Maine Governor 2014 – Cutler should have won!

This account by Warren D. Smith, Nov. 2014. Based on work by Nevin Brackett-Rozinsky, Clay Shentrup, Elijah Miller, Gerald Weinand, Aaron Hamlin, Warren D. Smith. Please notify warren.wds AT gmail.com of any errors.

This election was a train wreck. The candidates were Paul LePage, unpopular Republican governor seeking re-election (one of the least popular governors in the USA according to his approval numbers in polls); Mike Michaud, Democratic congressman with a rarely-surpassed lifetime record of hypocrisy, and Eliot Cutler, wealthy Independent lawyer whose only previous experience directly working in government apparently had been when he'd served with the OMB – appointed by US president J.Carter – over 30 years prior. Cutler was initially endorsed by (also Independent, former Maine Gov. 1994-2002 and its sitting US Senator) Angus King. Election day was 4 November 2014. (Cf. the preceding 2010 election.)

Official Results
(Plurality Voting)
Approval Voting Instant Runoff Score Voting
(four levels 0-3)
LePage   48.19% 
Michaud  43.38%   
Cutler    8.43%   
[source: 
State of Maine
rescaled so L+M+C=100.00%. 
July 2015 supplants earlier 
data in New York Times 
differing in last decimal.]

         CES    PPP
LePage   46.35  43/53 
Michaud  48.76  47/46 
Cutler   54.95  39/38 
[source: reweighted
CES exit poll, PPP
favorable/unfavorable
poll 22-23 Oct.]
Round One:
LePage   36.83
Michaud  29.29
Cutler   32.00
Two: M eliminated
then Cutler wins.
[source: reweighted
CES exit poll.]
LePage   1.29
Michaud  1.65
Cutler   1.68
[source: averages 2 Oct.
score-style polls by
Bangor Daily News/Ipsos
& 1 by PanAtlantic/SMS.]
  With median-based MJ
methods Cutler also wins 
in all 3 of these polls.
"Head-to-head" Pairwise ContestsCondorcet*
Cutler-LePage 57.4-42.6 50-40, 53-40, 49-38, 58.6-38.2       Cutler wins.
Michaud-LePage 51.0-49.0 50-46, 50-41, 43-50, 45.5-42.7, 48-46, 49.3-48.6       Borda voting*
Cutler-Michaud 59.9-40.1 54.6-36.5       Cutler wins.
[Sources: The last pairwise poll in each row is our reweighted CES exit poll; the others are from polls by Frederick, Bangor Daily News/Ipsos, Public Policy Polling, Portland Press-Herald/Univ. New Hampshire survey center, and PanAtlantic/SMS. All polls agree on all pairwise contests except for the italicized poll; the pink column is the unweighted-average poll result rescaled to sum to 100. Actually we believe LePage would have defeated Michaud pairwise, despite the averaged-poll conclusion above indicating Michaud would win 51-49, which was a statistical fluctuation. This is the only pairwise race that is in doubt, but fortunately the other two are the ones that matter for the purpose of seeing Cutler should've won.]
* Inferred from IRV ballots and/or pairwise polls; not tested directly.

A somewhat simplified version of the above data tables also is available in graphical form:

graphical form of Maine election data tables

Discussion

This is one of the most dramatic findings ever seen in an election with multi voting system polling data, and is a huge indictment of Maine's (and most of the world's) horrible plurality voting system.

Voting reformers: rejoice. Citizens of Maine: not so much. You will pay.

And the best voting systems are...

In this election,

Who are you calling a "spoiler"?

