Epistemology & Philosophy of Science

2.1 Why Epistemology Comes Before Policy

Every political argument rests on a knowledge claim. "The minimum wage causes unemployment." "Immigration increases crime." "Regulation stifles innovation." Before you can evaluate whether any of these claims is true, you must ask a prior question: what would it mean for such a claim to be true, and what kind of evidence could settle the matter? That is the domain of epistemology: the study of knowledge, its structure, and its limits.

This toolkit helps you avoid two symmetrical errors: accepting claims that sound scientific but aren't, and rejecting claims that are scientific because they cut against your priors. Both happen constantly in political debate.

2.2 Popper's Falsificationism

Karl Popper's central insight is negative: science does not proceed by proving things true. It proceeds by trying to prove things false and failing to do so. A theory earns its keep not by accumulating confirmations — confirmations are cheap — but by surviving serious attempts at refutation.

The criterion is falsifiability: a statement is scientific if and only if there exists, in principle, an observation that could prove it wrong. "All swans are white" is scientific because observing a single black swan would refute it. "Everything happens for a reason" is not scientific — not because it is necessarily false, but because no conceivable observation could refute it. It is compatible with every possible state of the world, and a claim compatible with everything explains nothing.

This gives us the demarcation criterion, the line between science and non-science. The line is not between "true" and "false" but between testable and untestable. A claim on the wrong side of that line may still be meaningful (ethics, aesthetics, metaphysics), but it cannot claim the authority of empirical science.

Why this matters politically: Unfalsifiable claims are the backbone of ideological immunization. When a policy fails, watch for the retreat to unfalsifiability: "It would have worked if it had been implemented properly." "The conditions weren't right." "External interference prevented success." Each of these moves the goalposts from a testable prediction ("this policy will produce outcome X") to an unfalsifiable excuse. Popper's framework gives you the vocabulary to name this maneuver.

2.3 The Demarcation Problem: Popper, Kuhn, Lakatos

Popper's criterion is clean. Too clean, his critics argued.

Thomas Kuhn (The Structure of Scientific Revolutions, 1962) observed that actual scientists do not behave the way Popper prescribed. In practice, scientists work within a paradigm, a shared framework of assumptions, methods, and exemplar problems. Normal science is puzzle-solving within the paradigm: filling in details, extending the framework, resolving anomalies. When anomalies accumulate beyond what the paradigm can absorb, a crisis develops, and eventually a paradigm shift replaces the old framework with a new one (Copernicus replacing Ptolemy, Einstein replacing Newton at high velocities, plate tectonics replacing static-earth geology).

The uncomfortable implication: during normal science, scientists do immunize their theories against refutation. An anomalous result doesn't usually overthrow a paradigm. Instead, it gets filed as a puzzle to be solved later, attributed to experimental error, or absorbed through an ad hoc modification. This is not pathological. It is how productive science actually works. You cannot abandon your framework every time a single experiment gives a surprising result.

Imre Lakatos tried to reconcile Popper and Kuhn with his concept of research programmes. A research programme has a hard core (unfalsifiable by methodological decision, e.g., "rational agents maximize utility") surrounded by a protective belt of auxiliary hypotheses that can be modified. A programme is progressive if its modifications predict novel facts; degenerating if its modifications are purely defensive, saving the theory without generating new predictions. The test is not "has it been falsified?" but "is it still generating new knowledge, or merely defending itself?"

This matters for evaluating social science: Marxism, neoclassical economics, and psychoanalysis can each absorb almost any anomaly through protective-belt adjustments. The Lakatosian question is not "is it falsifiable in principle?" but "is the research programme still producing novel predictions that turn out to be correct, or has it become a purely defensive exercise?"

2.4 Scientific Consensus

"The science is settled" is a phrase that Popper, Kuhn, and Lakatos would each reject, for different reasons. Popper: no empirical claim is ever permanently settled; it simply hasn't been refuted yet. Kuhn: what feels "settled" is the current paradigm, which is always in principle replaceable. Lakatos: "settled" confuses the hard core (a methodological choice) with established truth.

