YZ Sözlüğü
Yapay Zekanın tam sözlüğü
Causal Structural Model (CSM)
Mathematical formalization combining structural equations and directed acyclic graphs to represent causal relationships between variables.
Directed Acyclic Graph (DAG)
Graphical representation of causal relationships where nodes are variables and directed edges indicate direct causal influence.
Average Treatment Effect (ATE)
Expected mean difference between potential outcomes under treatment and control for the entire target population.
Average Treatment Effect on the Treated (ATT)
Average causal effect calculated specifically for the subgroup of individuals who actually received the treatment.
Mediation Analysis
Decomposition of the total causal effect into direct and indirect effects through intermediate variables called mediators.
Conditional Randomization
Experimental procedure where treatment assignment is random conditional on specific covariates to ensure balance.
Unmeasured Confounder
Unobserved confounding variable in the data that can bias causal estimates and requires robust identification methods.
Treatment Effect Heterogeneity
Variation of the causal effect across different subgroups or individuals, requiring conditional heterogeneity models.
Granger causality test
Statistical method based on temporal predictability to determine if a time series causes another in the Granger sense.
Difference-in-differences
Causal identification strategy comparing before-after changes between treatment and control groups to eliminate common trends.
Local Average Treatment Effect (LATE)
Causal effect identified by instrumental variables, applying specifically to individuals whose treatment status is modified by the instrument.
Pearl's causality
Formal approach to causality based on structural models and do-calculus, distinguishing correlation, intervention, and counterfactual.
Structural equation
Mathematical relationship describing how a variable is generated from its direct causes and a random error term.
Sensitivity analysis
Assessment of the robustness of causal estimates to potential unmeasured confounders or violations of identification assumptions.