Detractors (mainly Michaud supporters) loudly fretted that Cutler was a "spoiler," a claim repeated by the media. But this claim was incorrect, in view of Maine's Official Totals plus these pre-election poll results (asked only of Cutler supporters):

If Cutler were out, then his supporters would break
   45.8% Michaud,   18.6%  LePage   (A)
   64%   Michaud,   27%    LePage   (B)
   53%   Michaud,   32%    LePage   (C)
   53%   Michaud,   24%    LePage   (D, inferred from Q1 & Q2)
   50%   Michaud,   31%    LePage   (E, inferred from Q2 & Q3)
   46%   Michaud,   31%    LePage   (F, inferred from Q2 & Q3)
A: PanAtlantic/SMS 15-21 October 2014. (See poll table at end.)
B: UNH survey center, 15-21 october.
C: PPP 8-9 September 2014.
D: PPP 22-23 October 2014.
E: Bangor Daily News/Ipsos 6-12 October.
F: Bangor Daily News/Ipsos 23-29 October.

No matter which of these polls we choose, we conclude that if Cutler had dropped out of the race, his 8.4% of the voters, even if they broke 75-25 for Michaud (more severely than any poll indicated), still would not have helped Michaud enough to overcome LePage's 4.9 percentage-point lead. (This conclusion actually contradicts the pairwise-poll average which was 51-49 for LePage over Michaud, but the present calculation, since 92% based on the Official Totals, trumps them for accuracy.) So Cutler was not a spoiler!

But, ironically enough, Michaud was a spoiler. That is because, if Michaud had dropped out, then pairwise polls unanimously show Cutler would have defeated LePage, and by an enormous estimated 58-42 margin. And the vast majority of Michaud's supporters are known to have preferred Cutler over LePage, and hence "should" have preferred that Michaud drop out of the race.

What makes this even crazier it that the same thing happened in the preceding 2010 election – Cutler then also was not a spoiler, but the Democrat Libby Mitchell was. Cutler got robbed two elections in a row by plurality voting system pathologies!

There also is a sense in which LePage was a spoiler. Namely, if LePage had dropped out, then instead of Michaud's 43.3% official total dwarfing Cutler's 8.4%, in fact pairwise data indicates Cutler would then have defeated Michaud 60-40.

So the media had it exactly wrong. But that does help explain why voters were so afraid to vote Cutler in the official election. There was an enormous amount of strategic/dishonest official voting in this election. Why we know that: just compare the "IRV round one" election (presumably a comparatively honest kind of plurality voting) versus the official plurality vote totals. What a difference! And: just over one-third of Michaud voters in our exit poll admitted (in the sense that their IRV rank-order sub-ballot said so) they really preferred Cutler; and just over one-fifth of our LePage voters did the same. The reason the voters were so dishonest obviously was strategic – they did not want to "waste their vote" on Cutler since they believed Cutler's winning chances were too tiny. That is a known major defect of the Plurality system – "vote splitting" and "spoiler" fears motivate massively intentionally-dishonest voting, which in this case shafted Cutler. Another defect of plurality which also is highlit by this election, is that it forces voters to express essentially the least amount of information they possibly could, about the candidates. A voter who likes two candidates, has no way to say so.

Poll questions asking pollees for their predictions of who would win:
       50% LePage,  34% Michaud,  2% Cutler.    (A)
       47% LePage,  33% Michaud,  5% Cutler.    (B)
A: UNH survey center, 15-21 october.  (See poll table at end.)
B: Bangor Daily News/Ipsos 23-29 Oct.

In contrast score and approval voting are immune to (do not exhibit) "vote splitting" and votes are not "wasted."

Crazy developments in this race

Cutler was endorsed by Senator & former Governor, Angus King (Independent). But on 28 October Cutler announced that his supporters should feel free to vote for LePage or Michaud (i.e. their 2nd choice) instead of for him (Cutler) to try to avoid a "spoiler" scenario. Which then evidently happened in great numbers. As usual, left unmentioned, both by Cutler and by the media, was the idea that with a better voting system, such crazy dishonest voting behavior would become unnecessary. Indeed, TV journalist Rachel Maddow complained extensively that she "didn't understand" what Cutler was saying. His statement was perfectly clear to me. Here, let me quote it, so you can see if you can understand it:

[From story "Cutler won't quit, but says supporters who think he can't win should pick LePage or Michaud," Portland Press Herald, 30 Oct 2014]    "I truly believe in democracy and the ultimate authority of voters to vote for whomever they want for whatever reason, and I don't think any voter, whether a supporter of mine or not, now needs or ever has needed my permission or my blessing to vote for one of my opponents. Nevertheless, I want to reiterate what I said six months ago: Anyone who has supported me, but who now worries that I cannot win and is thereby compelled by their fears or by their conscience to vote instead for Mr. LePage or Mr. Michaud, should do so."

