Causation, Inference, and Algorithmic Prediction

Vocabulary flashcards reviewing core topics in causation, inference, observational vs randomized studies, and algorithmic/AI modeling based on lecture notes.

Counterfactual Causation

A philosophical definition of causality articulated by David Hume, stating that an event causes another if, had the first event not occurred, the second would never have existed.

Fundamental Problem of Causal Inference

The inherent scientific limitation that researchers cannot directly observe the counterfactual outcome of what would have occurred to the exact same subject in the absence of a treatment.

Hill's Causal Criteria

A set of nine epidemiological guidelines established by Sir Austin Bradford Hill in 1965—including strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, and analogy—used to evaluate evidence for causation.

Gene 'for' a Phenotype (Dawkins View)

The concept formulated by Richard Dawkins in 1982 that a gene is considered 'for' a trait if, holding all other factors equal, possessing the gene increases the probability of an organism exhibiting that trait in a population.

Gene 'for' a Phenotype (Natural History View)

The framework proposed by Kaplan and Pigliucci requiring that a gene must have a statistical association with a trait, an understood biochemical pathway, and evidence that the trait was selected for and maintained in recent evolutionary history.

Pleiotropy

A genetic condition where a single gene influences multiple distinct biological processes or phenotypic traits, such as the DSCAM gene in Drosophila.

Developmental Systems Theory

A biological model proposing that causal explanations cannot be reduced solely to genes, because higher-level cell signaling and cellular protein machinery actively regulate gene expression.

Inference to the Best Explanation (IBE)

A form of abductive reasoning described by C.S. Peirce that infers a hypothesis to be true because it offers the simplest and most plausible explanation for a set of observed facts.

Affirming the Consequent

A formal logical fallacy that invalidly concludes the truth of an antecedent from a true conditional statement and its outcome; it shares the exact logical structure as abductive reasoning.

Prior Probability

In Bayesian statistics, the baseline estimated probability of a hypothesis before any new empirical evidence or experimental data is gathered.

Posterior Probability

In Bayesian statistics, the revised probability of a hypothesis calculated after updating the prior probability with newly observed evidence using Bayes' theorem.

Bayes' Theorem

A mathematical formula $P(A|B) = \frac{P(B|A) P(A)}{P(B)}$ that calculates the posterior probability of hypothesis $A$ given evidence $B$ by multiplying likelihood by prior probability and dividing by total evidence probability.

Base Rate Fallacy

The cognitive tendency to ignore general background statistical probabilities (base rates) in favor of specific, localized test results or case details.

Randomized Controlled Trial (RCT)

An experimental study design where subjects are randomly assigned to either a treatment or control group to minimize bias and isolate causal relationships.

Observational Study

A study design where researchers observe variables without manipulating them, allowing for larger sample sizes and broader portability but risking confounding factors and selection bias.

Confounding Variable

An unmeasured or extraneous variable that correlates with both the presumed cause and effect, potentially generating a spurious correlation.

Matching

A technique used in observational studies to control for confounding variables by comparing individuals or groups that experience equivalent levels of those confounders.

Selection Bias

The distortion introduced when study participants or data are selected non-randomly, creating a sample that is not representative of the target population.

GRADE System

Grading of Recommendations Assessment, Development and Evaluation; a structured process for evaluating the quality of medical evidence to establish clinical practice guidelines, favoring RCTs over observational studies.

Mathematical Algorithm (Model)

A defined set of mathematical rules or quantitative relationships used to map inputs to outputs and predict outcomes.

Mechanical Prediction Superiority

The empirical phenomenon demonstrated by Paul Meehl (1954) and Grove et al. (2000) showing that simple algorithmic predictions consistently meet or exceed human expert predictions in accuracy.

Noise (in Judgment)

The unwanted, random variability in human decisions across identical or similar cases, which automated algorithms eliminate by strictly following fixed rules.

Broken Leg Rule

The guideline stating that human experts should override an algorithmic prediction only when a rare, decisive factor is present that is completely absent from the algorithm's variables.

Overfitting

The error where a model or algorithm is tailored too closely to a specific dataset (including its noise), causing it to fail when applied to new or unseen data.

Machine Learning

A subset of artificial intelligence where algorithms evaluate random combinations of parameters, score their performance against training data, and automatically refine themselves over time.

Training Set

The portion of a dataset supplied to a machine learning model to allow it to discover patterns, adjust parameters, and learn rules.

Testing Set

A separate portion of a dataset withheld during algorithm training, used strictly to evaluate model accuracy on unlearned data and detect overfitting.