Twelve weeks in, the mechanics are on the table, and the column now turns around and points at itself. Movement III is for the objections, and I want to start with the one that would end the project if I could not answer it. An equation that accommodates every outcome is not an equation. If every news cycle can be scored after the fact, and every score can be explained after that, then the FTSeffect is astrology with better typography.
Karl Popper set the standard I am going to hold myself to here, and it is narrower than people usually remember. Confirmation is cheap; almost any theory can be confirmed if you go looking. His test was whether a theory forbids anything: whether there is some possible observation that, if it turned up, would be incompatible with the theory. A framework that fits every conceivable event is empty rather than powerful, and its universal applicability is the symptom.
Anything can be applied. The test that matters is what a rubric forbids.
So let me answer it. The FTSeffect forbids two things. It forbids saturation events with comparable numerators and very different denominators landing alike. And it forbids the defending side’s speed and preparation being irrelevant to outcomes. Both are testable in principle, and if either failed the framework is finished in its current form. Last week’s post would be the first casualty.
Now the honest inventory.
The parts of this model that are empirical are, almost entirely, other people’s findings. That false stories travel farther and faster than true ones, with novelty and affect doing much of the work, is the Vosoughi result, and I have already spent a post on what it does and does not license. That corrections fail to fully clear the original claim is the continued influence literature, replicated many times over. That paid campaigns can be documented when someone gets inside the operation is what the Kenya files showed. Each can be taken away from me by better evidence, which is exactly the property it needs. I borrowed them; I cannot revoke them.
The part that is not empirical is the arithmetic, and I would rather say so than have a reviewer say it for me. Five terms multiplied, divided by two, each scored one to ten, each carrying a weight. Every one of those choices is mine. None of them is handed up by the data. The OECD and the European Commission’s Joint Research Centre published the standard handbook on building composite indicators, and its central methodological warning applies to me without modification: in any composite index, normalization, weighting, and aggregation are decisions made by the analyst, and different defensible choices produce different orderings of the same underlying cases. The FTSeffect is a composite index. It inherits that problem whole. There is no version of this where the number that comes out the other end is a measurement.
There is a sharper objection still, and it is the one I would lead with if I were reviewing this book. The question is whether the five numerator variables are actually separable. In the field they move together. A funded campaign buys volume, frequency, and amplification in a single transaction, as the Nairobi rate card shows plainly. If those three are substantially one thing observed three ways, then multiplying them triple-counts a single underlying quantity and inflates the score by construction. I do not have a clean answer to that, and the honest position is that the multiplicative form is a claim about how these variables compound. I have not shown that they do.
And there is a challenge larger than my arithmetic. A careful recent survey of this evidence, by Ceren Budak, Brendan Nyhan, and colleagues, argues that the public conversation about online misinformation misunderstands its harms in three specific ways: average exposure to false and inflammatory content is low rather than high, algorithms are less responsible for that exposure than commonly claimed, and social media has not been established as a primary cause of problems like polarization. Exposure, they find, is rare and concentrated in a narrow fringe already motivated to seek it out. Duncan Watts, one of its authors, also sits behind the base-rate corrective I cited in week nine; the same research program has now disciplined me twice.
I think that finding is substantially right, and I do not think it kills the framework, for a reason worth stating precisely. The FTSeffect was never a claim about mass persuasion. The people who walked into the Praça dos Três Poderes had been camped outside army headquarters for weeks; they were already motivated and already organized. What the equation scores is how much pressure was applied to a system and how little capacity that system had to answer inside the window. It does not count minds changed. Budak and colleagues locate the action in the tails. That is where this framework has always operated. Their finding narrows my claim, and a narrowed claim is a better one.
So, concretely, what would make me abandon it. If high-scoring saturation events proved no more likely than low-scoring ones to produce institutional consequences. If response speed and rational deliberation showed no relationship to outcomes across a decent sample. If the numerator variables proved statistically inseparable, in which case the multiplicative form is wrong even where the intuition holds. Those are the conditions. I would like them tested by someone who would enjoy doing it.
What the rubric earns is smaller than an equation usually promises and more useful than the alternative. “The news cycle is overwhelming” cannot be wrong. “This event ran high on frequency and low on rational deliberation, and here is the cadence data” can be wrong, in public, with a date on it. Moving from the first sentence to the second is the whole contribution. Everything else is bookkeeping.
Next week: the liar’s dividend, seven years on, and what has happened since Bobby Chesney and Danielle Citron named it.
— J.W.B.
Note 13 of 24