Cutler also said to the Bangor Daily News [29 Oct 2014 "Cutler says victory is 'long shot,' tells supporters to vote accordingly"]

If you don't think that I can win, and you want to vote strategically, do that. But I'd be grateful for your support.

But all that still was not clear to Maddow, as she repeatedly said. Apparently Maddow was unsatisfied with Cutler's statement because he did not officially withdraw from the race in order to force his voters to desert him, whether they wanted to or not. Apparently only that would have satisfied her.

And she actually had the utter gall to suggest that by not actually withdrawing from the race, Cutler was committing some sort of sin or suffering from some kind of mental problem. But, again as usual, Maddow in her piece on this said not a single word about how the real problem was the voting system, nor that better voting systems exist which do not have the vote-splitting problem.

Angus King then (later on the same day, October 28) duly followed Cutler's advice and indeed publicly withdrew his endorsement of Cutler, switching it to Michaud. Which didn't work, since as we've seen, Cutler was not actually a spoiler; Michaud was the spoiler.

Michaud would have been the first openly-gay US state governor if elected, albeit he'd kept his homosexuality secret for 30 years and had (while a statehouse legislator) voted against gay rights / anti-discrimination bills at least 18 times. That made him, according to his voting record, one of the top 5 most anti-gay legislators in Maine (LePage also was anti-gay) – in contrast to Cutler (who would seem to be heterosexual, considering his wife and several children) who in the past had strongly vocally and monetarily supported gay rights!

Explaining this to a reporter shortly before the 2014 election, Michaud said "I could make excuses but I won't. I regret some of my earlier votes and feel fortunate that friends and constituents continued to talk with me and share their stories so that I could gain a greater understanding of the issue."

Cutler responded "I find it unforgivable. It is inconceivable to me how someone could have voted no to equality unless they put crass politics above principle. And that happened every time. I would never vote for him."

Michaud replied: "It's easy to say one thing, but [Cutler] has never had a history or a record... never ran for office [before]." (Which actually was false. Cutler had also run for the Maine governorship in the preceding 2010 election, finishing a fairly close second behind LePage, actually closer by a factor of 2 than Michaud's own second-place finish in this 2014 election.)

Cutler: "Oh, please. Many others who represented more conservative districts found that it was possible to vote yes."

Michaud also had opposed gun control and abortion rights, yet for the 2014 race got the endorsement of Planned Parenthood after he declared after he made it to the US Congress: "My views on abortion have evolved." Michaud also supported ENDA (Employment Non-Discrimination Act) once in congress, contrary to all his earlier votes.

LePage, during the lead-up to his 2014 re-election, made a fool of himself on nationwide news by trying to place Maine nurse Kaci Hickox in a 21-day quarantine because she'd been volunteering to help Ebola patients in Africa. Hickox had tested negative for Ebola and it was known that Ebola victims were not contagious unless showing symptoms (i.e. quite sick). But she was perfectly healthy. The courts overruled attempts by LePage and New Jersey Governor Chris Christie to quarantine Hickox. This would seem to win LePage the "fomenting unreasoning fear and imprisoning heroines as primary 'leadership' method" award.

Criticism related to the rather massive correction/reweighting needed for our exit poll

Unfortunately, the CES exit poll was conducted in geographic locations (map) that were almost pessimally chosen, resulting in a poor Maine-wide sample. We corrected this sample-bias via an unfortunately-large post-correction we'll explain.

