Nashua NH, 8-11 Oct. 2015. Straw poll conducted for RLC by Center for Election Science. ≈777 voters. Sample ballot below which we've here filled in with the poll results. (See also: actual ballot image jpg.)
The blue-green "mean rating" column and "total" row of course were not actually present on the ballots; we have added them for results-display purposes. The boxes on each ballot were either left empty or checked (✓) by voters; we here have filled each box with the percentage of voters who marked that box. The rating "_" was not included on the ballots, which of course only allowed 0,1,2,3,4,5 as ratings; we have included the "_" column to count voters who left those all blank i.e. did not provide any numerical rating for that candidate. We counted blanks as 0 for purpose of determining score-voting winners. (jpg1, jpg2)
1. Which one candidate would you vote for? 2. Which candidates would you vote for if you could vote for as many as you want?
3. How would you rate each candidate? (0 means worst and 5 means best. Omitted candidates are given a 0 rating.)
You may write in up to two candidates. Your response will remain anonymous.
|1. Vote for one (776)||
2. Vote for
one or more (779)
|3. Rate each candidate (769)|
|Winners||Plurality: Paul||Approval: Paul||Score: Paul|
The ordering of the candidates shown was not the same as that on the preprinted ballots (jpg) because for display purposes we have reordered the lines to sort them into descending order reckoned by approval count.
Voters and winners: There were 776 plurality voters, defined as ballots containing at least one vote for at least one candidate. There were 779 approval voters, defined as ballots approving at least one candidate. There were 769 score voters, defined as ballots scoring at least one candidate nonzero; their mean score (on-ballot candidates only) was 0.95 on an 0-to-5 range. Rand Paul won reckoned by any of our three voting systems (all agreed).
The 16 candidates polled here were the 15 Republican contenders who'd participated in televised debates up til then, plus Mark Everson [Businessman (vice chairman of AlliantGroup, a tax firm), the 46th IRS Commissioner (2003-2007), and the CEO of the Red Cross (brief stint in 2007). http://markforamerica.com, https://en.wikipedia.org/wiki/Mark_Everson, http://www.nationalreview.com/article/422302/mark-everson-2016-presidential-candidate]. All are or were US senators or state governors except for Donald Trump, billionaire; Carly Fiorina, former CEO of Hewlett-Packard co; and Ben Carson, former head of pediatric neurosurgery at Johns Hopkins – all three of whom had never held any government office – and Everson. The three top-approved candidates receiving votes as "write ins" who were not preprinted on ballots were:
The first write-in row of the table pertains to Fulbright; all the non-Fulbright write-ins have been combined in the second write-in row.
Why were these results so different from nationwide polls? Because the sample (i.e. the voters) at the RLC convention was highly unrepresentative of Republicans nationwide:
At a first approximation, the pollbots all voted for their man with top rating, rating everybody else zero. That simplistic behavior makes it pretty easy for you (if you want) to detect and mentally-remove them from the results. But actually, there were exactly 464 voters who rated (Paul=5,Cruz=0) or (Cruz=5,Paul=0) and of them, exactly 366 rated everybody else 0 (of whom 240 were PaulBots and 126 CruzBots). These counts suggest some of the pollbots may not have been quite that simplistic, or that our estimate of 600 pollbots was too large.
If we remove these 240 PaulBots and 126 CruzBots from the electorate, then Paul's mean score drops to 3.00, Cruz's mean rises to 3.63, and Carson's rises to 2.68. This suggests that Paul's "victory" in this poll might have been caused by his strategic-voter conspiracy; perhaps Cruz really should have won.
Meanwhile, the non-pollbots approved varying numbers of candidates, averaging about 4. Of course all four subsets A,B,C,D were highly unrepresentative of the nation, although (C) presumably came the closest to being representative.What lessons do we learn from this poll? This is one of the first, perhaps even the first, polls in which the voters used all three voting systems simultaneously. That allows us to compare them in action.
