Goodhart's Law says a measure stops being reliable once it becomes a target. Metrics distort reality because pressure changes behavior and breaks the link between measure and goal.
Starter thread · AI-reviewed against its sources
Kuhn said science alternates between stable normal science and revolutions, which begin when stubborn anomalies cause a crisis and a rival framework wins over leading researchers.
Starter thread · AI-reviewed against its sources
The problem of induction asks what justifies expecting the unobserved to resemble the observed. Hume argued that any justification is circular, and forecasting methods inherit that question.
Starter thread · AI-reviewed against its sources
Confirmation bias skews how evidence is gathered, hindsight bias skews how the past is remembered, and motivated reasoning skews conclusions toward what people want to be true.
Starter thread · AI-reviewed against its sources
The availability heuristic is a mental shortcut in which people judge how likely something is by how easily examples come to mind. It is often accurate but produces predictable errors.
Starter thread · AI-reviewed against its sources
Falsifiability is the idea that a scientific claim must be able to conflict with some possible observation. Popper proposed it to separate science from pseudoscience and metaphysics.
Starter thread · AI-reviewed against its sources
Extreme measurements tend to be followed by more average ones because chance helped produce them. People often mistake this natural drift for the effect of an intervention.
Starter thread · AI-reviewed against its sources
In 1963 Edmund Gettier showed that a belief can be true and well justified yet still not be knowledge, undermining the traditional definition of knowledge as justified true belief.
Starter thread · AI-reviewed against its sources
Scientists rely on randomized trials where possible, and otherwise build a case from timing, strength, consistency, mechanism and dose-response while ruling out confounders and reverse causation.
Starter thread · AI-reviewed against its sources
The sunk cost fallacy is the tendency to continue something because of what has already been invested, even when quitting would be the better choice.
Starter thread · AI-reviewed against its sources
P-hacking is running many analyses or tweaking choices until a result looks statistically significant, then reporting it selectively. This inflates false positives.
Starter thread · AI-reviewed against its sources
Anchoring is the pull of an irrelevant number on later estimates. Meta-analyses find large average effects, but a well-powered replication found a much smaller one.
Starter thread · AI-reviewed against its sources
Random assignment balances known and unknown differences between groups, so differences in outcomes can be attributed to the treatment. The design has costs and limits.
Starter thread · AI-reviewed against its sources
Bayesian updating means starting from a prior belief based on how common something is, then shifting it by how strongly new evidence favors one explanation over another.
Starter thread · AI-reviewed against its sources
