Unpopular Front

Unpopular Front

False Purposes; Statistical Garbage; To Math or not to Math

Reading, Watching 11.22.26

John Ganz
Nov 22, 2026
∙ Paid

This is a regular feature for paid subscribers wherein I share some of what I’ve been reading and watching.

If you regularly read, enjoy, or share Unpopular Front, please consider becoming a paid subscriber. This newsletter is entirely reader-supported and is my primary source of income. Your $7 a month makes it possible for me to keep writing independently.

My book, When the Clock Broke, is out in paperback in the US and UK, with copies turning up as far afield as Kuala Lumpur and Christchurch.

I also co-host the film podcast Unclear and Present Danger with Jamelle Bouie of The New York Times. Our Patreon features bonus episodes and a weekly politics discussion.


Joseph Wright of Derby, An Experiment on a Bird in the Air Pump (1768). Oil on canvas. The National Gallery, London.

The old saying goes, “Victory has a thousand fathers, but defeat is an orphan.” Everyone is still trying to establish the true paternity of the Democratic wave: was it the progressives or the moderates? I kinda don’t care who the real daddy is, partly because we didn’t have a single election, but hundreds of them, each of them in different localities with their own particular qualities. Show me the specific election, and then we can discuss what happened. There are some common national conditions that are not difficult to divine: unpopular president, bad policies, high prices, etc., etc., etc. All of these are interrelated facts that are interesting to arrange in the correct order, and there are real political stakes in doing so. But there is this desire to assume there is such a thing as “election science,” which drives me crazy because it’s mostly bullshit and highly tendentious without admitting as much.

Jonathan Chait recently had the gall to compare “denial” of “election science” to climate change denial. This is absolutely absurd and intellectually indefensible. Climate change is based on classical physics, thermodynamics, and chemistry, that is to say, experimentally solid theories in the hard sciences backed up by years—centuries in some cases—of experimental validation that establishes invariable laws across time and space. Nothing—nothing—in the social sciences approaches this level of verification, and “electoral science” certainly has no remotely comparable theoretical foundation. Its models all depend on historically contingent relationships among human beings, institutions, and living, changing political circumstances.

When I pointed this out, statistics guru Nate Silver chimed in and insisted that his models were “falsifiable.” But this is exactly what they are not. In Karl Popper’s famous and extremely demanding criterion of good science, a scientific theory must risk being wrong: It must state conditions by which it would be refuted. If Einstein were wrong, then light would not bend around the Sun according to his equations. If a theory can encompass all possible outcomes, it is not falsifiable. These election models deal with probabilities. If you make a good prediction, you say, "Yep, our model is right," if not, you can say, "Well, we did say it had a 20 percent or so chance." That's the very definition of being unfalsifiable. What would falsify the entire model? And don’t ask me; ask the guy who invented the very idea. Here’s Sir Karl himself in The Logic of Scientific Discovery:

…if we use probability estimates in this way for the ‘explanation’ of observable regularities without introducing special precautions, then we may immediately become involved in speculations which in accordance with general usage can well be described as typical of Speculative metaphysics.

For since probability statements are not falsifiable, it must always be possible in this way to ‘explain’, by probability estimates, any regularity we please.

To be fair, Popper’s falsificationism is widely understood as too demanding a stricture, and it doesn’t capture how science is actually practiced, but I didn’t bring it up—they did. Moreover, it would be crazy to insist that probabilities and statistics have no scientific validity. The question is what they actually establish, what they leave unexplained, and what additional assumptions are required to turn mere statistical regularities into robust theories.

But Popper’s warning about “speculative metaphysics” gets closer to my problem here. No one seriously denies that electoral forecasting works, or that it can be done with increasing statistical refinement. Fundamentally, though, much of it involves asking people how they intend to vote and then making adjustments to account for various sources of error. It’s not exactly groundbreaking that asking people what they intend to do might yield some knowledge of what they will do. And it’s obviously a legitimate and useful exercise, but it doesn’t tell us how the electorate was constituted, why people hold the political allegiances and opinions they do, or how those allegiances and opinions might change. Yet from the ability to predict the behavior of an existing electorate, an entire theory of politics is being generated! What voters really want, what kinds of candidates are possible, and what political strategies are rational and so on and so forth. This is a metaphysical idea of politics masquerading as empirical science. Indeed, I would say it’s a particularly insidious kind of metaphysics, because its positivist pretensions prevent its practitioners from recognizing that they are engaged in metaphysical speculation at all. Nothing against speculation per se, but let’s be honest about what we are doing.

What I think makes this positivist picture of politics so poor is that it adds very little to a pre-theoretical, commonsense understanding of political rhetoric. It’s obvious that you have to understand your audience, appeal to its interests and passions, adapt your message to circumstances, and avoid gratuitously alienating people you hope to reach. Aristotle wrote about all this more than two thousand years ago. The supposedly “scientific” discovery that candidates should adopt positions popular with the voters they need to win is hardly some revolution in our understanding of politics. The statistical methods may refine our knowledge of what particular audiences believe, but they don’t establish any new theory of political action. You could say that they take elementary principles of rhetoric, translate them into the language of quantitative optimization, and then present the result as a scientific discovery.

In fact, their theoretical apparatus is less sophisticated than classical rhetoric because it doesn’t even have a good theory of change. The point of political practice isn’t just to discover what people want to hear: It’s to change people’s minds, reshape their understanding of their interests, and even to call into existence new publics. Political rhetoric can bind scattered grievances into a common cause and people who had never thought of themselves as belonging together into a single political movement. The electorate is not some fixed object out there awaiting measurement; It doesn’t really exist until election day, and the campaign is shaping its potential. The old electorate didn’t exist without Trump’s campaign, and this new one doesn’t exist without the failure of his government. Electorates themselves are a product of unfolding historical change and political activity. Now that in no way says anything definitive about being “progressive” or "moderate,” or says that there are no limits to what a politician can accomplish with the right message. But it says that the conditions are considerably more involuted than the positivists think they are.


If you’re not convinced by my own unregenerate humanism, then take a look at Matt Bruenig’s essay, “How Do You Know Which Candidates Performed Well?”, which examines the statistical methods used to determine whether candidates “overperformed” or “underperformed” expectations. Bruenig shows just how much depends on the construction of the hypothetical candidate against whom actual politicians are measured. Different methods produce different answers, and even the most elaborate models can’t settle the dispute over moderates and progressives. So, even on its own terms, the supposed science is inconclusive. This is just another way of saying this is all a lot more complicated than these people want to believe. Or, as I learned to say in my undergraduate Quantitative Methods in Social Science course, “There are too many confounding variables.”

User's avatar

Continue reading this post for free, courtesy of John Ganz.

Or purchase a paid subscription.
© 2026 John Ganz · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for writing, podcasting, and media founders.