It is clear the plurality vote was massively distorted. The approval and score votes, however, arrived at approximately the same conclusions, suggesting they both performed pretty well. Specifically the same ordering was output by both except for adjacent-transposition of (Trump,Jindal) and a cyclic shift of the last 3 finishers (Graham,Gilmore,Everson).
This also is one of the first, perhaps even the first, polls in which it was fully acknowledged by all that there was a huge attempt to manipulate the election to make Paul and Cruz win (two opposed strategic-voter teams, each organized and recruited). The organizers of the conference even encouraged this in an attempt to get more attendance-fee income – oh sorry, actually to "provide a test of the organizing and recruiting skills of the candidates." (Perhaps a better wording would have been "bribing skills" but we digress. Actually even as I write this in Oct. 2015, some of Rand Paul's top campaign workers are under indictment right now for a bribery that was part of his father Ron Paul's previous campaign for US president.)
This allows us to assess the impact of massive organized strategic voting efforts on the three voting systems. First of all, it is obvious that plurality voting was reduced to smoking rubble by the two strategic voter teams.
Second, the so-called "majority judgment" voting scheme was recommended by Balinski & Laraki, based not on the average rating, but rather on the median rating (plus a certain tie-breaking scheme if necessary). The top alleged reason for preferring median was that it would provide a better defense versus strategic voting, and/or discourage voters from even bothering to be strategic at all.
We here are in a position to perform the first-ever real-world test of that claim in an election with (as all involved acknowledged) massive organized strategic voting conspiracies.
The answer is about as clear as it could possibly be (albeit only based on this one election): MJ massively failed to provide such a defense. Plain average bent, but did not break, under the massive strategic assault. But MJ totally collapsed, claiming Paul had a "perfect" 5.0 median score – the same as presumably would have been obtained by "God" – while all rivals had "worst possible" median score 0.0 (the same as would have been obtained by the "Devil"), except for Cruz with 4.0.
("Paul=5, all others≤X" also holds regardless of whether we regard blank as 0 score, or as any other value≤X.)
Meanwhile with ordinary average-based range voting, Paul only got average score 3.17 even with all strategists included. Without strategy, I believe Paul's average would have been about 3.6±0.6. Cruz got average score 2.77 with all strategists included. Without strategy, I believe Cruz's average would have been about 3.5±0.6. Note average also outperforms median reckoned by the difference 0.4 between Paul's and Cruz's averages versus the difference 1.0 between Paul's and Cruz's medians. (This remains true even if we multiply the former by 5/3.17.)
So there is no question plain average outperforms MJ as a defender against strategic voting, at least in this election.
The problem in this election is Paul & Cruz imported such large numbers of "pollbot" voters. Median initially sounds like a good defense, because it and trimmed means sort of "go with the majority" and trim off the outliers. But in this election, the majority of voters were saying Paul=5 (most of them strategists) and therefore as far as median was concerned essentially every honest voter was an "outlier" who "should be trimmed off"! So MJ's "defense" actually made it worse, not better.
Meanwhile plain average was hurt little, because the strategists and counterstrategists largely canceled each other out in terms of their effect on Paul's and Cruz's averages (but the counterstrategists unfortunately had no effect on Paul's median). And actually I suspect this kind of near-cancellation is a common state of affairs, not merely a lucky break only just for this election. Which is why I suspect in the real political election world, medians are not an improvement even though Balinski & Laraki propagandize they are. And while the evidence from real world studies like this is small (since few of them) but to the extent it exists, it supports my view quite well.
It's amazing how MJ keeps performing demonstrably worse than plain averages when tried in the real world, and how the pro-MJ propagandists then often cover their eyes. But this particular poll appears to be the worst MJ-suck yet... since it's hard to be worse than worst-possible.
We also can consider the effect of trimmed means with different trim-off fractions for this election.
Want the raw data? Here is a computer program which contains the raw data, and uses it to compute various things.
Return to main page