Our 4 polling locations were in Portland (pop. 66318), Cape Elizabeth (pop. 9015), South Portland (pop. 25002), and Rockland (pop. 7297). The first three are in Cumberland County (Portland is the Cumberland County seat and Maine's largest city; South Portland is Maine's 4th largest) while Rockland is the seat of Knox County. All 4 lie in Southeast Maine. Michaud and Cutler were strongest in the South & North of Maine while LePage's strength lay in a wide belt in the mid-latitudes. Knox and Cumberland were the only two counties which Michaud officially won; LePage won all the others. And Cape Elizabeth was Cutler's home town.

In our defense, all of the largest towns in Maine, namely Portland, Bangor, Bar Harbor, Kennebunkport, and the capital Augusta, are in the comparatively pro-Michaud southern part of the state. We'd have had to travel to some comparatively remote places such as Greenville or Alexander to reach LePage strongholds.

The correction procedure we adopted was to assume/pretend that Maine voters consisted of 3 sub-populations, namely the Cutler, LePage, and Michaud voters. Each ballot in the CES poll consisted of 3 sub-ballots:

  1. asking who they officially voted for,
  2. Approval-voting (approve or disapprove each of the 3 candidates)
  3. Rank-order (Instant Runoff); rank the 3 in order from best to worst.

The correction was simply to reweight the ballots according to whatever it said in sub-ballot (A). The three weights were chosen so that the reweighted ballot set would exactly match the 3 official totals. A fine point: this was complicated by the facts that

As a consequence of these fine points, the reweighted ballot set yields plurality totals, when these 7 illegal/unusable A-subballots are left uncounted, which actually do not perfectly match the official plurality vote percentages:

             Official   Our Reweighted Ballots (if scaled so ∑=100)
   Cutler      8.43       8.26
   LePage     48.23      48.05
   Michaud    43.34      43.68

We also considered discarding those 7 bad ballots, in which case our reweighted ballot set really would yield an exact match; but the results of our analysis remain almost unaltered if we use that alternate reweighting, so we'll only discuss my main reweighting from here on. Also, note that the reweighting we chose is more "conservative" in the sense that it "hurts Cutler," thus the conclusion according to our poll that Cutler still wins, is "safe" i.e. cannot be attributed to this choice. The weights then were:

Cutler=0.101946,    LePage=0.479112,    Michaud=0.084542,    Refuse=0.221867,    DoubleVote(C&M)=0.093244.

In other words, if a ballot's A-subvote was for Cutler, then that whole ballot got weighted by 0.10195. Note that the largest weight 0.47911 is 5.67 times the smallest 0.08454.

A critic could, with some justification, object that the necessary reweightings were so unequal as to cast all the results of this exit poll into doubt! There are two underlying reasons this critic's view could be justified:

  1. The reweighting introduces considerable extra "statistical sampling noise" into the poll data, both because high-weighted ballots are rare so that our effective sample size is reduced, and because the reweighting factors themselves are "noisy" which is a second source of extra-error. We have a good answer to this objection, which is that we did a computer-aided "bootstrap" statistical analysis, which of course knew about both these kinds of noise and correctly reckoned them, and it turned out that most of our poll results still had quite high statistical significance even after correctly-accounting for this objection (details later).
  2. Perhaps the model that there are 3 voter subpopulations in Maine, is poor. For example, maybe there really were two different types A & B of LePage voters, and our poll, due to its poor geography, oversampled type A (which liked Cutler) relative to type B (who liked Michaud, but our poll was unable to perceive them). If so, our reweighting procedure would not have worked. The reweighting procedure works only to the extent the "three subtypes" model is a good approximation to the truth. But this particular speculation seems unlikely because the middle areas of Maine our poll failed to sample, historically have been pro-Republican, pro-Independent, and anti-Democrat. Therefore, if our poll erred about LePage voters of subtypes A & B, it probably did so in the other direction, i.e. Cutler would have beaten Michaud pairwise by an even greater margin than our poll thought. So again, our poll's "Cutler should've won" conclusion seems "safe."

