What Nate Silver's Haters Tell Us About Climate Risk
Progressives’ Constructed Ignorance on Polling and Climate
By Alex Trembath
It’s election season, which means a return to the quadrennial tradition of yelling at Nate Silver on the Internet. While I think most people at this point have made their peace with his probabilistic forecasts, many progressives, demonstrably confused by basic statistics, regularly accuse Silver of deliberately underestimating Democrats’ electoral fortunes. And it occurred to me that their confusion mirrors a similar mistake progressives make in evaluating climate impacts. In each case, many observers fail to understand the ways in which relatively modest changes in statistical averages are associated with larger relative shifts in the tails of the probability distribution. The funny thing is, the misunderstanding runs in opposite directions in the case of climate versus the case of election forecasting.
Start with the politics. Silver has been publishing his election forecasting model for the better part of twenty years, over five different presidential races and thousands of Congressional races. Progressives with less training in statistics have had that entire time to learn the basics of probabilistic forecasting. And many of them have steadfastly refused to do so.
Here’s Markos Moultisas, founder of The Daily Kos, reacting last week to the “wild swings” in Silver’s active election forecast:
Moultisas’s expectation that Silver’s model should yield a steadier forecast is based on his lack of awareness that a small nominal change in averages can result in large nominal changes in expected outcomes. This is really basic stuff and is easily illustrated. If, hypothetically, Harris were polling at 55% to Trump’s 45%, she would have overwhelming odds of winning the election. If, on the other hand, Harris were polling at 51% to Trump’s 49%, her odds of winning would be substantially lower and plausibly underwater given Electoral College effects, even though the electorate is broadly similar in both scenarios. Shifts in polling within these margins would be completely expected to yield “wild swings” in expected outcomes.
Progressives got really mad at Silver after he failed to predict a Trump victory in 2016, even though his model gave Trump a higher chance of winning than the other forecasters (including the Daily Kos). The whiplash of the Trump victory perhaps excused their frustrations at the time, but now they’ve had eight years to grapple with the basic statistics behind election forecasting and they’ve chosen instead to bury their heads in the sand. Any prognostication that does not herald an overwhelming progressive groundswell is antithetical to many progressives’ fantasies about the state of public opinion, so they ascribe bad faith and even corruption on anyone, like Silver, who simply evaluates polling data.
And we see a similar, and arguably much more common, willful misunderstanding of the statistical relationship between climate change and extreme weather.
The rising concentration of carbon in the atmosphere adds more moisture and energy to the climate system, influencing weather and, definitionally, weather extremes. This is simultaneously true and gets communicated poorly all the time. For instance, in 2022, World Weather Attribution (WWA) wrote that “Climate Change made devastating early heat in India and Pakistan 30 times more likely.” That is a large probabilistic change. But the exact same analysis found that “The same event would have been about 1°C cooler in a preindustrial climate.” As with election polling, a small change in the system is associated with a large change in probabilities. WWA is not saying that climate change caused the heat wave. They’re saying climate change made a heat wave that is one degree warmer than their counterfactual much more likely.
My colleague Patrick Brown has exhaustively covered this deliberate obfuscation of extreme weather trends, which is rampant in climate advocacy and journalism. As Ted Nordhaus wrote in the New Atlantis earlier this year, these obfuscations “have given journalists license to ignore the enormous body of research and evidence on the long-term drivers of natural disasters and the impact that climate change has had on them.”
I can think of a number of experts who are schooled in statistics, and who would defend Silver’s work, yet still fall prey to this kind of sloppy thinking on climate and extremes. But intellectual consistency requires probabilistic approaches to both. In elections and the climate, a nominally small change in the system can result in much larger shifts in the probability of outcomes.
Now with the election, the outcome difference is substantial – a Trump Administration would be quite different from a Harris Administration. With climate change, the outcome difference is more subtle: a heat wave of 100 degrees is not radically hotter than a heat wave of 99 degrees. Differences in extreme event intensity matter of course, since climate impacts occur on the margin. So it’s noticeable that progressives tend to emphasize the more abstract shifts in probabilities, since these yield a larger number than the shifts in the climate system itself. Most conspicuously, it’s precisely the opposite mistake that occurs in election forecasting criticism. With the election, progressives have a tendency to ignore the large probabilistic swings associated with small shifts in averages. With climate change, they instead tend to emphasize the statistical swings, and ignore the modesty of the shifts in the actual climate.
The unpredictability of the election and the relative stability of long-term extreme weather trends are what Steve Rayner called “uncomfortable knowledge” for some progressives. It’s uncomfortable for progressives to acknowledge that their political project is not massively popular, so they reject basic statistics forecasting uncertain electoral outcomes. And it’s uncomfortable for them to reconcile their climate catastrophism with IPCC science showing little observable trend in extreme weather, so they ignore the science. Rayner called this the “social construction of ignorance,” and the way you know it’s socially constructed is that progressives will wield the counterintuitive statistical relationships when it suits them (as with climate change) and pretend they’re voodoo when they don’t (as with election forecasting).
Braver men than I have tried to get people to grapple with this parallel between constructed climate ignorance and constructed polling ignorance. But for many, the knowledge is just too uncomfortable.
Journalist always seem to take liberties with relative and real risk data. In health a 50 % reduction in the chance of dying of a heart attack, could be 2 people out of 1,000 die of a heart attack without a drug, instead of 1 out of 1,000 if they take the drug. But the patients only get the first data point. And like climate change mitigation, the cure becomes worse than the potential future disease.
Interesting post. In regards to your example from World Weather Attribution - is a 1ºC change to a heat wave insignificant from the point of view of excess mortality and stress on the heating grid? And it feels sloppy to conflate this 1ºC change with a 1ºF change as you do three paragraphs later - "a heat wave of 100 degrees is not radically hotter than a heat wave of 99 degrees" - especially when chiding others for poor quantitative thinking!