Topics:
Correlation and Causation
Why we do RCTs
Announcements:
HW2 Assigned (Due Wednesday at 11:59 PM)
Optional Reading:
Intuitive Biostatistics Chapter 32
Definition: Indicates an association between two variables.
Types of variables:
Categorical/Binary
Example: Men are more likely to have kidney stones.
One binary and one continuous
Example: 11th graders score higher in statistics than 10th graders.
Both continuous
Example: Older adults have more compliant achilles tendons.
Knowing one variable provides information on the likely values of another, but correlation does not imply causation.
Definition: Causation means changes in one variable directly lead to changes in another.
Temporal Precedence: The cause must precede the effect.
Causation is:
Typically probabilistic in nature.
Example: Relationship between height (y) and weight (x).
Importance of Causation: Provides explanations of how systems work and guides interventions.
Study by Straus et al (1997):
Spanking and antisocial behavior correlation observed over 2 years.
Observational studies don't prove causation due to lack of control over grouping.
Example: Link between smoking and lung cancer shows correlation rather than direct causation.
Mistake in science: inferring causation from mere correlation.
Only true experiments (RCTs) can consolidate causation proof.
Examples:
Rehabilitation experiments must consider confounding factors.
Treatment comparisons need careful structuring to avoid bias.
Example: Treatment choice may be influenced by stone size in kidney stone treatment.
Common confounding factors include:
Baseline condition severity
Socioeconomic status
Comorbidities
Addressing confounds is crucial for establishing causal relationships.
Intervention: One variable must be manipulated.
Randomization: Participants are randomly assigned to groups to ensure comparability.
Controlling: Maintaining a control group helps mitigate biases and ensures valid comparisons.
Importance of Randomization: Eliminates selection bias and balances confounders.
Block Randomization: Ensures equal representation across treatment groups based on identified confounders.
Situations where RCTs are not feasible:
Ethical considerations (e.g., smoking)
Long-term interventions (e.g., diet)
Rare disease incidence studies
Definition: Graphical representation of causal assumptions.
Uses: Helps conceptualize relationships between variables and potential confounders.
Example Pathways:
Direct and backdoor pathways affect the perceived causality.
Causation vs. correlation: critical differentiation.
RCTs provide robust evidence for causality through intervention and randomization.
Causal DAGs are useful tools for representing and understanding research relationships.
Criteria for evaluating causation in observational studies include:
Strength, Consistency, Specificity, Temporality, Biological Gradient, Plausibility, Coherence, Experiment, and Analogy.