None of this means that scientific consensus is worthless. The consensus that the Earth orbits the Sun, that evolution by natural selection is the mechanism of speciation, that vaccines do not cause autism — these rest on mountains of converging evidence from independent lines of inquiry. Dismissing them is denialism.

The useful distinction is between earned consensus and manufactured consensus. Earned consensus emerges from decades of independent replication, failed attempts at refutation, and convergence across methods. Manufactured consensus emerges from institutional pressure, funding incentives, social conformity (see Chapter 1, §1.5 on motivated reasoning), and the suppression of dissent. The COVID-19 lab-leak hypothesis is an instructive case. Early in the pandemic a scientific "consensus" formed that the idea was a conspiracy theory. Later investigation showed that consensus had been shaped partly by institutional conflicts of interest and partly by plain social conformity under crisis. The underlying evidence had not changed; the social permission to examine it had.

The takeaway is not "distrust all consensus." It is: ask how the consensus was formed. Was it forged through open debate and failed refutations, or through institutional gatekeeping and social pressure? The distinction matters enormously.

2.5 Feynman's Cargo Cult Science

Richard Feynman's 1974 Caltech commencement address introduced the metaphor of cargo cult science. During World War II, Pacific islanders observed that American military bases attracted planes full of cargo. After the war, some communities built replica runways, control towers, and headsets out of wood and straw to perform the rituals of the airbase without the mechanism. Planes did not come.

Cargo cult science has the appearance of rigor (peer review, statistical tests, citations, institutional affiliation) without the substance. The substance, in Feynman's telling, is a specific kind of integrity: reporting everything that might make your result unreliable, especially the things that undermine your own conclusion. Most fraud in science is not outright fabrication. It is the quiet exclusion of inconvenient data and the framing of exploratory analysis as confirmatory, what Andrew Gelman later called the garden of forking paths.

Feynman's standard is simple and brutal: "The first principle is that you must not fool yourself — and you are the easiest person to fool." In a domain like political economy, where your priors are strong and your tribal affiliations are engaged, this principle is both maximally important and maximally difficult to follow.

2.6 The Replication Crisis

If cargo cult science is the disease, the replication crisis is the diagnosis.

In 2015, the Open Science Collaboration attempted to replicate 100 published psychology studies. Only 36% replicated, meaning the original result was obtained again with statistical significance. The crisis is not confined to psychology. Medicine (e.g., Ioannidis's 2005 paper, provocatively titled "Why Most Published Research Findings Are False"), economics, and nutrition science have all revealed similar patterns.

The mechanisms are now well-understood:

Publication bias: journals publish positive results and reject null results, so the published literature is a systematically distorted sample of what was actually found.
p-hacking: researchers (consciously or unconsciously) run multiple analyses until one crosses the p < 0.05 threshold, a practice that inflates false-positive rates from the nominal 5% to dramatically higher levels.
HARKing (Hypothesizing After Results are Known): framing a post-hoc finding as if it were a pre-registered prediction, converting exploration into fake confirmation.

The political implication is direct. The next time someone tells you "studies show that X," ask: how many studies? Did they replicate? Was the effect size meaningful or just statistically significant? Were the studies pre-registered? Was the literature systematically reviewed, or is the speaker citing one paper that supports their position?

2.7 Scientism

Scientism is the overextension of scientific authority to domains where the scientific method does not apply. It is a critique of the misapplication of science.

Science can tell you that a carbon tax will reduce emissions by X% (an empirical question). It cannot tell you whether the distributional consequences of that tax are just (a normative question). Science can tell you the expected mortality rate of a pandemic under different intervention scenarios. It cannot tell you how to weigh mortality against economic destruction, educational loss, and civil liberties — those are trade-offs that require value judgments, and value judgments are not derivable from data.

The slogan "follow the science" commits the scientistic error when it implies that policy follows mechanically from scientific findings. It never does. Policy always involves a normative step (a judgment about what matters and how much) that science cannot supply.