counterfactual
A hypothetical alternative scenario used to estimate what would have happened absent a given factor or intervention.
As defined in cwt_mark-zuckerberg-interviews-patrick-collison-and-tyler-cowen
Where it comes up · 43
- Susan Athey on Machine Learning, Big Data, and Causation8 claims
- 76 | Ned Hall on Possible Worlds and the Laws of Nature6 claims
- 164 - Cause Effect5 claims
- 196 | Judea Pearl on Cause and Effect5 claims
- 215 | Barry Loewer on Physics, Counterfactuals, and the Macroworld4 claims
- Valerie Ramey on Stimulus and Multipliers 10/24/20113 claims
- Philip E. Tetlock3 claims
- Making Your 80,000 Hours Count (with Benjamin Todd)3 claims
and 35 more
Most important claims · 10 of 81Search all claims for “counterfactual” →
- Causal reasoning has three hierarchical levels: seeing (passive observation, the domain of statistics and machine learning where belief changes with evidence), doing (intervention/experiment, where you manipulate the world), and imagining (counterfactuals, where you manipulate a stored symbolic model of the world in your mind), each strictly more powerful than the last.164 - Cause Effect
- Reasoning is organized into a three-level hierarchy: level one is association/statistics (inferring the likelihood of one event from observing another, which covers correlation and 99% of machine learning); level two is intervention/action (what happens if I do X, which changes the probability space and is learnable from randomized experiments); and level three is counterfactual reasoning (explanation, retrospection, imagination, asking what would have happened had something not occurred).196 | Judea Pearl on Cause and Effect
- The Scared Straight program, which took youthful offenders to tour prisons to deter crime, appeared to work because participants committed fewer crimes afterward, but compared to a control group that never went through it the participants actually committed more crimes, showing that wanting to make a difference and even apparent before-after improvement is not enough without counterfactual comparison.Making Your 80,000 Hours Count (with Benjamin Todd)
- For some causal questions, the relevant interventionist counterfactual is incoherent or uninformative: asking how stressed Joey would be if he had two parents but were systematically teased for having one parent produces a Kafkaesque scenario that fails to illuminate the original causal question, showing minimal-change comparison can mislead.76 | Ned Hall on Possible Worlds and the Laws of Nature
- On the interventionist approach (championed by figures like Judea Pearl), causal relations are relations exploitable in principle for manipulation: our interest in causation is an interest in facts we can use to settle what would happen if we did one thing or another, with the cleanest test being a god-like localized intervention that tweaks one variable while holding everything else fixed, modeling an ideal controlled experiment.76 | Ned Hall on Possible Worlds and the Laws of Nature
- Modeling phenomena at the micro level rather than macro level is powerful because instead of a few national data points (e.g., two recessions), each locality experiences its own mini-recession, providing far more variation and many mini-experiments to build better models, but only when the phenomenon is genuinely micro-level rather than a macro effect like NAFTA's cross-city movements.Susan Athey on Machine Learning, Big Data, and Causation
- Policies like NAFTA are hard to evaluate because one side of the ledger (a factory shutting down, jobs lost) is concentrated and easy to count, while the benefits (cheaper goods raising disposable income for many consumers) are diffuse and hard to measure, biasing perception toward the visible costs.Susan Athey on Machine Learning, Big Data, and Causation
- We never ran the counterfactual of what would have happened without the 2008 bailouts and interventions, so whether those policies were justified remains a genuinely open question; by analogy, it took a very long time to reach firm understanding of the Great Depression's causes and recovery.Lars Peter Hansen on Risk, Ambiguity, and Measurement
- Counterfactual reasoning — asking what outcome would have prevailed had an event not occurred (had Oswald not killed Kennedy, had Caesar not crossed the Rubicon) — allows historians to communicate and form consensus about trajectories that conflict with what actually happened, despite the fact that the real events did occur.164 - Cause Effect
- Because we rarely get many repeated takes at consequential decisions and cannot observe counterfactuals, a good outcome does not mean a good process was used and vice versa; like blackjack played perfectly over many hands, the goal is a process that gives an advantage across many people even when individual decisions go wrong.David Epstein on Mastery, Specialization, and Range