Now one common objection, often made by Trump supporters, is that the overwhelmingly pro-Democrat class of hacks and data analysts is biased and ignorant of sentiment in the “Real America” and this colors their judgment, as it purportedly did in 2016*.
Conversely, their liberal counterparts rejoinder that popular prediction markets, which are strongly out of sync with polls-based models, are inundated by delusional volumes of “MAGA money”.
- Augur (crypto prediction market) – 42%
- PredictIt (prediction market) – 40%
- Election Betting Odds (aggregator of odds from betting sites and PredictIt) – 36%
- Hypermind (superforecasters) – 27%
- Metaculus (prediction market more geared towards rationalists) – 20%
- Good Judgment Project (superforecasters) – 16%
- 538 model (most prominent polls-based model) – 10%
So let’s try to do something original and look at the opinions of a group of pundits who, in net terms, don’t really have much of a dog in this fight: The broad Russian equivalents of Americans who read Nate Silver.
Perhaps the best place to look for that is amongst the commenters at Alexander Kireev’s blog, a Russian-American elections analyst who also maintains a large database of historical elections data and electoral maps from around the world. Although Kireev himself has (neo)liberal political values, including a strong distaste for Putin and a preference for centrist Democrats over populists from both the Left and Right, there is scant evidence that he lets partisan sympathies get in the way of his analysis. He is most often amongst the best or even the overall best in the election prediction competitions that he runs before many elections. (Indeed, my own contest was partly inspired by his).
In any case, we are going to be looking at aggregates here, so his audience are of more relevance anyway. Relative to their American elections-watching counterparts, they are not particularly biased against Trump: Some 47% are pro-Biden, while 35% are pro-Trump. This is actually more or less reflective of the American electorate as a whole, which according to 538’s summary as of the time of writing is 53% pro-Biden and 44% pro-Trump (even if Kireev’s crew are much less pro-Trump than the Russian normie, who to the extent he has an opinion favors Trump over Biden by a factor of 2:1 according to polls from Levada and IPSOS). Certainly they are much more “Trumpist” than, say, Nate Silver’s employees, whom I would guess are at least 90% pro-Biden.
However, as is already evident from the table above, they are nonetheless much more skeptical about Trump’s chances, with only 26% expecting him to win vs. 56% for Biden when that poll run on a couple of weeks ago.
It is worth noting that Kireev’s commentariat has a very good predictive record on elections, even if – like almost everyone – they underestimated Trump in 2016. Still, even that underestimation was very modest relative to American norms. In the 2016 predictions of the percentage vote tally**, their average prediction was 47.9% for HRC and 45.1% for Trump, or a difference of 2.8%. (I excluded one troll response giving HRC 0% and Trump 88%, which if included diminishes that gap to a mere 1.9%). For comparison, the 538 forecast back in 2016 projected 48.5% for HRC and 44.9% for Trump, or a difference of 3.6%. In the event, HRC got 48.2% and Trump got 46.1%, with the resulting difference of 2.1% being sufficient to roll out a win in the Electoral College. This means that Kireev’s commenters were substantially closer to the mark during the 2016 elections than 538.
So what are Kireev’s commenters saying today? They are still more positive about Trump than 538, but the good news for MAGA stops there. As of the time of writing, the 538 forecast projects 53.3% of the raw vote going to Biden and 45.5% going to Trump. That is a difference of 7.8% points, hence explaining Trump’s chances plummeting from about a third in 2016 to just a tenth now. Meanwhile, Kireev’s commenters – also, obviously, as of the time of writing – are projecting an average of 51.3% for Biden and 45.8% for Trump, or a gap of 5.5%. That is obviously much better than with 538, but still, the difference is twice as big as in 2016. Even a popular vote gap of 5.5% in favor of Biden translates into just a ~3% of Trump winning in the EC (see right). Even if they are wrong by 1% point – i.e., even more so than in 2016 – that only increases Trump’s chances to 30%.
Finally, while I am not going to waste time bothering to aggregate the state-specific predictions from Kireev’s competition, I will note that virtually all of the more notable individuals in that community are projecting a win for Biden.
(1) Kireev, who probably has the best predictive record of anybody else on his own site,
gives Biden a narrow win with 273, with Biden losing PA but winning FL and AZ EDIT NOV 3: apologies, I mistakenly thought the example map he included in his contest post was his prediction; it isn’t, in his actual prediction which he posted just now, Kireev gives Biden 350 including NC, AZ, FL, and even Georgia. He also projects 53.01% for Biden and 44.92% for Trump in absolute percentages.
(5) The commenter putiny80percent, who has been very good at nailing Russian elections, gives Biden 334.
(8) alacsony was the champion (out of 97 people) of the raw scores predictions back in 2016, giving HRC 48.6% (real: 48.2%) and Trump 46.4% (real: 46.1%). He also got Trump winning the Presidency correct as well (though he only had Trump winning by a marginal 270 versus his actual result of 304). Still, very impressive, especially considering that Trump clinched three big states only by a very narrow (<1%) margin. But this time round, he is projecting 280 for Biden (see map right).
EDIT Nov 3, 6am Moscow:
Seva Bashirov (see above) has done the work of compiling a “consensus map” of predictions from the first 42 participants in the states-level competition. Here it is:
With a few hours to go until Kireev’s competition closes, raw percentage projections are 51.34% for Biden and 45.69% for Trump.
* This was certainly true of some people and institutions, e.g. Huffington Post’s infamous 98% chance that HRC would win. On the other hand, 538 assigning Trump a 29% chance in 2016 was actually very reasonable even in hindsight, considering the <1% margins by which he won Michigan, Wisconsin, and Pennsylvania.
** Although Kireev inconveniently lists predictions in the form of a table in a jpg file, happily there is now free-source OCR software that can convert it into Excel files.