Apparent connection between approval voting and candidate-honesty ============Warren D. Smith====Sept. 2016======================== SUMMARY: It appears that the most-approved candidates in approval-style polling coincide with the most-truthful candidates according to fact-checking organizations. Previously, it had not been clear why real-world approval voters approved or disapproved candidates or what approval "meant" to them. This observation may go a long way toward clarifying that, although further data, and possibly then some refined rewording of the Hypothesis, are desirable. But our present data already is good enough to yield at least three "nines" worth of confidence. 2016 USA PRESIDENTIAL ELECTION I am writing this in September 2016 in the midst of the 2016 US presidential race. I noticed an interesting possible connection, see the data table below. Candidate H0 H1 H2 H3 H4 H5 FC ApprovalRCP ApprovalHP AppNum Hillary Clinton(DN) 2 11 15 22 28 22 249 42.3/52.3 41.6/56.2 0.775 Bernie Sanders(D)* 0 11 17 20 39 13 106 50.0/36.8* 55.8/35.6* 1.463 Elizabeth Warren(Dx)* 0 0 11 11 33 45 9 28.0/26.3* 25.2/27.1 0.997 Joe Biden (Dx) 5 13 15 28 21 17 75 45.7/37.7* 51.3/36.6* 1.307 Martin O'Malley(D) 0 6 17 56 22 0 18 17.5/27.0 -- 0.648 Donald Trump(RN) 18 35 18 15 11 4 250 38.8/58.6 38.3/59.0 0.656 Ted Cruz(R) 7 27 31 13 16 7 114 29.7/54.8 23.8/54.8 0.488 John Kasich(R)* 5 13 14 16 28 25 55 39.1/30.0* 37.4/37.7 1.148 Marco Rubio(R) 3 15 23 22 24 13 142 36.3/40.3 34.9/46.0 0.830 Jeb Bush(R) 3 6 22 22 30 14 79 30.8/52.4 30.6/53.5 0.580 Rand Paul(R) 6 16 14 20 24 22 51 25.7/41.0 27.7/44.0 0.628 Chris Christie(R) 8 17 10 28 19 21 102 29.8/44.5 30.4/48.2 0.650 Rick Santorum(R) 8 27 20 22 12 10 59 24.0/42.0 19.9/46.3 0.510 Rick Perry(R) 11 18 18 25 14 15 169 21.0/39.3 23.6/43.5 0.538 Scott Walker(R) 6 24 15 19 23 13 172 23.5/29.3 33.7/37.4 0.852 Bobby Jindal(R) 0 10 0 40 40 10 10 18.7/30.3 17.1/35.9 0.547 Ben Carson(R) 14 43 25 11 7 0 28 -- 36.3/45.4 0.830 Lindsey Graham(R) 0 17 17 33 17 17 12 -- 16.3/42.1 0.387 George Pataki(R) 20 80 0 0 0 0 5 -- 14.8/37.9 0.391 Mike Huckabee(R) 10 12 29 22 7 20 41 -- 28.5/45.0 0.633 Carly Fiorina(R) 9 23 23 18 14 14 22 -- 24.6/39.0 0.631 Jill Stein(G) 0 0 100 0 0 0 1 -- 15.4/28.9 0.532 Gary Johnson(L) 0 33 33 0 33 0 3 -- 24.1/28.9 0.834 KEY: FC=Number of "facts" claimed by that candidate and checked by "politifact.com, winner of pulitzer prize" as of 13 Sept. 2016. Note: for the following people politifact did not check enough facts; therefore I have added all the facts checked by FactCheck.org for that candidate, along with numerical ratings provided by me, to their totals: George Pataki, Jill Stein, Gary Johnson, Elizabeth Warren. H0=Percentage of those facts given the worst "pants on fire" rating. H1=Percentage given 2nd worst "false" rating. H2=Percentage "mostly false." H3=Percentage "half true." H4=Percentage "mostly true." H5=Percentage "true." Parentheses: (D/R)=Democratic/Republican party candidate. N=eventual nominee of that party. x=never actually ran. ApprovalRCP=Favorable/Unfavorable rating from USA-wide poll averages compiled by RealClearPolitics.com, as of 13 Sept. 2016, mostly from http://www.realclearpolitics.com/epolls/2016/president/favorable_unfavorable.html ApprovalHP=same thing, but instead compiled by "HuffPost Pollster" as of 15 Sept 2016. AppNUm=numerical computation of the Fav/Unfav ratio (averaged if have two). Asterisks: Candidate-name awarded a star (*) if he/she manages to attain 51% or greater "true" or "mostly true" rating from politifact.com. Approval awarded asterisk if candidate manages to get more "favorable" than "unfavorable" ratings. (HP and RCP disagreed on approval-star-worthiness for Warren and Kasich, but we shall grant them the benefit of the doubt.) Amazingly enough, the two kinds of asterisks happened to coincide. I.e, the only two candidates (Sanders & Kasich) who got 51%-or-above honesty ratings from politifact, ALSO happened to be the only two who were more approved than disapproved by the US public. Is this just a fluke? Well, there are 21 candidates listed here, hence 21*20/2=210 possible candidate-pairs. If honesty and approval were completely unrelated, the chance the two pairs would coincide would be 1/210. (Actually, less, since it did not have to be a "pair.") Therefore, we conclude with confidence 99.5% versus the null hypothesis that this effect is real. If we include also the two