뒤로Experiments and Observational Studies: Association, Causation, and Confounding
스터디 가이드 - 스마트 노트
자료에 맞춘 맞춤형 노트, 핵심 정의, 예시, 맥락을 확장해 제공합니다.
Experiments and Observational Studies
Association and Causation
Understanding the relationship between variables is fundamental in statistics. The concepts of association and causation help distinguish whether variables are merely related or if one variable directly influences another.
Association (Relationship): Two variables are associated if knowing the value of one provides extra information about the other, beyond what summary statistics alone can offer. This extra information aids in prediction, explanation, or understanding.
Independence: If two variables are not associated, they are considered independent.
Causation (Causal Association): Two variables are causally associated if changing the value of one variable influences the value of the other.
Key Phrase: "Association is not causation." Not all associations imply a cause-and-effect relationship.
Example: "Taking a practice test improves your score" implies a causal association between practice tests and scores.
Confounding and Lurking Variables
Confounding variables can obscure the true relationship between explanatory and response variables. Understanding their role is crucial for interpreting statistical results.
Confounding Variable: A third variable associated with both the explanatory and response variable, offering a plausible explanation for an observed association.
Lurking Variable: A variable not considered in the study, often appearing in observational studies. It can explain observed associations if not accounted for.
Three Possible Explanations for Observed Association:
Causal association exists.
Association is due to a confounding variable.
Coincidence (random chance).
Example: Socioeconomic status (SES) may explain the negative association between birth rate and life expectancy.
Types of Studies
Statistical studies are classified based on whether researchers control variables or simply observe them.
Experiment: Researchers actively control one or more explanatory variables (treatments).
Observational Study: Researchers do not control variables; they observe values as they naturally exist.
Randomized Experiment: The value of the explanatory variable for each unit is determined randomly before measuring the response variable.
Key Point: Randomized experiments allow for establishing causal relationships, while observational studies often struggle to control confounding variables and cannot easily establish causation.
Example: Tobacco companies argued that personality traits, not smoking, caused lung cancer, highlighting the difficulty of establishing causation in observational studies.
Randomization Methods
Randomization is essential for reducing bias and establishing causality in experiments.
Randomized Comparative Experiment: Cases are randomly assigned to treatment groups, and results are compared.
Matched-Pairs Experiment: Each case receives both treatments in random order, or cases are paired, with each receiving a different treatment.
Control Groups, Placebos, and Blinding
Control groups and blinding techniques are used to ensure the validity of experimental results.
Control Group: Receives a treatment that does not directly influence the response variable. This may be no treatment, standard care, or a placebo.
Placebo: A no-treatment disguised as a treatment. The placebo effect occurs when participants respond to the belief they are being treated.
Blinding: Prevents participants or researchers from knowing which treatment is administered.
Single-Blind: Participants do not know their treatment group.
Double-Blind: Neither participants nor researchers know the treatment assignments.
Replication and Realism
Replication and realism are important for scientific validity and generalizability.
Replication: Results should be replicated across many experimental units before accepting an association as scientific fact.
Lack of Realism: Participants may behave differently in experiments than in real life, introducing response bias.
Summary Table: Types of Variables and Study Designs
Term | Definition | Example |
|---|---|---|
Association | Knowing one variable gives extra information about another | Birth rate and life expectancy |
Causation | Changing one variable influences another | Practice tests improve scores |
Confounding Variable | Third variable associated with both explanatory and response variables | Socioeconomic status |
Lurking Variable | Variable not considered in the study | Diet in heart disease studies |
Experiment | Researchers control explanatory variables | Randomized drug trial |
Observational Study | Researchers observe variables as they exist | Survey of smoking habits |
Additional info:
Random selection allows generalization from sample to population.
Random assignment allows attribution of observed association to cause-and-effect.
Replication and realism are necessary for scientific acceptance and practical relevance.