We do not have an optimal answer to the second criticism, but do have some further mildly decent indirect answers. First of all, the large pairwise victory of Cutler over LePage is not just coming from our exit poll. It also was found by 2 other, independently conducted, pre-election polls by Bangor Daily News/Ipsos during 6-12 and 23-29 October, as well as by the Frederick poll 16-17 September (we'll soon discuss these).

Regarding our even poll's even-larger claimed pairwise victory for Cutler over Michaud, unfortunately apparently no other pollster directly asked about this head-to-head contest so we cannot get direct independent confirmation. Nevertheless, there were other pollsters who asked questions whose answers allow us to make deductions about this question. (For instance we already saw independent polls indicating Cutler was the approval-voting and score-voting winner.) A poll paid for (and released to the press) by the Cutler campaign and conducted by FrederickPolls (16-17 Sept. 2014) asked:

A. "Which candidate for governor would be your second choice?"
     Cutler=54%,  Michaud=19%,   LePage=6%,   None/Won't say=22%
         which if the "Don't Know"s are ignored becomes
          Cutler=69.2%,  Michaud=24.4%,   LePage=7.7%
B. "In this year's election for Governor between Paul LePage, the Republican, Mike Michaud, the
 Democrat, and Eliot Cutler, an Independent – which one are you most likely to support?"
     LePage=35%,  Michaud=35%,   Cutler=19%,  Undecided=11%.
C. "If the election for Governor were just a choice between Paul LePage and Eliot Cutler, 
 which one would you support?" 
     Cutler=50%, LePage=40%, Undecided=10%.

Analysis: Slightly perturb answer B to become L=35, M=36, C=19, U=10 (we need this perturbation to make the following math come out exact). Then answer C is explained if Michaud's 36 points split 5→LePage and 31→Cutler as second choices. And then answer A is explained if LePage's 35 points from answer B split 23→Cutler. LePage's remaining 12 points have to go X→Michaud and (12-X)→Undecided. We cannot tell the value of X, but one might guess X≈5 so that the Michaud:Undecided ratio is the same within this 12, as within the whole population sample in question A. As a result, in the Cutler-Michaud pairwise match within this Frederick poll, Cutler necessarily gets 19+23=42 points, while Michaud gets 35+X points, which if X=5 is 40 yielding a 42-40 win for Cutler over Michaud. That was not exactly a huge independent confirmation of a Cutler>Michaud pairwise victory, because

     Frederick cross-examined his poll responses and claimed that 
  "Cutler is the second choice of 84% of Michaud voters and 42% of LePage voters,"
  each "far more than any party-label candidate or even 'none'."
  (I further deduce:  LePage is the second choice of 14% of Michaud voters.)

That 42% figure contradicts our 23/35=66%. It appears that the only explanation of this contradiction is either that Frederick made a mistake, or that his voters contradicted themselves.

The Bangor Daily News/Ipsos poll (6-12 Oct.) asked Likely Voters

A. "Thinking about the upcoming general election in November of this year, 
   if the election for Governor of Maine were held today, for whom would you vote?"
         Cutler=16%,   LePage=36%,   Michaud=42%,   Don't Know, Other=6%
B. "If the election for Governor of Maine were held today and Eliot Cutler 
   (Independent) was not on the ballot, for whom would you vote?"
               LePage=41%,   Michaud=50%,   Undecided=9%.
C. "If the election for Governor of Maine were held today and Mike Michaud 
   (Democrat) was not on the ballot, for whom would you vote?"
              Cutler=49%,   LePage=38%,   Undecided=13%.

Analysis: It would seem Cutler's 16 in answer A evidently split 8→Michaud, 5→LePage, 3→Undecided to yield answer B. And Michaud's 42 in answer A split 33→Cutler, 2→LePage, 7→Undecided to yield answer C.