non-candidates (but the press acted as though they were likely to be candidates for quite a while) Warren and Biden, then Warren also would have gotten a double-*, albeit based on only 9 checked facts. (She has the greatest True & MostlyTrue rating among those listed.) But Biden would have only a single *, although he did come close to getting the 51%-or-above honesty star with 49%. (Hillary Clinton and Bobby Jindal came even closer with 50% each.) So, all in all, if we consider adding Warren and Biden and lowering the honesty threshhold to "above 48%" we then would have 6 honesty stars and 4 approval stars among 23 candidates and close-to-candidates, with the approval stars forming a strict subset of the honesty stars. Is _this_ a mere coincidence? If the two kinds of stars were wholy unrelated then considering that the number of possible 6-element-subsets of 23 pseudo-candidates is 23*22*21*20*19*18/720=100947 and the number of such subsets that include one particular quadruple is 19*18/2=171, this would again be a low-probability (specifically chance=171/100947) event, yielding 99.8% confidence versus the null hypothesis that this is a real effect. If on the other hand we kept our honesty threshhold at "51% or above" then we would have 3 honesty-stars and 4 approval-stars and 99.8% confidence. CONCLUSION: Depending how we calculate it, we get 99.5 to 99.8% confidence based on this data that "candidate truthfulness" and "approval rating" are related. This could be an important insight into approval voter behavior. Meanwhile, the actual voting system used by the Dem/Repub parties to select their nominee -- which was NOT approval voting -- produced results (e.g, their nominees Clinton and Trump) far less correlated with honesty! Trump was the single most-dishonest person tabulated among those with at least 30 checked facts, with only 15% True/MostlyTrue (lowest by far) and 53% False/PantsOnFire (highest by far). Clinton was about the 4th most honest among the 21 listed, but could not quite manage to reach 51% True/MostlyTrue. Also, if one examines the scores of candidates in plurality-style polls and the official primary votes... then again, one sees far less correlation with candidate truthfulness than one would find with approval-style voting -- indeed I doubt that any statistically significant connection is discernible. Might the cause of all this be some sort of bias from that nasty evil organization "politifact.com"? Certainly their ratings are somewhat subjective (and I do not always agree with them, and different fact-checking organizations sometimes disagree), and their choice of which "facts" to check may be somewhat manipulable. To try to protect ourselves from that worry we alternatively could consult other independent fact-checking organizations. The book "Deciding What's True: The Rise of Political Fact-Checking in American Journalism" by Lucas Graves (Columbia University Press, Sept 2016) is helpful for learning about that. Apparently the fact-checking entity is a fairly recent invention and it may have been an American invention: Snopes.com - founded 1995 Spinsanity - founded 2001, over 400 facts checked during 2004 US election, but ended FactCheck.org - founded 2003 Annenberg public policy center at Univ. Pennsylvania; produces about 5 fact checks per week, supported by foundation philanthropy. Politifact.com - founded by St Petersburg Times / Congressional Quarterly in 2007, sold in 2009 to The Economist Group; checks about 20 facts per week making it the most productive fact checker (at least among USA-based ones if not worldwide). Washington Post fact checker - started September 2007, ended after 2008 campaign season, but revived in January 2011. Checks about 1 fact per day. CNN.com's "reality check" - not sure of its years of operation, but it has been checking Clinton & Trump during the latter part of the 2016 presidential campaign. Internationally, we have "Pagella politica" and "Il politicometro" (both Italy) The Guardian "reality check" (UK) FactChecker.in (India) MorsiMeter (Egypt) "El poligrafo" a fact-checking feature in "El Mercurio" (Chile) Africacheck.org (Africa) Le Monde ran (but discontinued?) a fact-checking feature "Les Decodeurs" (France). French fact checkers: Decodeurs: http://decodeurs.blog.lemonde.fr/ Founded by Le Monde in 2009 Les Pinocchios: http://tempsreel.nouvelobs.com/politique/les-pinocchios-de-l-obs/ OWNI veritometre: http://owni.fr/2012/02/16/veritometre-factchecking-presidentielle/ http://owni.fr/2012/05/03/veritometre-debat-hollande-sarkozy/ etc Desintox: http://www.liberation.fr/desintox,99721 Founded by Cedric Mathiot for Liberation.fr in 2008 Der Spiegel (Germany) also has perhaps the largest fact-checking team in the world. Unfortunately for our purposes, "Factcheck.org" does not provide numerical truth-ratings for checked facts nor summary ratings for candidates; they just provide lists of individual facts, usually claimed by or about candidates. (Anybody willing to work could tally them up and numerically rate them, however; and they do index their facts by candidate-name and provide "annual reviews.") The same problems pertain to "Fact checks of the 2016 election" by New York Times and most or all of the international and other groups. (Further, e.g, MorsiMeter focused solely on Morsi's campaign promises and not on fact-checking any of his rivals, making it nearly useless for our purposes.) "Fact checker at the Washington Post" does provide numerical "Pinocchio ratings" with occasional summary tallies. Here is the summary they published on 15 July 2016. Candidate 4P 3P 2P 1P GC FC Donald Trump 63 21 10 2 2 52 Hillary Clinton 13 36 30.5 5.5 14 36 KEY: 4P = percentage of facts regarded as "4 Pinocchios" -- outright whopper lies. 3P = "3 Pinocchios" -- "mostly false." 2P = "2 Pinocchios" -- "half true." 1P = "1 Pinocchio" -- "mostly true." GC = the coveted "Gepetto checkmark" -- completely true. FC = total number of facts checked for that candidate. As you can see the Washington Post agrees with Politifact that Clinton is far more truthful than Trump, but appears to have a tougher standard in that both Clinton and Trump score worse at Washington Post than they score at Politifact. (I also think the Washington Post checks fewer facts than Politifact, but the ones it checks, it checks more thoroughly.) The Washington Post went further by claiming that Trump was not merely the least-honest candidate in the 2016 race (that conclusion agrees with Politifact), but actually among all major politicians in the entire history of the Washington Post's fact-checking feature. FactCheck.org also agreed, calling Trump "The king of whoppers" on 21 Dec 2015 and noting "in the 12 years of FactCheck.org's existence, we've never seen his match." CNN's "reality check" one day after the first Clinton-Trump presidential debate on 26 September checked claims made by both during that debate and my tally of their checks (version of noon 27 Sept) was Candidate IC FA TB MT TR FC Donald Trump 0 14 2 0 3 19 Hillary Clinton 0 0 1 5 6 12 KEY: TR=true; MT="mostly true"; TB="true but misleading"; FA=false; or IC="it's complicated." And CNN concluded "From misrepresentations to half-truths and flat-out lies, Trump has talked around and away from the truth more brazenly than any major party presidential nominee in modern political history" agreeing with all the other fact-checking bodies. This debate also was fact-checked by Washington Post (https://www.washingtonpost.com/news/fact-checker/wp/2016/09/27/fact-checking-the-first-clinton-trump-presidential-debate) and by FactCheck.org (http://www.factcheck.org/2016/09/factchecking-the-first-debate). ============================================================== NOTE ABOUT GARY JOHNSON The data tabulated above shows that (at least as of mid-September 2016) the most-approved presidential candidate still running (based on approval/disapproval ratio) was NOT Hillary Clinton, and NOT Donald Trump. It was the Libertarian party's nominee (and former governor of New Mexico) Gary Johnson. Johnson was not permitted to participate in the Presidential Debates and got very little media attention. He will undoubtably get only a tiny percentage of the official vote on election day. Nevertheless it appears the USA wants him as president more than any rival. While it is true that Gary Johnson currently has the highest approval/disapproval ratio, it is unclear how seriously that should be taken because the US public is poorly informed about GJ compared to HRC and DJT. If our elections were using approval, then the media and his rivals would scrutinize GJ a lot more, everybody would be better informed, and then his approval might change a lot in one or the other direction, perhaps invalidating his frontrunner status. Nevertheless it is totally clear GJ deserves far more attention than he got, and deserves far more votes than he will get -- the USA's present voting system plainly is an absurd parody of democracy. ============================================================== WHAT ABOUT THE PRECEDING (2012) USA PRESIDENTIAL ELECTION? This election was largely but not entirely independent of the USA 2016 race. "Not entirely" because Perry and Santorum ran in both races. "Largely" because all the major candidates in 2016 were a wholy-disjoint set from those in 2012. 