Just as ranked choice voting offers a better way to run multi-candidate elections, it promises a boon to public opinion research. Here’s a demonstration of how data collected by a recent online ranked choice survey could have been used to anticipate Carly Fiorina’s and Marco Rubio’s recent surges in the GOP race.
The source data comes by way of the voting reform group FairVote, and Civinomics, a civic software startup based in Santa Cruz California which launched a new ranking site in August. Well over 3,000 ranked votes were submitted for the GOP poll by mid-September. Not surprisingly, the results offer fresh insights about that crowded race.
I’ve been monitoring activity there for a while. At first Rand Paul was in front. Then a huge wave of Donald Trump supporters showed up. Carly Fiorina briefly took the lead. Last time I checked, Ben Carson was far far ahead. Each surge of votes for a given candidate was probably triggered by an email sent out by his or her campaign, or perhaps because of a comment that appeared somewhere on a partisan social media feed. That kind of swarming behavior is fairly typical of online polls, though you usually don’t see so many distinct groups pop out.
With no control for demographic balance and no guard against participant bias, the raw Civinomics results can’t be considered representative of the voting public. Technically, it’s a “nonprobability poll.” Results aren’t deemed scientifically valid because of the way participants select themselves. The data collected don’t meet the highest standards for random sampling.
But accuracy isn’t the point. That’s not what Civinomics and FairVote are trying to achieve. Just like the Oscar Best Picture sample ballots that FairVote has promoted over the past few years, the goal is to get more people to see the benefits of ranked choice voting.
By the same token, a closer look at these ballots can enlighten people about the benefits of ranked choice polling. At a minimum, it can help advance development of a more rigorous and dependable methodology for it.
* * *
There were 3,153 ballots in the data set I downloaded from Civinomics on Saturday, September 5. Of those, nearly 800 participants had ranked just a single candidate. About half the participants ranked at least 6 candidates while 891 ranked all 17. Most of my analysis focused on that last subset. I’m interested in how ranked choice voting can serve “team players” who make the effort to evaluate the full list of candidates presented to them.
Carson supporters vastly dominated the original data, with Trump in second place, followed by Paul and Kasich. The same was true in the subset of voters who filled in all 17 ranks. I ran a transformation on that subset which was designed to match proportions with on-the-ground, real-world polling, where Trump dominates. I used a combination of polls available at Huffpost pollster.
This is where things get very technical, Suffice it to say that, rather than base the transformation on Huffpost’s statistical aggregate of polls, I used sources that had the lowest numbers of undecided voters. This allowed for relatively less fudging in comparison with a snapshot that would in any event become obsolete fast.
To match the real-world percentages, I calculated the multiples that would have to be applied to the sets of ballots cast, grouped by a voter’s first choice. That approach preserves the down-ballot sentiment for all the ranks submitted within each group. The calibration process generated a dataset with over 45,500 ballots. As it turned out, to get there, Carson’s multiple was 4. Trump’s was 42. The largest multiples by far went to Santorum and Huckabee (130 and 163), indicating that their supporters were most underrepresented in the Civinomics poll.
(Shortly after I drafted this post, I noticed that the Civinomics folks had also published their own reweighting based on standings in one of the prominent national polls.)
Given all these scrubbings and transformations, plus the normal caveats that go along with public opinion research, it would be foolish to pretend that the results presented here reflect today’s electorate. But the down ballot numbers do seem to align with traditional polling, giving reasonable confidence that this approach is sound.
For example, the results confirm Donald Trump’s controversial role in the contest. He is simultaneously the most popular and the most unpopular choice. As with “real world” polling among likely Republican voters, Marco Rubio and Ben Carson have notably low unfavorability scores while Jeb Bush and Lindsey Graham have relatively high ones.
Things get very interesting when we examine the data to investigate which candidates are best positioned as consensus choices. Here’s a paragraph I drafted shortly before the second GOP debate.
These results also underscore the extent to which Carly Fiorina and Marco Rubio are commanding considerable depth of support in this highly fragmented field. Those strong showings suggest they might perform very well in the Condorcet and Instant Runoff simulations that I plan to publish before long. That can have important implications as the Republican field begins to shrink.
This depth of support is indicated in the chart above, where their 6th place ranks (colored orange) reach considerably farther across the scale than downballot support for the former front runner Jeb Bush. That suggests they are in much better position to pick up support as the field shakes out at the bottom.
As it turned out, both Fiorina and Rubio surged after the debate. Their debate performances served them well, of course. But the analysis illustrates the extent to which they were positioned to break out from the pack. What’s being revealed here clearly strengthens arguments for the virtues of ranked choice polling.
* * *
It’s no secret that there are lots of people in the GOP who want to focus on defeating Donald Trump. They presumably recognize the virtues of coalescing around a strong opponent. If so, they might also recognize the utility of a polling tool that can help them locate that champion.
By the same token, Donald Trump has a strong interest in picking up new supporters as opponents drop out, and in getting a heads up about which opponent is most strongly positioned to coalesce his most vociferous opposition. This tool could help him, as well.
More importantly, for the long run, a tool like this can be used to help members of the electorate rank their own priorities and then evaluate candidates’ responses to questions on those priorities. Consider how useful mobile devices that support real-time interactive ranked choice feedback would be for straw poll and town hall participants. That’s the kind of tool worth building for a Democracy that wants to be worthy of the name. That’s what AimsPoll is ultimately designed to do.