The second Bangor Daily News/Ipsos poll (23-29 Oct.) asked Likely Voters

A. "If the election for Governor of Maine were held today, for whom would you vote?"
         Cutler=13%,   LePage=42%,   Michaud=42%,   Don't Know, Other=6%
B. "If the election for Governor of Maine were held today and Eliot Cutler 
   (Independent) was not on the ballot, for whom would you vote?"
               LePage=46%,   Michaud=48%,   Undecided=5%.
C. "If the election for Governor of Maine were held today and Mike Michaud 
   (Democrat) was not on the ballot, for whom would you vote?"
              Cutler=53%,   LePage=40%,   Undecided=7%.

Analysis: It would seem Cutler's 13 in answer A evidently split 6→Michaud, 4→LePage, 3→Undecided to yield answer B. Now note LePage actually got fewer votes with Michaud out of the race because evidently some number (at least 2) of LePage's 42 points of voters were being strategic/dishonest in answer A, and really preferred Cutler, which they told Ipsos's pollster in answer C. Assuming it was 2, then Michaud's 42 in answer A split 38→Cutler to yield answer C. Probably it was closer to 4, in which case Michaud's 42 in answer A split 40→Cutler to yield answer C.

In the light of all three of those other polls, the only way we can consistently explain the CES exit poll's claim that Cutler would have beaten Michaud pairwise by 60-40 is by postulating that a large fraction of the Michaud official votes were strategic/dishonest, and they honestly preferred Cutler. And in fact, that is exactly what the CES exit poll does claim: just over 1/3 of its pollees who said they'd officially voted Michaud, said that vote had been dishonest and they really preferred Cutler, see below.

Also note: The Frederick second-choice poll result (undecideds removed) when combined with the Official Maine Results indicates that over 75% of Michaud & LePage voters (merged) prefer Cutler as their second choice. That plus the fact from our exit poll that over a third of Michaud voters admitted they really preferred Cutler... also makes it difficult to imagine Cutler losing a Cutler-Michaud head-to-head match. Since polls are released selectively by campaigns this Frederick poll has somewhat less credibility, but as far as it goes this supports our C/M pairwise claim.

Our poll's reweightings about Cutler & Michaud voters are quite close (ratio=1.21) hence our poll is less questionable with respect to them alone. Our raw (unweighted) poll found:

Conclusion: Sorry, this has been a bit involved. Our claim Cutler beats LePage pairwise is indisputable since found by many polls with high confidence. Our claim that Cutler beats Michaud pairwise is disputable, since only found by one poll – ours – with no other polls directly asking this question. However, we've got several reasons to believe our poll has not deceived on this question.

What's a "Bootstrap statistical analysis"?

Our exit poll featured 696 respondents (plus one 697th who refused to fill out any of the three sub-ballots but did "write-in" another candidate's name; we ignored this). "Bootstrapping" consisted of, 100000 times,

  1. Taking a random subset of only two-thirds of them (which is 464),
  2. Finding the reweighting factors causing that 464 to agree with the three official plurality-totals,
  3. and then, using the resulting set of 464 weighted ballots, performing pairwise, IRV, and Approval voting elections.

(Why we chose 2/3 as the fractional subset size.) The point was this allowed us to assess the effect of statistical "noise" and to attach a numerical confidence to any conclusion from our poll (subject to belief in the reweighting assumptions). For example, the conclusion "Michaud beats LePage pairwise" is not at all significant – the opposite happens in 33% of the simulations. But most of the conclusions are highly significant, such as "Cutler is the approval winner" (99.7%), "Cutler beats Michaud pairwise" (100.000%), "Cutler beats LePage pairwise" (100.000%), and hence "Cutler is the Condorcet winner" (100.000%). Intermediate-significance conclusions include "Cutler is the IRV winner" (91.4%), as opposed to Michaud (6.3%) or LePage (2.2%); "Michaud is the IRV loser" (91.4%) as opposed to Cutler (8.6%) or LePage (0.003%).