2012 Candidate H0 H1 H2 H3 H4 H5 FC Apprv Mitt Romney(RN) 9 18 17 28 18 15 206 475 Rick Santorum(R6) 8 27 20 22 12 10 59 430 Ron Paul(R) 8 20 13 20 20 20 40 610 Newt Gingrich(R) 15 20 20 24 13 8 75 325 Herman Cain(R) 12 42 15 19 12 0 26 430 Tim Pawlenty(R) 6 18 12 18 35 12 17 420 Jon Huntsman Jr.(R) 6 6 28 28 22 11 18 490 Michele Bachmann(R) 26 36 13 10 7 8 61 420 Rick Perry(R6) 11 18 18 25 14 15 169 345 Barack Obama(DN6) 2 12 12 27 28 21 572 525 Notes: 6 inside the parentheses warns that candidate's politifact report card honesty stats may include many statements they made not just in the 2012 race, but also during the 2016 and/or 2008 races. Approval gives favorable/unfavorable percentages expressed as 1000F/(F+U) at latest date shown for that candidate in the multicolored approval-vs-time graph http://rangevoting.org/AppPrimaryGOP2012.png which I'd created about 4 years ago from a large number of approval-style polls (I put it toward the end of http://rangevoting.org/USA2012primary.html ). None of the 10 candidates listed managed to get True+MostlyTrue percentages 51% or greater according to Politifact; the highest three were Obama with 49%, Pawlenty 47% (but that was based on only 17 facts checked), and Ron Paul with 40%. Only two candidates got more approval than disapproval: Obama and Paul. Again, by an amazing coincidence, the two most-approved candidates coincided with the two most-truthful candidates -- at least, if the low-data candidate Pawlenty is ignored. Pawlenty dropped out very early in the race, on 14 August 2011, many months before the first vote in the first US state primary was cast and hence arguably should not even be counted as a "candidate" at all. All the others stayed in the race until at least two states had voted. This amazing coincidence corresponds to a confidence level of 88 to 98% depending on how I compute it. ========================================== THE FRENCH PRESIDENTIAL ELECTION OF 2012 OWNI "Veritometre" fact-checking data: http://owni.fr/2012/04/20/une-presidentielle-en-donnees/ "OWNI data journalists have checked nearly 1300 quantitative claims by the presidential candidates -- on employment, security, trade ... Of all the statements we checked, 40.2% are correct, 41.6% are incorrect and 18.6% turn out to be too vague." I am going to tabulate each candidate's correct/incorrect RATIO, ignoring claims that Veritometre called "en cours" (i.e. still unchecked; apparently they planned to check those later but I do not know if that ever happened) and for the present purpose I am going to count what Veritometre calls "declarations imprecises" (meaning numbers within +-10% but not within 5%) as "correct." Here are the 6 most-approved candidates in France 2012 election, who also were the six with over 100 claims checked by Veritometre as of 20 April 2012: Veritometre ratio Average CANDIDATE FC correct/incorrect Pinocchios Approval Fav/Unfav Jean-Luc Melenchon 164 2.82 1.83 (23) 39.07% 45/46=0.978 Francois Hollande 215 2.71 1.97 (35) 49.44% 61/32=1.906 winner Eva Joly 107 1.86 2.20 (20) 26.69% 30/61=0.492 Nicholas Sarkozy 392 1.11 2.35 (34) 40.47% 49/48=1.021 also in runoff Francois Bayrou 271 1.08 2.43 (23) 39.20% 48/42=1.143 Marine Le Pen 163 1.25 2.71 (21) 27.43% 33/62=0.532 KEY FC = #quantitative claims made by that candidate that they checked; correct/incorrect ratio as described above; Average Pinocchios: a different set of facts checked by Veritometre each were awarded "Pinocchio counts" (more Pinocchios==>more dishonest). For example "1.83 (23)" for Melenchon means of the 23 facts claimed by Melenchon he got an average of 1.83 Pinocchios. Approval: in a pseudo-election exit poll study of 2340 voters using approval-style voting. carried out by Antoinette Baujard, Frederic Gavril, Herrade Igersheim, Jean-Francois Laslier, and Isabelle Lebon and corrected by them for geographic biases. Fav/Unfav: IPSOS professional approval-voting-style France-wide poll 