Rob Richie gets it wrong about Louisiana Governor Race 2015

Warren D. Smith, 29 Nov. 2015.

This concerns the article Did Louisiana Just Elect 'Wrong' Governor Due to Flaw in Top 2 Runoff Method?, by Rob Richie (head of the Instant Runoff Voting propaganda organization "FairVote"), Huffpost 24 Nov 2015.

We point out that a better title would have been "Rob Richie wrong about wrong governor."

Rob Richie's claims:

1. On 23 November 2015, Democrat John Bel Edwards won the 2-man runoff against Republican U.S Senator David Vitter by 56.1% to 43.9%.

2. RR says it is "quite likely that either of the other two Republicans in that first round in October would have defeated Edwards." This first round, held 24 Oct. 2015, was:

  J. Bel Edwards  (Dem)  39.9%
  David Vitter    (Rep)  23.0%
  Scott Angelle   (Rep)  19.3%
  Jay Dardenne    (Rep)  15.0%
  Cary Deaton     (Dem)   1.1%
  S.L. Simpson    (Dem)   0.7%
  Beryl Billiot   (Ind)   0.5%
  Jeremy W. Odom  (Ind)   0.4% 
  Eric P. Orgeron (Ind)   0.2%

"Vitter was getting the most votes among Republicans, but... was also the weakest Republican candidate when paired against Edwards and also lost when matched one-on-one against either of his fellow Republicans." That claim was based on Clarus Research Group (pollster) head-to-head race polls finding

  Vitter (R) 41  vs  Edwards  (D) 45
  Vitter (R) 35  vs  Angelle  (R) 40
  Vitter (R) 35  vs  Dardenne (R) 42

reported in this 27 sept 2015 news story http://theadvocate.com/news/13542209-123/poll-among-david-vitter-john (800 likely voters, claimed "margin of error" ±3.46%), and on these pairwise results from a PPP poll released 24 Sept:

   Vitter   (R) 38  vs  Edwards (D) 50
   Angelle  (R) 40  vs  Edwards (D) 40
   Dardenne (R) 42  vs  Edwards (D) 40
   LouisianaPollSeptember2015.pdf

3. Richie does not mention this, but that same PPP poll (616 likely voters, 21-22 sept 2015, claimed "margin of error" ±4%) also did an approval voting subpoll. And the same Clarus poll cited by Richie went further by reporting score voting ratings on a "very favorable" (VF), "somewhat favorable" (SF), "somewhat unfavorable" (SU), "very unfavorable" (VU) 4-level scale.

                PPP        CLARUS    Mean Clarus
   CANDIDATE  fav/unfv  VF,SF,SU,VU    score
   Edwards     35/27    11,28,10,5%    1.833
   Dardenne    33/32    10,33,10,5     1.828
   Angelle     29/29    10,29,10,6     1.782
   Vitter      34/51    13,32,16,28    1.337

PPP found Edwards would have been the approval voting winner, while Clarus found Edwards would have been the score voting winner. These findings, however, were damaging to the theory Richie was trying to fabricate (even though the PPP approval voting results actually were more convincing than the data from the same poll that Richie did use) so as usual whenever that happens, Richie ignored them.

4. Richie then figured that if IRV (instant runoff voting) with full rank ordering ballots had been used, Angelle would have gone ahead of Vitter thanks to vote transfers from Dardenne, thereupon likely winning against Edwards. The first part of that theory (Angelle beating Vitter) indeed is supported by Richie's poll data. The second part (Angelle beating Edwards) however, is a 50-50 coin toss based on the poll data Richie was able to find. Therefore for that part Richie ignored his own data and simply based it on "Republicans now typically [win] in statewide offices in Louisiana."

Richie then concludes "John Bel Edwards should be congratulated for winning his election under the 'Top Two' runoff rules used in Louisiana. Second, Edwards almost certainly owes his victory to a flaw in those rules..." He then urges that Louisiana switch from top-2 to IRV.

Analysis:

5. However, Richie forgets to mention that actually the very polls he himself cited showed Edwards was the correct winner with approval voting, and with score voting, both of which are systems with lower Bayesian Regret than IRV.

Oops.

6. Also, let us ask ourselves the following. Suppose a professional statistician had the very same poll data Richie examined, showing Edwards was the approval-voting winner, and score-voting winner, and showing pairwise results

Vitter 38 vs Edwards 50;    Angelle 40 vs Edwards 40;    Dardenne 42 vs Edwards 40

(with claimed margin of error ±4%). Would any professional statistician on the planet have concluded from this that it was "almost certain" that Edwards would have been defeated? Then still kept his job?

And Richie also concluded it was "quite likely that either of the other two Republicans in that first round in October would have defeated Edwards." (My italics.) Again: would any professional statistician on the planet have concluded that and still kept his job? The answer, I hope, is "no." In fact, the chance "either Republican would have defeated Edwards" is actually about 30% based on Richie's own data, which is not "quite likely," and the chance Richie's whole theory that Angelle would with IRV have defeated Edwards, is not "almost certain," it is "less than 50%." These are mathematical facts, based on Richie's own poll data.

7. And there actually is more poll data, and more recent poll data, available than the data Richie cited. What does it say?

Here's a score voting statewide telephone poll by Market Research Insight 15-19 Oct 2015 using scores "very favorable" (VF), "somewhat favorable" (SF), "somewhat unfavorable" (SU), "very unfavorable" (VU) and "don't know" (x). I will calculate mean score based on a (3,2,1,0) numeric scale for (VF,SF,SU,VU):

CandidateScores (VF,SF,SU,VU,x)Mean Score
J. Bel Edwards(128,137,60,67,208)1.83
Scott Angelle(61,196,51,58,234)1.71
David Vitter(74,144,83,177,122)1.24

An independent score voting statewide telephone poll was done by Harper Polling (612 likely voters, Oct. 16-17, percentages given unlike above poll where counts given):

CandidateScores (VF,SF,SU,VU,x)Mean Score
J. Bel Edwards(35,26,15,14,11)1.89
Scott Angelle(16,28,23,14,19)1.57
Jay Dardenne(18,27,25,15,15)1.56
David Vitter(21,15,18,38,8)1.21

Keep in mind that score voting is the best voting system discussed here, with lowest Bayesian Regret, and that the election was held 24 Oct.

Revised conclusion

Edwards fully deserved his victory under the "top two runoff" method, as is shown by two score voting polls both more recent, and more convincing, than the polls Richie misused; and also by the same polls Richie used (but Richie ignored these inconvenient data within his own polls). Although top two runoff indeed has its "flaws," they were not in this election responsible for a "wrong governor." Richie employed standard techniques from the book "How to lie with statistics" by Darrell Huff, such as cherrypicking his data, ignoring even the data he did cite whenever it disagreed with the conclusion he wanted to reach, and using utterly unjustified phrases like "almost certain." However, in addition to being an incompetent and fraudulent statistician, Richie also wasn't even a very good liar, because even the most crude and shallow attempt to check his facts was enough to refute him – had HuffPost bothered to do even a crude fact check (which obviously, they didn't). In particular it took me less than a single day to write this (the same day I first saw Richie's article).


Prior Louisiana governor elections

Return to main page