Score voting polls

All of these polls used this 4-level scale

    very favorable,  somewhat favorable,  somewhat unfavorable,  very unfavorable
 which we have listed in this order and regarded for average-score computing purposes as
         3                   2                    1                    0

The polls were:

PanAtlantic/SMS, phone, 23-29 sept (although called the "October omnibus poll"):
 LePage     VF=28.5 SF=21.0 SU=13.0 VU=34.5  DN= 2.1  AverageScore=1.448
 Michaud    VF=22.0 SF=28.5 SU=21.5 VU=18.8  DN= 9.3  AverageScore=1.591
 Cutler     VF=13.3 SF=39.5 SU=15.8 VU=10.5  DN=21.0  AverageScore=1.703
Note LePage's 2-humped distribution, which poses a problem for "majority judgment" type 
median-based methods. Majority Judgment in this case, however, clearly awarded the win to Cutler.
Bangor Daily News/Ipsos, online, 6-12 october:
                VF  SF   SU  VU     average
 Mike Michaud   31  32   19  18      1.76
 Eliot Cutler   19  46   25  10      1.74
 Paul LePage    20  22   13  45      1.17
Again note LePage's 2-humped distribution which poses a problem for "majority judgment" type 
median-based methods. Majority Judgment in this case, however, clearly awarded the win to Cutler,
even though Michaud slightly ahead of Cutler on average score.
Bangor Daily News/Ipsos, online, 23-29 october:
           VF  SF  SU  VU  average
Cutler     15  43  28  14   1.59   
Michaud    27  28  23  22   1.60
LePage     27  18   9  46   1.26
Michaud slighly ahead of Cutler on average score, but Cutler is MJ winner.

Michaud wins 2 and Cutler 1 of those three score polls, but Cutler wins the super-poll got by averaging all three. If we were to change the four score levels from (3,2,1,0) to (+2,+1,-1,-2) with zero missing, or to (+1,+1,-1,-1) which in a sense converts score to approval voting, then either way Cutler then would win all three score polls using average score. There also was a fourth score-style poll (same scale) conducted by Rasmussen Reports 7-9 October online. But I do not have their data (hidden behind paywall).

Voter Error & Dishonesty Rates with different voting systems

I. Officially voted for for somebody different than top rank-order preference:

Total ballots: 696.  Official votes: C=80, L=99, M=510.
Of 510 M voters, 174=34.1% top-ranked somebody else, namely C=171 and L=3.
Of  99 L voters,  21=21.2% top-ranked somebody else, namely C=19  and M=2.
Of  80 C voters,   1=1.25% top-ranked somebody else, namely M.
In all: 196=28.2% of the voters (unweighted raw numbers) officially voted dishonestly.

II. Officially voted for somebody they did not approve (and approved at least one):

Of 510 M voters, 42=8.2%  approved C alone.
Of  99 L voters,  5=5.05% approved C alone.
In all: 47=6.75% of the voters (unweighted raw numbers) officially voted for somebody
other than Cutler even though Cutler was the only candidate they approved.
  Type-I and type-II dishonest simultaneously:  45=6.5% of the voters
officially voted for somebody other than C even though C was simultaneously
the only candidate they approved and the candidate they top-ranked.

III. Illegal plurality ballots:

5 undervotes or refusals to divulge;  also 2 overvotes.   
That's 0.29% overvotes  and  1.0% illegal in all (under+over).
Comparison with official figures:
2010 Maine Gov official totals:  1.3% spoilage rate (included both
under- & overvotes and all other spoilage modes).
2014 Maine Gov official 
totals:  not announced yet as of end of Nov. 2014.  Update July 2015: 
Maine now claims the final totals were Cutler=51515, LePage=294519, Michaud=265114,
Other=79, Blank/Spoiled=5740, TotalCounted=616967=51515+294519+265114+79+5740.
The official spoilage rate is therefore 5740/616967=0.93%.

IV. Illegal Approval ballots: zero.

V. Illegal IRV ballots:

In total in 6=0.86% of IRV ballots they tried to vote, but failed to give a valid
top-rank choice. In 5 of those 6 cases, some voting machines could have succeeded in
interpreting anyhow, by interpreting the 2nd-ranked or 3rd-ranked choice as "top" ranked.
But this "correction" is dangerous.

