Lecture 6 (causation, variation, stat sig) Flashcards
Types of asso
- Noncausal asso (flaw –> bias)
- Noncausal asso (confounding)
- Random asso (random variation)
- Causal asso
Internal validity
- Assessing design, execution, analysis
- Was it done right or was asso due to bias/confounding/random var?
External validity
- Generalizability
- Reqs internal validity
- Externally valid if asso = result of basic human physiology
Types of causes (2, not combos)
- Necessary: must precede, w/o which doesn’t develop
- Sufficient: leads to disease, but w/o which disease could develop due to smtg else
Types of causes (combos)
- Necessary and sufficient: very rare
- Necessary but not sufficient: all req’d
- Sufficient but not necessary: exposure alone causes, but so do others
- Neither: most chronic diseases
Criteria for causality
- Temporal relationship
- Strength of asso
- Dose-response
- Replication of findings
- Biologic plausibility
- Consider alternate explanations
- Cessation of exposure
- Specificity of disease
- Consistency w/other knowledge
Inarguable criteria for causality
-Temporal relationship: exposure must precede onset of disease
Strength of asso: what does it means for causality
- Stronger asso less likely to be explained by confounding
- But weak assos can be causal, strong assos can be non-causal
Overall assessment of asso btwn alc + breast cancer
- Temporality: yes
- Strength: weak (<2)
- Dose-response: yes
- Replication: yes (many not stat sig though)
- Biologic plausibility: some
- Alt explanations considered: yes
- Cessation: no ev
- Specificity: no
Point estimate
-Estimate of the asso a study is measuring
Test of stat sig
-Prob that an asso occurred due to random variation
P value (def and what it means when not sig)
- Prob that asso at least as large as that observed could occur by chance
- When it’s not sig: doesn’t mean it was due to random chance, ev just not strong enough to reject/accept Ho
Sig level
-Prob that a sig asso will occur when there really is none
Type I error
-Rejectct Ho when it’s really true
-Saying asso exists when it really doesn’t
=alpha
Confidence Interval (def and benefits)
- Interval that, w/given prob, contains true value
- Provide more info than p-value
Power (def and formula)
-Prob that study will demonstrate an asso if it truly exists
=1-beta
Beta
-Prob that a true asso will not be observed
Type II error
- Failing to reject false Ho
- Saying asso doesn’t exist when it really does
What determines power
- Incidence (cohort): max when 1/2 develop
- Prev of exposure (case-control): max when 1/2 exposed
- Strength of true asso
alpha
-Give % of assos expected to be sig due to chance alone (when studying multiple)
