Thursday, July 30, 2015

Wanted: A Condorcet Experiment for the GOP

The headlines scream 'Donald Trump is the Front Runner!'

It is the case that Trump has polled at the front of the pack for the GOP nomination for a couple of weeks. His support level reaches the atmospheric heights of . . .  24 percent. And those numbers have settled back to 18% as of the end of July.

The trend of support for the 16 current GOP candidates (plus undecideds) appear in the graphics at right. Much like in 2012, when a variety of new flavors surged to the top of the polls only to fade back into obscurity, there is tremendous oscillation in support.

[A related take on this argument can be found over at the 538 Blog]

To call any of these candidates a 'frontrunner' becomes comical when you show their support on a 100-point scale.  Everyone is muddling around near the ground.  To the extent that Donald Trump is a frontrunner, it a consequence of unprecedented fragmentation of the GOP electorate across a tremendous number of candidates. 

[See also Natalie Jackson at HuffPo who first called into question the frontrunner status of #TheDonald based on poll data]

The truth is, we just don't know who leads this pack. Sam Wang argued correctly over at the New Republic that "Single-choice, horserace polls provide the wrong kind of data, and it’s misleading much of the media and the public—not to mention Trump himself." One pollster attempted to get at this by asking a question favored by myself and other advocates of runoffs, instant runoffs, and other majoritarian systems, that of 'second preferences.' A recent Monmouth poll revealed that Trump had 24% support and an additional 8% of voters indicated he was their second preference, for a potential 32% support level.

What we need is a Condorcet experiment. The Condorcet candidate in an election is the candidate who defeats all other candidates in a head-to-head competition. For Trump to be a definitive frontrunner, he should be able to defeat any other candidate head-to-head, absent other choices. In other words, he would be the majority preference among those people who do not name either him or any one other opponent as their first choice.

One way to tease out Trump's Condorcet potential is to ask voters to consider every potential trial heat. Now, unfortunately for testing the current GOP primary field, there are 120 possible head-to-head matchups, so ascertaining preferences might prove tiresome. But, when support levels are this fragmented (the Rae Fracturization Index for the GOP primary is .10, which indicates extreme fragmentation), it is possible that a candidate who is scoring in single-digits might be the Condorcet candidate, capable of beating anyone else head to head.

We need to start testing the pairs to figure out the front-runner.

Another alternative is to ask voters to rank-order their preferences -- not unlike the AP Football Poll. This is an application of the Borda Count, which ranks candidates (teams) and assigns points based on the inverse of their ranking.

This method is within the reach of voters (I think we can assume we voters are  nearly as capable of ranking politicians as we are of ranking football teams.)  This approach allows us to disentangle the ranked preferences and, therefore, the presumed preferences among voters  in any pairwise voting contest.  And, it would provide us with meaningful information about which candidates are taken most seriously by the electorate, and which are only deemed viable by a fringe or fraction of the voters.