VI. IRV ballots where we suspect they tried to vote... in reverse order! (But our guess about their intent might be wrong!) In other words these ballots were probably wholy or partly caused by voter-errors, even though they were not "spoiled," i.e. would have been counted.

For 5=0.72% of the 696 total IRV ballots, the ballot was fully legal,
but probably sometimes not what they intended, e.g. seemed to be in reverse, or
partially-reversed, order versus what would have been expected from their
official-plurality & approval sub-ballots.

VII. The combined effect of the CES poll reweighting and the dishonest official voting behavior led to the surprising output (pointed out by Mark Frohnmayer) that the CES exit poll actually claimed LePage had fewer approvals than official plurality votes!

How that happened:
(A)--There were 6 pollees who plurality-voted LePage but did not approve LePage,
in 5 cases apparently figuring now was their chance to honestly bullet-approve Cutler;
in the 6th case they approved nobody.

(B)--LePage despite (A) got more approvals than plurality-votes in the raw
ballot data set, due to 17 non-LePage voters approving him.   However, because the non-LePage
voters had so low weight in the weighted ballot set,  (A) overwhelmed (B)
resulting in LePage getting fewer weighted approvals than weighted plurality votes.

Conclusions: This CES exit poll supplies yet more confirmation of the usual finding that under equal conditions, IRV ballots cause the greatest voter error rates (6+5=11 ballots), plurality voting intermediate rates (2 to 7), and approval voting the lowest (0). Our finding about "erroneous but legal" IRV ballots is probably new, since we are unaware of any prior multi-voting-system study where the voters tried all three voting systems at the same time; and in the absence of such multi-data these errors would have been undetectable. Obviously "order reversal" errors are very serious, at least about twice as serious as just discarding an IRV ballot.

Other reweighting schemes

We also considered bilinear ballot-reweighting schemes with weighting based both on who that voter officially voted for and on which town that ballot was from. I will not describe these schemes here because I am unsure whether they are a good idea at all. I'll just remark that they seem to yield the same results we've claimed based on our simpler pure-linear reweighting, only more strongly.

Links to underlying poll data

PollSample(s)Field date in 2014Methods
Previous 2010 Maine governor election 580538 voters2 Nov. 2010official election
Frederick (for Cutler campaign) local plaintext copy 400 Likely Voters16-17 Sept86% landline, 14% cell phone inteviews
PPP (Public Policy Polling) for MCAF1059 LVs8-9 Sep.interviews
PanAtlantic/SMS400 Maine LVs age≥1823-29 Sep.random telephone
Bangor Daily News/Ipsos 1004 Maine residents, including 903 registered voters and 540 likely voters ages≥18. 6-12 Oct.online
UNH/Portland Press-Herald (PPH published 25 or 26 Oct.) 639 LVs15-21 Oct.random phone
PanAtlantic/SMS400 LVs age≥1815-21 Oct.random phone
PPP for Maine Conservation Action Fund660 LVs22-23 Oct.interviews
Bangor Daily News/Ipsos 1005 residents including 946 registered voters, 488 of them likely voters 23-29 Oct.online
Maine Government, New York Times ≈610000 actual voters 4 Nov.Official election results
CBS News 1008 actual voters (exit poll) 4 Nov.In person
Center for Election Science 696 actual voters (exit poll) 4 Nov.In person; they filled out paper "ballots"

None of the pollsters in the above list used "robocalls," i.e. all interviews were human↔human. But the Bangor Daily News/Ipsos polls used online methods, which are at present usually a worse (although cheaper) polling method. Fortunately Ipsos has a lot of experience with those methods; inexperienced users of online polling methods produce "results" truly horrible to behold.

I (Warren D. Smith) did the majority of the data analysis for the CES exit poll, which was organized by Nevin Brackett-Rozinsky and carried out by him plus several volunteers in 4 locations in Maine. You can get our raw data, weighted and unweighted totals (also includes raw data in 2nd format), C source code for bootstrap statistical analysis program (also includes some output in comment at end).


Update: LePage's performance as governor

Return to main page