19 April 2012. (Apparently by telephone with about 956 respondents.) I have also computed their F/U ratio numerically. Hollande not only won the official election (both first round, and runoff), he also would have won with approval voting (either poll) and also apparently would have been the "Condorcet beats-all winner" according to pairwise polls. And, Hollande was the second-most-truthful candidate, slightly behind Melenchon, and was actually THE most truthful within the statistical margin of error in the truthfulness measurement [i.e. given that only 164, a finite number, of facts were checked for Melenchon, his truthfulness is only measurable to within error at least 1 part in squareroot(164)]. After France's first voting round, the runoff was to be between Hollande and (the incumbent seeking re-election) Sarkozy. Hence there was a Hollande-Sarkozy debate, and further fact-checking of statements by Hollande & Sarkozy. http://owni.fr/2012/04/29/credibilite-hollande-56-sarkozy-44/ "Data journalists from OWNI checked 132 statements by Nicolas Sarkozy or Francois Hollande made during recent debates or speeches." They gave a score of 56% correct to Hollande versus 44% for the incumbent Sarkozy. So our truthfulness<->approval connection arguably is supported by Hollande's victory in France 2012, in the sense that he was the second-most-honest candidate (and considering the statistical margins of error may actually have been the most honest). This support is far less clear if we go to the next most approved candidate -- namely either Melenchon, Bayrou, or Sarkozy. All three were comparably approved, within statistical margin of error, but their truthfulnesses differed considerably according to Veritometre. So certainly our whole hypothesized connection is less supported by France 2012 than by either USA 2012 or USA 2016, but nevertheless I would contend France 2012, on overall balance, supports the hypothesis. UNFORTUNATELY, Veritometre for our table's column about correct/incorrect ratio only checked numerical claims, and their +-10% cutoff, while objective, probably does not have greatly relate to what voters care about. For example a candidate who said "three million" when the correct number was 2.6 million would be rated "incorrect" by Veritometre -- but speaking as a voter, I probably would have been satisfied with that candidate's approximation. Meanwhile, such blatantly false (according to several fact-checking entities unanimously) claims by Donald Trump in the USA 2016 campaign, such as "I opposed the Iraq war from the beginning," "Hillary Clinton started the 'birther' movement, but I finished it" and "Hillary Clinton has been fighting ISIS her entire adult life" would not matter to this Veritometre column at all since those were *unquantitative* claims! But those claims probably would matter to voters! This problem might be overcomable by also obtaining data from Le Monde's "Les decodeurs" fact-checking entity. Unfortunately Le Monde only has made Decodeurs archives available online starting soon AFTER the 2012 election ended, so I haven't been able to investigate that. And Desintox did not provide tallies and hence also was not very useful. Fortunately an independent (?) collection of facts were checked by Veritometre using a "Pinocchio scale" over several months and I obtained our "average Pinocchio" tallies from Vincent Flores (http://frenchflairdata.blogspot.com/2012/08/french-presidential-candidates-fact.html) -- albeit unfortunately the number of facts checked for each candidate remains 3-10 times less than Politifact's corresponding fact-check-counts for the major USA 2016 candidates. ========================================== CONCLUSION If we COMBINE the USA 2012 coincidence with the USA year-2016 coincidence (we also optionally could expunge the Santorum and Perry data from the 2016 race to make 2012 and 2016 more-independent; this has little effect on our conclusions), then we deduce combined confidence levels of between 99.93 and 99.997% (depending on how calculated) that the following hypothesis is correct (versus the null hypothesis) HYPOTHESIS: High approval happens for, and only for, candidates with high truthfulness. If we also throw in Hollande in France 2012 that would bring us a further factor of about 3 nearer to 100% confidence, reaching between 99.98 and 99.999% confidence that highest approval is connected to highest truthfulness. (end.)