this: different countries can yield different conclusions. The Australians, who for their entire lives had had compulsory voting, felt differently than the Americans who, for their entire lives had had noncompulsory voting.

But that seems exceptional. In the case of IRV (issue #1 above) we've given polls from several countries including ones with a long history of IRV use and ones with a long history of IRV non-use. In the case of score and approval voting, I do not see why France and (say) the USA should have any great disparity on this, and in the case of MJ, presumably France would be expected to be more pro-MJ than the USA (since MJ had been highly and favorably publicized there) but MJ still lost by landslide in France.

A less-convincingly supported election USA reform: change election day from Tuesday to Weekend

POLL QUESTION: Some people have suggested that election day should be changed from Tuesday to an entire weekend so that people would be able to vote on either Saturday or Sunday. Do you think this is a good idea or a bad idea?

Quinnipiac University Polling Institute, 275 Mount Carmel Avenue Hamden, CT 06518-1940. NY field date 6-12 Feb, NJ field date 30 Jan-5 Feb, both in 2001. Method: telephone. Sample Sizes: NY=1412, NJ=1156.

New YorkNew JerseyCombined
whiteblacktotaltotaltotal
good idea43%71%48%48%48%
bad idea48%25%44%47%45%
don't know/no answer8%5%8%5%7%

Unfortunately this was not a nationwide poll, it was only a poll within New York state plus another poll within New Jersey state.

That racial disparity is rather interesting. New York (2010 census) is 65.7% white, 15.9% black, and 18.4% other races (as self-described). USA-wide is 72.4% white, 12.6% black, and 15.0% other. (NJ: 72.6% white, 13.6% black, 13.8% other.) Extrapolation from NY+NJ to USA-wide (discarding don't-know's and using all 3 races) therefore suggests "weekend" would win by about 51.3% to 48.7%. This is not hugely convincing because putting in statistical 1-standard-deviation error bars, that "51.3%" would really be more like (51.3±1.3)%.

So in summary, this "change to weekend" idea seemed (when polled in 2001) to have more popular support than "stay tuesday," but not convincingly and not hugely.


See also surprising poll results about comprehensibilities of voting methods.

Which scales score voters prefer.

Return to main page