Say what you will about Andrew Yang, he seems to genuinely enjoy running for things. The former entrepreneur started his pursuit of public service the way many of us might … by running for president. Although he was mocked by some, he was earnest and had some big ideas — most notably Universal Basic Income — and he was not shy about talking about them. He was young and wealthy (although not as wealthy as some of the other candidates), and not afraid to do some rather unconventional campaign stops.
The result of his happy, enthusiastic campaign was a somewhat improbable fan club — the Yang Gang — that boosted Yang online. Personal friends that I never knew to be political would ask me if I had heard of Andrew Yang, and if I thought he could win. This attention created some rather unusual pricing of the odds for his presidential run, much to the consternation of Nate Silver.
Traders on PredictIt consistently bought up shares that he would win primaries in every state, defying the polls that suggested otherwise. Nearly 8 million shares of Yang to be the Democratic nominee for president would eventually change hands in that market. This phenomenon was well described by jipkin, a well-known trader in political markets, and provided some great profits for political bettors with deep pockets. But Twitter, as they say, is not real life, and Yang’s campaign for president ended before the New Hampshire primary.
When it seemed that Andrew Yang might follow up his presidential campaign with a run for Mayor of New York City, traders in information markets were initially skeptical. Yang was effectively a meme candidate — not a serious contender — and could be bet against for a tidy profit. But 2021 is the year of the meme stocks. Like Gamestop, shares of Yang to become the Democratic nominee for mayor of New York City rose, and rose, and rose. On Polymarket, his probability topped out at 76% in early May:
As of June 18th, shortly before the primary is set to occur, nearly $1 million has been traded on Polymarket on this question.
Unlike Gamestop, however, there appeared to actually be some support underlying Yang’s apparently unstoppable campaign.
Polls in early April showed Yang with 22% in a crowded field and a convincing 9-point lead over Eric Adams, his next closest rival. He also raised more individual donations from New Yorkers than his rivals (despite the fact that a greater percentage of his donations came from outside New York). Other polls from around that time showed him leading the pack in terms of favorable opinions. Yang was cruising to an easy victory.
Until he wasn’t. I think two primary factors account for the dramatic fall in Yang’s chances:
Name recognition bias inflated his early lead
Everyone loves talking about Andrew Yang. He ran for president! He wears a hat that says MATH! He has bad opinions about street vendors! And everyone loves to write about him:
He generates attention, and so he gets stories — he gets stories, so more New Yorkers are generally aware of him. More New Yorkers are generally aware of him, so more will tend to rank him in polls … especially before most people are really paying attention. A January poll suggested that 84% of respondents had heard of Yang, compared to only 60% having heard of Eric Adams. And of course, more people ranking him in polls generates more stories about how Andrew Yang is leading the polls as an unstoppable juggernaut. High name recognition becomes a self-fulfilling prophecy.
Name recognition also impacts the information markets. The structure of information markets on the Polymarket platform removes the trading caps that seem to have distorted the PredictIt markets for Yang and allowed the “Yang Gang” to effectively max out bearish traders. But these traders still retain some of the same biases. Information market traders skew young, male, and affluent — and in many cases, extremely online — coincidentally, a demographic that is likely even more exposed to a constant barrage of Yang content than most New Yorkers.
Ranked choice voting is very weird and hard to poll
New York will use a new ranked choice voting system in this mayoral primary — each voter marks each candidate they like in order of preference, giving a first choice candidate, a second choice candidate, and so on (voters can also mark just one, or up to five candidates as they like).
As the last-place candidate is eliminated, their votes are redistributed to the remaining candidates according to the ranked preferences of the voters. This system is great in ensuring that even if your first-choice candidate doesn’t win, your preferences are still accounted for in the final selection — you don’t need to hold your nose and vote for someone you hate just to block someone you hate more. But it is more complex than traditional “pick one” voting, and so New York has gone to great lengths to try to explain it to voters.
But if you thought it was difficult to poll a presidential election with just two candidates, asking people to rank their top five choices among a field of thirteen is even tougher. The earliest polls showing Yang leading generally had high numbers of undecided voters, and did not attempt to simulate a ranked choice election — simply asking voters for their top choice. Where newer polls attempt to simulate RCV, the margins are slim — a recent poll has Yang eliminated in 4th place, giving his votes to Kathryn Garcia who then edges out Maya Wiley by less than a percentage point for second place and loses to Eric Adams by 4 points after all votes are distributed.
The impact, then, is that any slight bias toward one candidate, like Yang, can have an outsized impact in the final results by changing the redistribution of ranked choices votes in the simulation. And, in early polls featuring more undecided voters, fewer respondents included choices beyond perhaps first and second — leaving less votes to be redistributed and again potentially skewing RCV simulations. New York mayoral primaries have tended to have lower turnout in past years — Yang himself has never voted for mayor — which further challenges pollsters attempting to size the electorate in projecting their results. These complexities make this election one of the most difficult to forecast that I can recall in recent memory. And it won’t even be easy to determine a winner once the votes have actually been cast: it looks like it may take weeks for final results to be available.
In the end, information market platforms such as Polymarket are a source of information about the world — but their prices are a derivative of what information is available. In the case of Andrew Yang, while markets may have initially overstated the odds that his mayoral campaign had a commanding lead they subsequently quickly incorporated new information — from later polling, to more information about how RCV will play out, to Yang’s performances in the debate. Each data point gets incorporated into the price, theoretically making the forecast increasingly accurate as the election approaches. Unfortunately for Yang, however, the odds don’t look good.
Full disclosure: Polymarket has compensated me for this post.
